Ai-agent based system and method for real-time multilingual and context-aware linguistic transformation in telecommunications

Abstract

A system and method for real-time or near real-time multilingual language transformation in telecommunications environments using distributed artificial intelligence (Al). The system includes at least one processor configured to instantiate a plurality of Al agents, each corresponding to a user in a communication session, and to manage their operation via an Al-agent runtime adapter comprising a microkernel agent manager and a virtual machine layer. A language model abstraction layer enables access to language models and Al algorithm libraries through publish-subscribe interfaces. The system supports contextual transformation of speech based on user profile data such as sentiment, emotional tone, and cultural background. Execution can occur on telecommunication-class switches, edge devices, or general-purpose computing hardware. Al agents operate asynchronously, access shared or private memory, and produce personalized, bidirectional linguistic output. The architecture supports secure, scalable deployment across heterogeneous devices and networks.

Description

INTERNATIONAL APPLICATION UNDER THE PATENT COOPERATION TREATY Non Provisional Utility Patent Application APPLICANTS: Cheryl Ee-Lin Cunningham INVENTORS: Cheryl Ee-Lin Cunningham James David Harlow TITLE: AI-AGENT BASED SYSTEM AND METHOD FOR REAL-TIME MULTILINGUAL AND CONTEXT-AWARE LINGUISTIC TRANSFORMATION IN TELECOMMUNICATIONS. FIELD:

[0001] The present invention relates generally to the fields of computational linguistics and artificial intelligence (AI), and more particularly to systems and methods for real-time or near real-time multilingual transformation of spoken and textual communication using AI agents, language models, and algorithmic interpretation within telecommunication environments. PRIORITY CLAIM:

[0002] This application claims priority to U.S. Provisional Patent Application No. 63 / 670,713, filed on July 12, 2024, the contents of which are incorporated herein by reference. BACKGROUND ART

[0003] The present disclosure relates generally to telecommunications infrastructures and, more particularly, to systems and architectures for dynamic, real-time language transformation of speech and text using artificial intelligence (AI). Specifically, the invention addresses the integration of AI-driven computational linguistics into telecommunications environments to enable multilingual communication across diverse user devices and platforms.

[0004] Existing language translation systems in telecommunication settings often rely on static processing pipelines, centralized engines, or non-interactive translation layers, resulting in latency, generic output, and limited adaptation to the conversational context. These systems typically do not account for individual user characteristics such as sentiment, tone, or cultural profile, and are constrained in their ability to operate flexibly across heterogeneous networkconditions and devices.

[0005] As AI becomes increasingly embedded in communication technologies, there is a growing need for systems that can operate with greater contextual awareness and responsiveness. Recent developments point toward the importance of enabling AI agents to process language in real time, drawing on user-specific inputs such as demographic or emotional data to shape communication more naturally. This trend reflects a broader movement toward language-level personalization, where AI systems can dynamically adjust linguistic output to better match the speaker’s intent, tone, and audience.

[0006] While not all existing systems fully address these challenges, architectures that leverage distributed AI agents, real-time model abstraction, and modular algorithm libraries offer a foundation for advancing in this direction. Such systems can be particularly effective when deployed within telecommunications infrastructures, enabling device-agnostic operation, reduced latency, and broader accessibility for users with varying levels of connectivity. Accordingly, there is value in expanding these capabilities within real-time communication systems to support increasingly complex and personalized language interactions. DISCLOSURE OF INVENTION

[0007] The present invention provides systems, methods, and computer-readable media for performing real-time or near real-time linguistic transformation of speech within telecommunication sessions using distributed artificial intelligence (AI) agents. The invention enables interpretation, transcription, translation, and transliteration of speech in a dynamic and context-aware manner, based on user-specific profile data such as demographics, emotion, cultural background, and communication intent.

[0008] In one embodiment, the invention comprises a plurality of AI agents instantiated during a telecommunications session, each corresponding to a session participant. These AI agents operate within a shared runtime environment facilitated by an AI-agent runtime adapter, which includes a microkernel-based agent manager and a language model abstraction layer. The abstraction layer enables the agents to interact with one or more language models and AI algorithm libraries through a publish-subscribe architecture. These components collectively support low-latency, scalable processing across heterogeneous devices and networks.

[0009] The AI agents are capable of executing personalized language operations, including sentiment scoring, emotional interpretation, tone modulation, and context-sensitivetranslation. The system may further enable device-agnostic deployment, allowing seamless operation across mobile devices, VoIP endpoints, conferencing tools, and other networked communication platforms. This allows users with minimal hardware requirements and basic connectivity to access sophisticated multilingual transformation services.

[0010] The invention addresses the growing need for language-level hyper-personalization in AI-mediated communication. By leveraging AI agents that adapt to individual user profiles and interaction contexts, the system enhances clarity, cultural appropriateness, and emotional fidelity in cross-lingual communication. In doing so, it lays a foundation for future AI systems capable of simulating human communication behaviors with increasing accuracy and nuance. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Claimed subject matter is particularly pointed out and distinctly claimed in the concluding portion of the specification. However, both as to organization and / or method of operation, together with objects, features, and / or advantages thereof, it may best be understood by reference to the following detailed description if read with the accompanying drawings in which:

[0012] Figure 1 illustrates a multi-layered telecommunications infrastructure (2500) designed to support real-time or near real-time computational language conversion using AI agents. At the top, the Geosynchronous Earth Orbit (GEO) satellite layer (2510) provides broad, high-latency global coverage. Below it, the Medium Earth Orbit (MEO) layer (2520) offers a balance between coverage and latency while the Low Earth Orbit (LEO) cubesats layer (2530) delivers low-latency, high-speed data transmission ideal for real-time communication. This space-based network is integrated with a terrestrial layer (2540) encompassing IoT consumer goods and gaming applications enabling connectivity for smart devices and interactive platforms. Supporting mobile and transport-based connectivity, the infrastructure includes aircraft (2532), vehicles and fleets (2542) and maritime systems (2544). End-user access is enabled through 5G / 6G smartphones and internet-enabled devices forming a cohesive global communication ecosystem capable of supporting seamless, AI-driven multilingual interaction across diverse domains and platforms.

[0013] Figure 2 depicts a telecommunications network (2200) that operates in a one-to-one configuration, enabling real-time or near real-time language conversion between individual users. In this implementation a Japanese speaker (330) communicates with an Arabic receiver (2620) through a network-enabled translation process. The system performs translationof foreign text to Japanese text (2602) and foreign text to Arabic text (2604) depending on the direction of communication. For audio-based interaction the translated text is further converted into spoken language audio files with foreign text translated to Japanese and converted to a Japanese .wav file (2612) for playback to the Japanese user and foreign text translated to Arabic and converted to an Arabic .wav file (2614) for playback to the Arabic user. This configuration allows seamless multilingual communication through automated translation and speech synthesis, enabling natural and efficient exchanges between users of different languages over the telecommunications network.

[0014] Figure 3 illustrates an example telecommunications network (300) operating in a one-to-many configuration where a single English speaker (310) communicates simultaneously with multiple recipients in different languages. The process begins with speech-to-text conversion (350) where the English audible speech is transcribed to text (352). The transcribed English text then passes through an interpretation and translation module (360) which translates the content into multiple target languages, including Japanese (2602), Arabic (2604), English refinement or repetition (2606), and Mandarin (2608). Following translation, the system utilizes text-to-speech synthesis (370) to generate spoken audio outputs in the respective languages: translated text + Japanese .wav file (302), translated text + Arabic .wav file (304) and translated text + Mandarin .wav file (306). These audio files are delivered to the respective receivers, including the Japanese receiver (2610), Arabic receiver (2620) and Mandarin speaker (320). This configuration demonstrates a multilingual broadcasting capability, allowing one speaker to engage with a diverse, multilingual audience through AI-powered translation and voice synthesis in real or near real time.

[0015] Figures 3A–3D illustrate a schematic block diagram of an example one-to-many telecommunications network configuration designed to support real-time multilingual voice translation using AI-enabled cloud services. The system includes various speakers English Actor (310), Arabic Actor (330), Mandarin Actor (320) and Japanese Actor (340) each connected via laptops (312, 314, 316, 318) equipped with HTML5 browsers, user-specific channel management agents and a backend processing system called JOSH which stores pre- and post-translation text and .wav files in the DOM. The English speech is first processed through a Speech-to-Text GA (322) for low-latency, streaming transcription and trickled as JSON data to databases. This transcription is then handled by a Language Translation GA (324, 326) to convert text into target languages like Arabic, Mandarin, and Japanese. Each translated text is synthesized using a Text-to-Speech GA producing natural-sounding audio output in the respective language formats (e.g., Arabic.wav + Arabic.txt). The system relies on HTTP POSTrequests to IBM Watson endpoints for speech recognition and synthesis, and stores results in graph databases for personal and regional LLMs. A publish-subscribe message bus enables real-time updates and all data exchanges use secure APIs with authorization tokens. This architecture enables scalable multilingual communication where one speaker can be understood simultaneously by multiple recipients in different languages.

[0016] Figure 4 illustrates a telecommunications network (400) configured to operate in a many-to-one configuration enabling multilingual inputs from various speakers to be interpreted and delivered to a single receiver in their preferred language. In this setup, users such as a Mandarin speaker (320) a Japanese receiver (2610) and an Arabic receiver (2620) communicate toward a centralized English receiver (2630). Spoken input from the Mandarin speaker undergoes speech-to-text conversion (350) with Mandarin audible speech transcribed to text (358). The resulting text is processed through an interpretation and translation module (360) where it is translated into multiple languages including English (2606), Japanese (2602) Arabic (2604) and Mandarin (2608). Each translated text is then converted into an audio file using text-to-speech synthesis (370) producing outputs such as translated text + English .wav file (308) Japanese .wav file (302) and Arabic .wav file (304). These audio outputs are directed to their respective recipients ensuring that regardless of the source language, the final message is received in the listener’s native or selected language. This configuration supports converged multilingual communication toward a common recipient, streamlining global collaboration and multilingual conferencing.

[0017] Figure 5 presents a telecommunications network operating in a many-to-many configuration enabling real-time multilingual communication among speakers and receivers of different languages. Participants include an English speaker (310), Mandarin speaker (320), Japanese speaker (330) and Arabic speaker (340). Each speaker’s audible speech is first transcribed via speech-to-text modules Japanese (502), Mandarin (504), English (506) and Arabic (508). The transcribed text is then routed to a translation module which translates it into various target languages: to English (2606), to Mandarin (2608), to Japanese (2602) and to Arabic (2604). Each translated text is then processed through text-to-speech conversion resulting in audio files: Japanese .wav (302), Arabic .wav (304), Mandarin .wav (306) and English .wav (308). These outputs are delivered to the corresponding receivers English receiver (2630), Japanese receiver (2610), Arabic receiver (2620) and Mandarin receiver (2640) ensuring that each participant receives communications in their native or selected language. This configuration demonstrates a highly integrated AI-powered network capable of facilitating seamless, real-time multilingual dialogue across global participants.

[0018] Figure 6 illustrates a schematic block diagram (600) of an example AI-agent runtime adapter system, highlighting how AI agents perceive, process and act within a dynamic environment. At the core is the AI agent runtime environment (630) implemented within a wavefront array (SISD Single Instruction, Single Data) architecture which facilitates efficient sequential decision-making. The system begins with agent sensors (610) capturing inputs from the environment (608) interpreted as percepts (606) signals that inform the agent of “what the world is like now” (602). These inputs are processed using condition-action rules (606) to decide “what action I should do now” (604). The system maintains and updates internal states such as current state (612a) how the world evolves (612b) and what the agent’s actions do (612c). The outcomes are carried out via actuators completing the perception-decision-action loop. This runtime framework enables adaptive, context-aware responses from AI agents operating in real-time telecommunications or language processing environments.

[0019] Figure 7 compares two architectural models of operating systems: a monolithic kernel-based system (710) and a microkernel-based system (720). In the monolithic kernel architecture components such as application system calls, virtual file system (VFS), interprocess communication (IPC), file system, scheduler, virtual memory, device drivers and dispatcher all operate within the kernel mode providing direct and high-performance access to hardware. Both user mode and kernel mode interact closely, but this tightly coupled design can be less stable or secure if one component fails. In contrast, the microkernel architecture (720) separates key functionalities into isolated modules. Here applications, IPC, UNIX server, device drivers and file servers operate primarily in user mode while only essential services such as basic IPC, virtual memory management and scheduling remain in kernel mode. This separation enhances system stability, modularity and security by minimizing kernel responsibilities and isolating failures.

[0020] Figure 8 presents a schematic of a heterogeneous AI-agent runtime adapter (610) which may be implemented, either fully or partially within a uniprocessor architecture such as SISD (Single Instruction, Single Data). The system models the core perception-decision-action loop of an AI agent. Input is received from the environment (608) through sensors generating precepts that represent “what the world is like now” (602). The agent processes these precepts using condition-action rules (606) to determine “what action I should do now” (604). The chosen response is executed through actuators allowing the agent to interact with the environment. This runtime adapter supports adaptive, context-aware behavior, even within resource-constrained processing environments, by efficiently managing sensing, reasoning, and action cycles in a modular and unified system.

[0021] Figure 9 illustrates a model-reflexive AI-agent runtime adapter (620) which maybe implemented either fully or partially within a wavefront array architecture such as SISD, MISD or similar computational structures. This advanced AI-agent framework expands on traditional perception-action models by incorporating self-awareness and predictive modeling. The agent receives input from the environment (608) as precepts (606) representing “what the world is like now” (602). Using condition-action rules (606) and internal state representations (612a) the agent determines “what action I should do now” (604). Additionally, the agent maintains models of how the world evolves (612b) and what my actions do (612c) enabling forward-looking, reflexive decision-making. Actuators then carry out the selected actions in the environment. This configuration allows the agent not only to react to current stimuli but also to adapt based on predicted outcomes, leading to more intelligent and context-aware behavior in real-time or near real-time operations.

[0022] Figure 10 illustrates a goal-type AI-agent runtime adapter (1000) which may be implemented, wholly or partially, within a wavefront-array architecture (e.g., SISD, etc.). This AI-agent framework is designed to support goal-driven decision-making. The agent interacts with the environment (608) by receiving precepts forming an understanding of “what the world is like now” (602). It maintains an internal state (612a) and uses models to predict how the world evolves (612b) and what its actions do (612c). Before acting, the agent evaluates potential outcomes, such as “what it will be like if I do action A” (614) and compares them against its defined goals (616). Based on this analysis, the agent decides “what action I should do now” (604) and executes the selected action via actuators. This architecture enables the AI agent to plan purposefully, anticipate consequences and act intelligently toward achieving specific objectives in dynamic environments.

[0023] Figure 11 illustrates the architecture of a cybersecurity AI-agent system (2700) that integrates intelligent mobile agents, game theory, and distributed detection to defend against network attacks. A network security officer (2702) oversees operations, supported by an intrusion detection server (2704) and a mobile agent platform factory (2714). These platforms deploy mobile agents through migration mechanisms (2706) across a network mode (2712) that includes multiple nodes such as PC1, PC2, and PC3. The system uses an intrusion detection processor trained on the KDD dataset classifying network traffic into categories like normal, smurf, Neptune, back and multihop with detection volumes reaching over 143,000 events. The agents monitor the network using sniffer tools and detection engines and interact with attackers through a multi-step process: (1) detecting the attack (2) engaging in a non-cooperative game to analyze adversarial behavior (3) computing a risk value using Nash equilibrium (4) entering a cooperative game to coordinate defense (5) activating Working Group Agents (WGAs) (6) computing Shapley values to evaluate contribution, (7) generating and sending an attack reportand (8) deploying Local Agents (LAs) to threatened nodes for direct intervention. This system enables real-time, intelligent and collaborative cyber defense across dynamic and distributed network environments.

[0024] Figure 12 illustrates a utility-type AI-agent runtime adapter (1200) which may be implemented, in whole or in part, within a wavefront array architecture such as SISD, MISD or MIMD. This agent model is designed to make intelligent, goal-aligned decisions based on utility optimization. The agent receives precepts from the environment (608) to assess “what the world is like now” (602) and maintains an internal state (612a). It uses models to understand how the world evolves (612b) and what its actions do (612c). The agent predicts “what it will be like if I do action A” (614) for various possible actions and evaluates each future state using a utility function (618) which determines “how happy I will be in such a state” (619). The action that yields the highest utility is selected and executed through actuators. This architecture enables AI agents to reason not only about outcomes but also about the desirability of those outcomes, making decisions that maximize satisfaction or benefit within complex, dynamic environments.

[0025] Figure 13 depicts an unsupervised and / or supervised learning AI-agent runtime adapter (1300) which may be implemented, in whole or in part, within a wavefront-array architecture (e.g., SISD, MISD, MIMD, etc.). This learning-oriented agent continuously improves its performance through interaction with the environment (608). The agent receives percepts from the environment via Critic Sensors (1302) which compare the agent’s behavior to a predefined performance standard and provide feedback. The Learning Element (1304) uses this feedback and defined learning goals to update the agent’s internal models and decision-making strategies. To support learning, a Problem Generator (1306) introduces new experiments or challenges, encouraging exploration and adaptation. The Performance Element (1308) applies learned behavior through effectors, producing actions that impact the environment, which in turn may change as a result. This adaptive feedback loop enables the agent to refine its behavior over time, whether through supervised learning (with feedback based on correct outcomes) or unsupervised learning (pattern discovery without explicit labels), enhancing autonomy and effectiveness.

[0026] Figure 14 illustrates an example multi-agent platform (1400) structured in a layered architecture to support intelligent task execution and communication across distributed components. The platform is divided into three functional layers: The Superior Layer (1402) the Intermediate Layer (1404) and the Reactive Layer (1406). The Superior Layer manages communication by protocol and performs high-level planning according to tasks often handled by superior agents or super users. The intermediate layer focuses on task-based planning through message passing communication facilitating coordination among agents via structured messages,orders, and information exchanges. The reactive layer (1406) handles centralized planning and real-time action / perception where reactive agents operate locally within locality a, b, and c responding to environmental stimuli. Components such as component a, b, and c and connectors allow the platform to dynamically create new components or connectors enabling scalability and adaptability. Both normal users and super users interact with the platform by sending user messages managing new connections or deploying new agents. This hierarchical, modular design allows the system to perform complex, coordinated tasks in a flexible and scalable multi-agent environment.

[0027] Figure 15 illustrates a system (1500) comprising an aggregator AI-agent architecture designed to process uncertain sensor data, generate predictions and support decision-making. Within the system, machine A (1510) operates in environment A (1520) and is monitored by a sensor agent (1502) which collects sensor data with uncertainty. This raw data is passed to the aggregator agent (1506) which combines and processes it into aggregated data subsequently stored in a historical database (1514). A predictor agent (1504) utilizes this historical and current aggregated data to generate predictive models, which are refined through a model trainer agent (1512). The trained models are then presented through a display model and visualized to users via the user interface agent (1516). Finally, insights and model outputs inform the decision maker agent (1508) enabling intelligent, data-driven decisions. This integrated agent-based system allows for robust handling of uncertain data, continuous learning and real-time interaction between human users and AI-driven processes.

[0028] Figure 16 depicts various aspects of a trusted platform module (TPM)-based AI-agent system (1600) designed to enhance cybersecurity in embedded or networked environments. The system is built to detect and respond to threats such as hacking (1602), spoofing (1604) and data falsification (1606). Central to the architecture is the TRN chip (1620) which works in conjunction with crypto software (1622) and secure boot software (1624) to establish a secure execution environment. Critical cryptographic operations like sealing, signing and sealed-signing (1610) protect data and ensure integrity. The system monitors real CAN packets (1632) and uses feature extraction (1634) to identify patterns. These features are fed into a classifier (1636) that performs a normal vs. attack decision (1638). For accurate detection, the system is trained using labeled CAN packets (1640) processed through a DNN (Deep Neural Network) structure (1644) comprising input layers, hidden layers, and output layers (1642). This AI-agent architecture combines trusted hardware and deep learning to enable secure, intelligent and adaptive cyber threat detection in real time.

[0029] Figure 17 presents a block diagram of an AI Agent Runtime Adapter system (1700) designed to manage a plurality of AI agents through a centralized control mechanismknown as the HuroBOSS Agent Manager (1704). This kernel-level process autonomously or semi-autonomously handles the deployment, scheduling, and provisioning of AI processes and daemons, enabling streamlined and scalable operation of intelligent systems. Each AI agent interacts with Language Model Math Modules (1702) for computational operations, such as matrix manipulation using 2x2 or 4x4 vectors, and relies on a robust AI Algorithm Library (1706) that includes functions for semiotic interpretation, hermeneutic analysis, sentiment and emotion scoring, and axiomatic rule-based logic. The entire system operates on an underlying operating system (1708)—which may be POSIX-compliant or based on single-threaded or multiplexed architectures—and executes code through machine-safe language interpreters (1710) supporting platforms like Python, Java VM, or Rust. These are run on a virtual machine or microkernel (1712), managed via a runtime adapter (1714) that supports and coordinates AI agent behavior (1716). At the hardware level, tasks are performed by a general-purpose CPU (1718) with support for various computational paradigms, including MIMD, SIMD, MISD, and SISD (1720), ensuring flexibility, parallelism, and efficiency in runtime AI operations.

[0030] Figure 18 (1800) illustrates the architecture of an AI Agent Control Unit, highlighting its internal organization and interactions between components that enable coordinated and efficient AI processing. At the core is the AI Agent Control Unit (1800), which orchestrates the behavior of multiple AI agents. These agents access a Library of Code and Algorithms (1802) that provides essential logic, computational methods, and predefined AI routines. Communication and data exchange are facilitated through a Shared Memory module (1804), allowing multiple agents to interact, coordinate tasks, and share learned insights in real time. Simultaneously, each AI agent maintains its own Private Memory (1806) for handling isolated computations, sensitive data, or agent-specific operations without interference. The legend (1810) maps the structural layout, showing how individual agents (labeled A, B, C, D, W, X, Y, Z) are equipped with adapters that connect them to the shared memory, private memory, and code library, enabling modular and flexible deployment across various computational tasks. This setup supports scalable, multi-agent AI operations with controlled memory access and centralized code management.

[0031] Figure 19 illustrates a generalized diagram of language model construction (1900) that outlines the multi-stage pipeline used to process raw natural language data into a fully trained and deployable language model. The process begins with raw text input (1902), which undergoes tokenization, machine transliteration, and transcription to convert human language into machine-processable units. Following this, various approaches (1904) such as grapheme-based, phoneme-based, or hybrid techniques are applied to standardize language representation. The next stage is preprocessing (1906), involving key tasks like Part-of-Speech(POS) tagging, stemming, lemmatization, and stop-word removal, all aimed at reducing noise and preparing the text for analysis. Subsequently, lexical and morphological analysis (1910) is performed, followed by feature extraction, which isolates key linguistic features for computational modeling. Machine learning algorithms are then used to interpret the data, often incorporating stages like machine translation, sentiment analysis (SA), and named entity recognition (NER) (1908). The system may also perform postprocessing and evaluation to measure the quality and accuracy of the model. The processed data is passed into a neural network (1925) composed of input layers, hidden layers, and output layers, enabling the final generation of intelligent language representations for downstream applications.

[0032] Figure 20 presents a schematic block diagram of a Language Model Abstraction Layer (2000), showcasing a multi-device, multi-language architecture for handling natural language prompts across various platforms. Personal computer (2010), cell phone (2020), and tablet (2030) are shown as interface devices, each capable of issuing prompts in their respective languages (Language 1, 2, and 3), with all language model outputs returned in the same originating language, ensuring seamless multilingual interaction. At the core of the architecture is a publish-and-subscribe communication bus (2002 and 2008) that operates asynchronously to manage message distribution between components. A central processing layer runs CICAI (2004)—an AI execution environment built within a POSIX-compliant or ternary / quad-threaded operating system, enabling robust multitasking. Within this layer is a Natural Language Prompts Paraphraser (2006) responsible for interpreting user inputs and executing diverse AI algorithms. These prompts are dispatched across multiple language model instances (2040a–2040e) that process and return the response in the appropriate format and language. This abstraction architecture ensures consistent, device-agnostic, and language-specific AI interaction through modular, publish-subscribe-driven execution.

[0033] Figure 21 illustrates a schematic diagram (1100) of an example computing environment implementation showcasing how multiple devices interact over a network and share resources for processing and data management. The system includes interconnected devices First Device (1102), Second Device (1104) and Third Device (1106) communicating via a network (1108). Each device comprises core computing components including a processor (1120) responsible for executing instructions and a memory hierarchy consisting of primary memory (1124) for fast-access data storage, secondary memory (1126) for long-term storage and general memory (1122). Data input and output operations are managed through an input / output interface (1132) while communication across devices and systems is handled by a communications interface (1130). Additionally, the system utilizes a computer-readable medium (1140) which can store instructions or data used by the processor. This modular architecturesupports distributed computing, remote resource access and scalable processing across networked systems.

[0034] Figure 22 presents a schematic block diagram (200) of an example Internet of Things (IoT)-type device illustrating its core components and functional architecture. At the center is the Processor (210) responsible for executing instructions and coordinating operations. The processor interfaces with memory (230) which stores data and program logic, including software / firmware code (232). The device interacts with its environment through various sensors (250) that collect data and counters / timers (260) that support time-based operations and event tracking. A display (240) provides user-facing feedback or system status. Communication with other devices or networks is enabled via a communications interface (220) allowing the IoT device to send or receive data wirelessly or through wired connections. This architecture enables intelligent, connected operations suitable for various IoT applications such as monitoring, automation or remote control.

[0035] Figure 23 illustrates a schematic diagram (1100) of an example computing environment implementation showcasing how multiple devices interact over a network and share resources for processing and data management. The system includes interconnected devices First Device (1102), Second Device (1104) and Third Device (1106) communicating via a network (1108). Each device comprises core computing components including a processor (1120) responsible for executing instructions and a memory hierarchy consisting of primary memory (1124) for fast-access data storage, secondary memory (1126) for long-term storage and general memory (1122). Data input and output operations are managed through an input / output interface (1132) while communication across devices and systems is handled by a communications interface (1130). Additionally, the system utilizes a computer-readable medium (1140) which can store instructions or data used by the processor. This modular architecture supports distributed computing, remote resource access and scalable processing across networked systems.

[0036] Reference is made in the following detailed description to accompanying drawings, which form a part hereof, wherein like numerals may designate like parts throughout that are corresponding and / or analogous. It will be appreciated that the figures have not necessarily been drawn to scale, such as for simplicity and / or clarity of illustration. For example, dimensions of some aspects may be exaggerated relative to others. Further, it is to be understood that other embodiments may be utilized. Furthermore, structural and / or other changes may be made without departing from claimed subject matter. References throughout this specification to “claimed subject matter” refer to subject matter intended to be covered by one or more claims, or any portion thereof, and are not necessarily intended to refer to a complete claim set, to aparticular combination of claim sets (e.g., method claims, apparatus claims, etc.), or to a particular claim. It should also be noted that directions and / or references, for example, such as up, down, top, bottom, and so on, may be used to facilitate discussion of drawings and are not intended to restrict application of claimed subject matter. Therefore, the following detailed description is not to be taken to limit claimed subject matter and / or equivalents. DETAILED DESCRIPTION

[0037] Reference is made in the following detailed description to accompanying drawings, which form a part hereof, wherein like numerals may designate like parts throughout that are corresponding or analogous.

[0038] It will be appreciated that the figures have not necessarily been drawn to scale, such as for simplicity or clarity of illustration.

[0039] For example, dimensions of some aspects may be exaggerated relative to others.

[0040] Further, it is to be understood that other embodiments may be utilized.

[0041] Furthermore, structural, or other changes may be made without departing from claimed subject matter.

[0042] References throughout this specification to “claimed subject matter” refer to subject matter intended to be covered by one or more claims, or any portion thereof, and are not necessarily intended to refer to a complete claim set, to a particular combination of claim sets (e.g., method claims, apparatus claims), or to a particular claim.

[0043] It should also be noted that directions or references, for example, such as up, down, top, bottom, and so on, may be used to facilitate discussion of drawings and are not intended to restrict application of claimed subject matter.

[0044] Therefore, the following detailed description is not to be taken to limit claimed subject matter or equivalents.

[0045] References throughout this specification to one implementation, an implementation, one embodiment, an embodiment, or the like means that a particular feature, structure, characteristic, or the like described in relation to a particular implementation or embodiment is included in at least one implementation or embodiment of claimed subject matter.

[0046] Thus, appearances of such phrases, for example, in various places throughout this specification are not necessarily intended to refer to the same implementation or embodiment or to any one particular implementation or embodiment.

[0047] Furthermore, it is to be understood that particular features, structures, characteristics, or the like described are capable of being combined in various ways in one ormore implementations or embodiments and, therefore, are within intended claim scope.

[0048] In general, of course, as has always been the case for the specification of a patent application, these and other issues have a potential to vary in a particular context of usage.

[0049] In other words, throughout the patent application, particular context of description or usage provides helpful guidance regarding reasonable inferences to be drawn; however, likewise, “in this context” in general without further qualification refers to the context of the present patent application.

[0050] As alluded to previously, in order to meet the ever-increasing demands for telecommunication or related services, efforts continue to be made to improve telecommunications or related technologies, such as, for example, to improve capacity, increase data transfer speeds, reduce costs, implement additional features or capabilities, or the like. Furthermore, as AI-type applications become increasingly sophisticated, utilization of suitable (e.g., specifically trained, tailored) AI agents, such as to assume a growing number of responsibilities, including helping to bring improvements to telecommunications or related technologies, for example, continues to be an area of development.

[0051] Embodiments described herein may involve telecommunications networks, such as digital or analog networks, for example, with one or more data processing units that may be communicatively connected or coupled to one or more computing devices or platforms so as to provide computational source-to-target language conversion services in real time or near-real time.

[0052] As will be described in greater detail below, these or like services includes, for example, interpretation, translation, transcription, transliteration, or like aspects or processes, such as implemented, at least in part, in connection with one, or a plurality of, sender(s) and one, or a plurality of, receiver(s) engaged in unilateral, bilateral, directional, or omnidirectional network-type communication, for example.

[0053] Thus, example embodiments may, for example, be utilized, in whole or in part, to facilitate or support one or more technological solutions to one or more global technological problems that may otherwise be unsolvable, including, for example, organizing, or collapsing discrete operational equipment (e.g., legacy of prior years of phone-service provisioning) into more effective or more efficient (e.g., smaller, greener, with lower power demands) footprints.

[0054] As a way of illustration, according to World Trade Organization (WTO), high costs (e.g., US$1.5 Trillion Dollars, currently) in non-tariff barriers to trade and cross-border commercial transaction inefficiencies may be due to ethnic, cultural, or language barriers.

[0055] As will also be seen, one or more embodiments described herein may address these or like inefficiencies, among other aspects.

[0056] Thus, “Computational Interpretation of Communication using Artificial Intelligence Algorithms” (CICAI) may comprise, for example, a communications network, such as a multi-agent network, as one particular example, of one or more processing units executing computer-readable instructions to facilitate or support one or more Artificial Intelligence-type processes, operations, or algorithms.

[0057] In some instances, these or like processes, operations, or algorithms includes neural-network algorithms, for example, or other suitable computer-readable instructions that may be directed to computational linguistics, including but not limited to syntactic parsing, semantic interpretation, sentiment analysis, and cross-language alignment, in one or more embodiments.

[0058] More specifically, in certain simulations or experiments, it has been observed that, in some instances, a CICAI applied to a particular area of computational linguistics, including but not limited to syntactic parsing, semantic interpretation, sentiment analysis, and cross-language alignment, such as relating to, for example, interpretation (e.g., semiological, hermeneutical, interpretation of a sender’s capabilities), transcription, translation, transliteration, or the like may prove beneficial.

[0059] In some implementations, the algorithmic processing includes sentiment scoring, emotional index evaluation, or semiotic analysis to reflect tone, intent, or cultural context in the transformed output.

[0060] As also described below, one or more embodiments includes, for example, multilingual aspects, operations, processes.

[0061] Due at least in part to multilingual capabilities described herein, in an implementation, a CICAI network may perform one-to-one, one-to-many, many-to-one, and many-to-many interpretations, translations, transcriptions, or transliterations to facilitate conversations or presentations, for example, so that one could have the entire United Nations speaking with each other via headset, with each individual speaking their own language and hearing their own language in return in real time or near real-time, for example.

[0062] Continuing with the U.N. example, the General Assembly could all be listening to the same speaker.

[0063] The speaker would speak the speaker’s own language and everybody in the audience would hear the speech in their respective languages in real time or near real-time (e.g., in some embodiments a minor delay of perhaps 200ms or so may be experienced).

[0064] That sort of stream of consciousness-type interaction involving interpretation, translation, transcription, or transliteration is meaningful, and it’s quite an advance of the state of the art.

[0065] In one or more embodiments, interpretation, translation, transcription, and transliteration operations may be performed concurrently within a processing pipeline. To enhance grammatical or contextual understanding across languages, user-specific information may be incorporated. For example, during transcription, diacritics may be applied to capture tone, emotional tension, or nuance in a conversation or presentation. In some implementations, the original utterance may be transcribed into a computer-readable form in the source language and stored in memory to support post-processing or verification tasks that minimize language conversion errors.

[0066] In one implementation, a telecommunications session between a first and second user communication device is intercepted at a media stream processor. The processor digitizes audio streams and forwards them to an AI-agent runtime adapter. The runtime adapter deploys two autonomous AI agents, one representing each user via a microkernel agent manager. Each AI agent performs real-time speech detection, translation, and re-synthesis using a language model accessed through a publish-subscribe abstraction layer. For example, the first agent receives English audio, transcribes and translates it to Spanish using transformer-based models, and generates a synthesized Spanish output for the second user. The second agent performs a reverse operation for bidirectional communication.

[0067] The system may further provide a linguistically transformed and re-synthesized output stream, which is transmitted to the destination user device in real time or near real time as part of the ongoing telecommunications session. In this example, the method includes receiving a digital audio stream representing the session, instantiating AI agents representing each user, and executing an AI-agent runtime adapter comprising a microkernel agent manager, a language model abstraction layer configured for publish-and-subscribe access, and an AI algorithm library for multilingual signal processing.

[0068] In one or more embodiments, a telecommunications-class switch infrastructure may support real-time or near real-time interpretation, transcription, translation, or transliteration by provisioning user-specific AI agents via a runtime adapter. This runtime adapter may include a microkernel agent manager, a language model abstraction layer, and an AI algorithm library. A media stream processor may digitize intercepted audio signals from user communication devices and route them through this architecture. In one embodiment, computer-executable instructions stored on a non-transitory medium may cause a processor to intercept digital audio, instantiate user-specific AI agents, and perform linguistic transformation of the stream for retransmission, using the microkernel-based runtime adapter.

[0069] Once a communication or presentation has been translated into a target language using electronic source-to-target language conversion, a complex set of diacritics may be addedto reflect pronunciation, tone, emotion, or timbre. For example, if Arabic is the source language and Mandarin is the target, the system may display Mandarin text with diacritic annotations and not simply produce an audio transcription, thereby guiding pronunciation and capturing emotive characteristics of speech.

[0070] Traditionally, real-time or near real-time language translation has relied on skilled human interpreters, such as those at the United Nations, where delays of several seconds are common. These delays—often 6 to 8 seconds—can disrupt fluid dialogue, hinder diplomatic engagement, and result in missed cultural nuances. Embodiments described herein seek to reduce such latency, supporting more seamless, culturally aware exchanges. By using AI agents to interpret speech in real time, the system may streamline diplomatic conversations and facilitate more effective communication between parties who may otherwise be limited by linguistic or cultural barriers.

[0071] In such settings, interpretation may go beyond literal translation to account for cultural sensitivities and individual speaker profiles. For example, a phrase acceptable in one language may be offensive in another. Embodiments may address this through AI-driven interpretation that adjusts phrasing based on user background, historical communication data, or known sensitivities. While “interpretation” is used here, in context it may refer interchangeably to interpretation, translation, transcription, or transliteration—depending on the nature of the linguistic transformation being performed.

[0072] The “translation” aspect or other aspects of embodiments described herein may be connected, at least in part, to interpretation aspects or to other aspects.

[0073] Again, although “interpretation” is specifically mentioned, depending on context, “interpretation,” “translation,” “transcription,” or “transliteration” may be used interchangeably, even though, at times, just one aspect may be mentioned for efficiency purposes.

[0074] In an example, a communication or presentation may involve Arabic, for example, that may be spoken in the Gulf versus Arabic, for example, spoken in Morocco.

[0075] They’re both Arabic, but may be spoken, written, differently, at least in part.

[0076] Further, different languages may have different loan words, and sometimes loan words may not pass well between languages, for example.

[0077] It may sometimes occur with such loanwords, for example, such as out of the United States where there may exist a larger number of technical names or acronyms, for example, that may be disseminated or otherwise communicated around in the world, at least in part, wherein such terminology may be adopted (e.g., suddenly) by other languages.

[0078] Such circumstances may make it relatively very difficult to interpret, translate, due at least in part to the human interpreters, translators themselves having some sort of domainexperience, subject matter expertise, to have an ability to translate potentially highly technical subjects, for example, to individuals, groups, who may not have similar experience, expertise.

[0079] To address these challenges, at least in part, it may be advantageous to gather content about individuals, groups, organizations, ., with their permission, of course. In an embodiment, content may be collected, such as from LinkedIn, for example, or from other sources. Such example profile content may provide a sense of an individual’s education level, literacy level, languages spoken, and so forth. With such profile content (e.g., data, information), dialogue may be structured to an elevated academic level or a modest academic level, for example, depending at least in part on the individual. Note that may not be an attempt to train the individual, but rather a CICAI may respond to individual(s) involved in terms and abstractions that may be more readily comprehended, for example.

[0080] One or more embodiments includes one or more processing units executing computer-readable instructions to facilitate or support one or more Artificial Intelligence-type processes, operations, or algorithms to, for example, detect that speech is being uttered.

[0081] Responsive, at least in part, to a detection of speech, analog electrical signals representative of incoming audio (e.g., speech) may be received from a microphone or other transducer or audio source, for example, or may be converted to digital signals or signal packets via an analog-to-digital converter (ADC), for example.

[0082] In embodiments, digital signals or signal packets may be collected as one or more “.wav” files or like.

[0083] Also, in one or more embodiments, digital content stored in a .wav file may be converted to a “.txt” text file or like using phenome extraction or analysis techniques or other algorithms to step through a .wav file to isolate individual quanta of utterances (e.g., words, phrases), for example.

[0084] From a .txt file, for example, original speech, having been through phenome extraction or analysis operations, for example, may be translated (or transcribed, interpreted, or transliterated, for example) from speech of an original speaker into text or language of a target listener, for example.

[0085] In doing so, pronunciation of vowels or consonants of target language(s) may be correctly replicated, for example.

[0086] For example, in one or more embodiments, a Unicode conversion, for example, may be made from language A to language B, or a more refined pronunciation of vowels or consonants may be produced, for example.

[0087] The processing units may be part of a switch-type infrastructure that includes memory coupled to the processor and a runtime adapter for instantiating user-specific AI agentsand managing language conversion processes.

[0088] In some embodiments, AI-agent output may be reconstituted as synthesized audio or structured text and routed onward through switch-based infrastructure in minimal time to support session continuity.

[0089] In one or more embodiments, with respect to the “transliteration” aspect, for example, transliteration may advantageously generate phenomes to test whether a way a particular portion of speech was heard (e.g., detected, interpreted, or transcribed) in a manner intended by a speaker (e.g., human individual).

[0090] Again, although “transliteration” is specifically mentioned, depending on context, “interpretation,” “translation,” “transcription,” or “transliteration” may be used interchangeably, even though, at times, just one aspect may be mentioned for efficiency purposes.

[0091] In one or more embodiments, an AI-agent may monitor text, streaming text, streaming analog, or like, which may comprise content being captured by a microphone or other sensors or that may be delivered to CICAI.

[0092] Also, as mentioned, unique profiles may be captured or generated for individuals, in one or more embodiments.

[0093] For example, content regarding which movies an individual may like, where an individual may shop, where an individual may live, and so forth.

[0094] With such content, for example, text that may have been generated, such as from captured audio, for example, may be altered to better correspond with a target listener’s culture, community, education level, socio-economic status, likes, dislikes, or like.

[0095] For example, an individual “T” may be from Texas and an individual “N” may be from New York. To help these two individuals perhaps create a better connection when they converse or otherwise communicate with each other, such as over the telephone, for example, a “y’all” expression by individual T may be transliterated, for example, to “you all” for individual N in New York.

[0096] Similarly, in one or more embodiments, a “you all” expressed by individual N may be transliterated into “y’all” for individual T, for example.

[0097] Other, perhaps more serious or more important, communications, conversations, or presentations, for example, may be adjusted in this manner, such as through example transliteration approaches, in one or more embodiments, to help individuals better communicate one with another or within groups, communities, organization, for example.

[0098] Also, although this particular example is English-to-English, similar transliteration or like operations may be applied to communications where communication between users of different languages may be involved, in one or more embodiments.

[0099] In one or more embodiments, responsive at least in part to uttered speech being detected, such as may be received at a microphone (e.g., incoming digital audio stream received via telecommunications network from a cell phone)—processing threads may be initiated to facilitate operations pertaining to interpretation, transcription, translation, or transliteration, for example.

[0100] In one or more embodiments, processes pertaining to one or more of interpretation, transcription, translation, or transliteration aspects may depend at least in part on values generated by other processes pertaining to one or more others of interpretation, transcription, translation, or transliteration.

[0101] Therefore, some delay (e.g., not perceivable to human individuals) may occur as one process waits for values to be available from another process.

[0102] However, in one or more embodiments, processes pertaining to interpretation, transcription, translation, or transliteration aspects may be performed concurrently.

[0103] As alluded to previously, cultural considerations or the like may affect one or more of the interpretation, transcription, translation, or transliteration aspects of CICAI, in one or more embodiments.

[0104] For example, consider a situation where, culturally, it may be advantageous or desirable to receive speech:

[0105] (a) With a particular tone;

[0106] (b) Seemingly from a particular gender; or

[0107] (c) Seemingly from a particular individual.

[0108] In one or more embodiments, as part of transcription or like operations, diacritics may be added to individual letters as they are transferred from a source language to a target language.

[0109] Therefore, in one or more embodiments, as a word is uttered, the letters may be created and diacritics may be added.

[0110] Such diacritics may reflect:

[0111] (a) Tone;

[0112] (b) Timbre; or

[0113] (c) Other emotional attributes of uttered speech.

[0114] In this manner, emotional content of a moment of speech may be captured via the diacritics. One or more example embodiments comprising diacritics are discussed more fully below.

[0115] One or more embodiments may also find advantageous utility in teleconference settings.

[0116] In such a setting, individual participants may:

[0117] (a) Speak a preferred (or specified) language; or

[0118] (b) Hear their preferred or specified language, regardless of the languages being spoken by others.

[0119] Often in a teleconference, participants may speak over one another—either intentionally or unintentionally—making it difficult to discern what is being said.

[0120] Additionally, it may be generally assumed that a particular language will be spoken.

[0121] One or more embodiments described herein—including the use of AI-agents—may address these types of challenges.

[0122] In one or more embodiments, the use of AI-agents may comprise one or more processing units executing computer-readable instructions to facilitate or support one or more Artificial Intelligence-type processes, operations, or algorithms.

[0123] For example, although a participant in a teleconference may have difficulty understanding what is being said when multiple participants are speaking at the same time, one or more embodiments may, via AI-based approaches such as example approaches described herein, have little difficulty tracking who said what and when it was said. In one or more embodiments, such as in connection with one or more interpretation, translation, transcription, or transliteration aspects, a transcript may be generated showing:

[0124] (a) What was said;

[0125] (b) Who said it; and

[0126] (c) When it was said.

[0127] In one or more embodiments, multiple transcripts may be generated, including:

[0128] (i) A first transcript for a first participant;

[0129] (ii) A second transcript for a second participant; and

[0130] (iii) Transcripts of their respective translated utterances.

[0131] Further, in one or more embodiments, generated transcripts may show:

[0132] (a) What was said in originally spoken languages; and

[0133] (b) What was said in a target language.

[0134] Additionally, in one or more embodiments, separate transcripts may be generated for individual participants, in one or more languages (e.g., source or target languages). In one or more embodiments, interpretation, transcription, translation, or transliteration operations may be involved in such processes.

[0135] Further, for teleconference-type or like circumstances, one or more embodiments described herein, including the use of AI-agents, may affect how a teleconference is presented atvarious computing devices of various participants. For example, in one or more embodiments, individual profile information may inform AI-agents involved in shaping a teleconference presentation, including:

[0136] (a) Historical content related to previous communications;

[0137] (b) Prior teleconferences; or

[0138] (c) Other contextual information.

[0139] As alluded to previously, AI-agents may comprise one or more processing units executing computer-readable instructions to facilitate or support one or more Artificial Intelligence-type processes, operations, or algorithms.

[0140] As a result, in one or more embodiments, a teleconference application (e.g., computer-readable instructions executable by one or more processors) may dynamically alter presentation parameters to focus more on individuals that are contextually important to a session, for example by:

[0141] (a) Altering the size of a display window;

[0142] (b) Adjusting microphone volume; or

[0143] (c) Modifying other audio parameters.

[0144] Additionally, in one or more embodiments, the AI-agent-powered application may monitor or adjust the teleconference presentation in real time, including:

[0145] (i) Muting microphones;

[0146] (ii) Adjusting volume levels; or

[0147] (iii) Configuring display windows (e.g., count, size, or participant ordering).

[0148] In some implementations, participants may replay portions of the stream. For example, a participant may back up or rewind the stream by a few seconds to focus on what another participant said. Replay control may target specific speakers. In this way, speech from overlapping participants can be disambiguated. In one or more embodiments, video and audio playback may be managed by processing units configured for teleconference stream storage and retrieval.

[0149] Any or all of the above-described teleconference features may incorporate aspects of the CICAI system including real-time or near real-time interpretation, transcription, translation, or transliteration. These linguistic transformations may be performed on captured audio, synthesized speech, or textual transcripts across participants.

[0150] Further, in one or more embodiments, ubiquity of service in a telecommunication network may be another aspect that may be quite advantageous or that may provide a significant improvement to the state of the art. In one or more example embodiments, such as may be discussed below, CICAI functionality, including the interpretation, translation, transcription, ortransliteration aspects, may be implemented, in whole or in part, in telecommunications switches (e.g., class-4 Telco switches). In one or more embodiments, an already-existing infrastructure may be utilized, in whole or in part, to provide real-time or near real-time services to allow people from around the globe to much more readily or effectively, for example, communicate with other people from around the globe. One or more such embodiments are discussed more fully below, for example in connection with FIGS.17-18.

[0151] “Artificial Intelligence” (AI) or the like refers to a software or hardware system configured to interact with its environment at least in part by processing input or generating output, attributing meaning, making decisions or eliciting actions to resolve a problem or to achieve a desired state shift. “Agent,” “AI-agent,” or the like refers to an AI entity capable of acting autonomously or independently or configured to act autonomously or independently.

[0152] In one or more embodiments, CICAI may comprise, for example, a communications network, such as a multi-agent network, as one particular example, of one or more processing units executing computer-readable instructions to facilitate or support one or more Artificial Intelligence-type processes, operations, or algorithms.

[0153] In one or more embodiments, CICAI may also comprise one or more neural-network algorithms or other approaches using, for example, Systolic Arrays or Wavefront-Arrays of parallel processing units, for example, using synchronous or asynchronous nodal communication, in some circumstances, for example.

[0154] In one or more embodiments, CICAI may be implemented, in whole or in part, in a mesh-type network configuration, for example, in a TRILL or Shortest Path Bridging network topology, for example, which may be one or many of the following Layer 2 Link-state switching architectures such as, but not limited to, cell-switched, frame-switched or packet-switched / routed network architecture comprising one, or a plurality of, computers, networking equipment, Telco Equipment comprising one or more multi-core microprocessors, FPGAs, SOCs and ASICs, for example.

[0155] In one or more embodiments, an advantage of a parallel-processing-type multi-agent-type network implemented, in whole or in part, in a Layer 2 network mesh-network topology, such as TRILL, may be plug-n-play nature.

[0156] For example, a network administrator may be relieved of heavy configuration, unlike in a Layer 3 network, in some circumstances.

[0157] In one or more embodiments, TRILL may achieve this with a Dynamic Resource Allocation Protocol (DRAP), for example, where individual nodes derive its own nickname and a protocol may help ensure no duplicity.

[0158] In one or more embodiments, a configuration involvement of TRILL is relativelysmall (e.g., minimal), for example.

[0159] In some implementations, the AI-agent runtime adapter may include a language model abstraction layer that accesses multiple language models using a publish-and-subscribe architecture to facilitate dynamic model selection and real-time inference.

[0160] In one or more embodiments, a plug-n-play CICAI TRILL-type network may include a plurality of AI-agents executing computer instructions across a plurality of processing units interconnected via a Transparent Interconnection of Lots of Links (TRILL) network switch.

[0161] In one or more embodiments, such TRILL network switches may operate at OSI Layer 2 and may be configured, in whole or in part, as asynchronous wavefront-arrays to enable Runtime Adapters for one or more AI-agents acting on behalf of a sender or receiver.

[0162] In one or more embodiments, AI-agents may perform one or more of the following in-stream-type operations in real-time:

[0163] (a) Interpretation;

[0164] (b) Translation;

[0165] (c) Transcription; or

[0166] (d) Transliteration.

[0167] In one or more embodiments, such operations may include, for example, generating subtitles in a language of the receiver or providing voice-over synthesis.

[0168] In one or more embodiments, a plurality of language requests may be handled concurrently across different geographies via the same TRILL-based CICAI infrastructure.

[0169] In one or more embodiments, due at least in part to asynchronous time delays inherent in long-distance communication, the use of wavefront-arrays may be particularly effective for implementing CICAI globally.

[0170] In one or more embodiments, wavefront-arrays may be substituted, in whole or in part, with systolic-arrays when deployed in Metropolitan Area Networks (MANs) due to reduced distances and time delays and increased packet, cell, or frame switching speeds.

[0171] In one or more embodiments, “real-time” refers to processing events—such as interpretation, translation, transcription, or transliteration—at a speed sufficient to avoid perceptible latency to a human user.

[0172] In one or more embodiments, “near real-time” refers to processing events with a delay perceptible to a human individual, such as a delay of less than 250 milliseconds.

[0173] More generally, “real-time” or “near real-time” may refer to processes occurring approximately concurrently with the associated real-world events or inputs.

[0174] In one or more embodiments, such real-time or near real-time linguisticprocessing may be performed within telecommunications-class switches as digital audio is received.

[0175] In one or more embodiments, AI-agents may be instantiated and operated via a Runtime Adapter within such switches to minimize latency experienced by users.

[0176] In one or more embodiments, CICAI may include one or more TRILL-based networks comprising networked computing resources.

[0177] In one or more embodiments, such computing resources may include one or more of the following:

[0178] (a) Computer memory;

[0179] (b) Processor farms;

[0180] (c) Optical interconnect buses, such as Infiniband;

[0181] (d) Solid-state storage devices (e.g., NVMe); or

[0182] (e) Mechanical storage devices.

[0183] In one or more embodiments, these resources may execute computer instructions that codify AI-agents, including:

[0184] (a) Artificial Intelligence Algorithms as described herein;

[0185] (b) Public domain neural network algorithms;

[0186] (c) Supervised or unsupervised machine-learning algorithms;

[0187] (d) Computer vision algorithms; or

[0188] (e) Artifact detection algorithms.

[0189] In one or more embodiments, CICAI may enable a user to dial any telephone number globally using standard telephony infrastructure.

[0190] In such embodiments, a network may provide in-stream interpretation, translation, transcription, or transliteration services.

[0191] In one or more embodiments, such services may help eliminate non-tariff language barriers in contexts such as:

[0192] (a) International commerce;

[0193] (b) Personal communication;

[0194] (c) Group communication; or

[0195] (d) Cross-cultural text communication.

[0196] In one or more embodiments, the disclosed method may be executed within a telecommunications switch infrastructure configured to support digital stream interception and linguistic transformation within the network core.

[0197] In one or more embodiments, CICAI may provide language interpretation, translation, transcription, or transliteration services as a foundational service of the networkitself. These services may be embedded directly into telecommunications infrastructure such that language transformation occurs natively within session transport and switching systems.

[0198] In one or more embodiments, the Transport Layer of a CICAI Network may utilize one or more example protocols to enable integrated linguistic services across various transmission media. These protocols may operate jointly or severally and may be tailored for different communication contexts, including real-time voice, email, document processing, and web content.

[0199] Example protocols include, but are not limited to:

[0200] (a) ARP – Address Resolution Protocol;

[0201] (b) ATM – RSVP, VPC with bundled VCCs;

[0202] (c) BGP – Border Gateway Protocol;

[0203] (d) Bluetooth – SDP, TCS, AVCTP, OBEX, LMP, BNEP, RFCOMM;

[0204] (e) DRAP – Dynamic Resource Allocation Protocol;

[0205] (f) DNS – Domain Name System;

[0206] (g) DHCP – Dynamic Host Configuration Protocol;

[0207] (h) FTP – For document interpretation and translation;

[0208] (i) HSRP – Hierarchical Satellite Routing Protocol;

[0209] (j) HTTP – For webpage interpretation and translation;

[0210] (k) ICMP – Internet Control Message Protocol;

[0211] (l) ISL – Inter-Switch Links;

[0212] (m) LoRaWAN – LPWA multicast protocols for IoT;

[0213] (n) OSPF – Open Shortest Path First;

[0214] (o) PDH – Plesiochronous Digital Hierarchy;

[0215] (p) RIP – Routing Information Protocol;

[0216] (q) SGRP – Satellite Grouping and Routing Protocol;

[0217] (r) SMTP – For interpretation and translation of emails;

[0218] (s) SONET / SDH – OC-2 through OC-192; STM-1 units;

[0219] (t) STP – Satellite Transport Protocol;

[0220] (u) TCP / IP – Designed by the U.S. DoD;

[0221] (v) Telnet – Plain Old Telephone System;

[0222] (w) UDP – User Datagram Protocol;

[0223] (x) WiFi / WiMax / LTE / 4G / 5G / 6G.

[0224] The foregoing may, in-part or whole, describe one or more embodiments comprising artificial intelligence-based Network Services implemented, in whole or in part, within Space-Air-Ground Integrated Networks (SAGINs) for use by any or all devices,terrestrial or otherwise, connected thereto; including, for example, deep-space relay satellites for communication to future colonies, for example. In one or more embodiments, with CICAI implemented, in whole or in part, in a systolic-array or a wavefront-array to provision autonomous-type AI-agents, for example, with Runtime Adapters, CICAI infrastructure, due at least in part to its smaller footprint in at least some circumstances or its modest power or cooling requirements, for example, may be lifted into orbit (e.g., cloud-server inside LEO / MEO / GEO / DSR satellites) to provide telecom-type services to SAGINs layers, for example, within a LEO, MEO, or GEO configuration, for example. See, for example, FIG.1. For example, such a configuration among LEO Cubesats may accelerate removal of non-tariff inefficiencies globally because “Computational Interpretation of Communication using Artificial Intelligence Algorithms” may be performed in handheld 4g & 5g cell-phones, for example, without a need for terrestrial cell-towers requiring optical, copper cables to connect them to large regional land masses, for example.

[0225] In one or more embodiments, a CICAI Network implemented, in whole or in part, within SAGINs may comprise, but is not limit to, example permutations of network architectures that may be used in state-of-the-art SAGINs, for example: (a) Optical Channel Layer of OTN Model (Layer 1); (b) Hybrid RF / Optical Link Layer Model (10+Gbps laser-uplink to MEO / GEO Satellite with high-rate Ka-band or Ku-band downlinks; (c) Data Link and Network Layers of ATM Model; (d) Data Link and Network Layers of OSI Model (Open Systems Interconnection Model); (e) Network Interface Layer of TCP / IP model and; (f) Link and Application Layers of Low Power Wide Area (LPWA Models) of networking; (g) OSBN / SBN - Optical-switched Broadband / Satellite Broadband Network which may possess inter-satellite multi-Gbps optical crosslinks with up to 6,000 km range each; (h) TRILL Data Link Layer of OSI Model (Open Systems Interconnection Model).

[0226] An example infrastructure, such as depicted in FIG.1, for example, may also be considered in connection with FIG.20, discussed below. In one or more embodiments, example infrastructures of FIG.1 and FIG.20 may share at least some characteristics. Of course, subject matter is not limited in scope in these respects.

[0227] In one or more embodiments, a CICAI network may operate in any of one or more permutations of peer-to-peer, one-to-many, many-to-one, or many-to-many configurations, for example. See FIGS. 2-5. FIGS. 3A-3D, for example, depicting an example network operating in a one-to-many configuration wherein individuals speaking Arabic, English, Japanese, or Mandarin, for example, may be able to communicate one with another in their respective languages with little or no perceivable delay, and while maintaining the appropriate context, sentiment, intention, emotion, .

[0228] One or more embodiments may describe one, or a plurality of, unique networked computing models in the field of ‘Computational Linguistics’ or, more specifically, “Computational Interpretive Communication using Artificial Intelligence Algorithms” (CICAI). CICAI may comprise a parallel-computing-type architecture implemented in hardware or software to support AI-agent deployment across diverse instruction / data stream paradigms.

[0229] In one or more embodiments, CICAI may utilize a Single Instruction Stream, Single Data Stream (SISD) model. For example, this may involve uni-core, uni-cell, or uni-node processing units. Software microkernel-based agents or heterogeneous multi-agent-type architectures may be applied advantageously within SISD frameworks to facilitate localized, streamlined language processing operations.

[0230] In another embodiment, CICAI may leverage Multiple Instruction, Single Data (MISD) architecture. MISD may include multi-core or multi-node processing units performing distinct operations on the same data stream. Example applications may include voice interpretation pipelines, real-time translation, or space / military systems such as drones or hypersonic telemetry processing. Inter-agent communication in such contexts may be managed via intra- or inter-process communication (IPC).

[0231] CICAI may further utilize Multiple Instruction, Multiple Data (MIMD) models. These parallel computing frameworks allow multiple units to operate asynchronously on different data streams. MIMD architectures may be beneficial in computer-aided design, telecom switches, or simulation workloads. They may rely on shared memory, distributed memory (e.g., mesh, hypercube), or both. CICAI agents within MIMD may be microkernel-based, heterogeneous, and communicate via hardware or software IPC.

[0232] In yet another embodiment, CICAI may use Single Instruction, Multiple Data (SIMD) architecture. In SIMD, multiple processing elements perform the same operation across multiple data points simultaneously. This model may be especially useful where systolic arrays are deployed for high-throughput parallelism. CICAI-compatible AI-agents—homogeneous or heterogeneous—can leverage internal or external IPC to enable efficient interpretation and transformation pipelines under SIMD.

[0233] In one or more embodiments, CICAI may be implemented, in whole or in part, within a system architecture affording mixed permutations of MISD, SISD, MIMD, or SIMD-conforming paradigms. The selected compute model may depend on a particular application, use case, or deployment constraints. CICAI may often adopt a MIMD-type approach, but comparable functionality may be derived from any combination of SISD, MISD, SIMD, or MIMD.

[0234] In use cases where dedicated hardware support for SIMD, MISD, MIMD, orSISD is not available, CICAI may be implemented, in whole or in part, using a SWAR (SIMD Within A Register) architecture. This may involve general-purpose processors comprising one or more cores, cells, or nodes lacking SIMD micro-instructions in hardware, yet still capable of performing SIMD-like operations through software emulation.

[0235] A key difference between SIMD and SWAR architectures is that SIMD-capable processors typically include microcoded instructions accessible via low-level programming (e.g., assembler, C / C++) for true hardware-based parallel operations. In contrast, SWAR relies on general-purpose processors that lack these native SIMD instructions but can still manipulate data stored in sub-words or fields within registers to simulate SIMD functionality.

[0236] In one or more embodiments, SWAR may be applied in heterogeneous computing environments, including those using Harvard or Princeton architecture models. Such environments may allow single-core processors to execute CICAI’s multi-agent or heterogeneous agent instructions asynchronously. SWAR’s asynchronous execution model may be well-suited for parallel interpretation, transcription, or transformation tasks, even when agents operate at different instruction execution clock speeds.

[0237] In one or more embodiments, CICAI may comprise a parallel-processing multi-agent-type architecture that allows for mixed permutations of Runtime Adapters implemented, in whole or in part, as processing unit instructions.

[0238] In one or more embodiments, such Runtime Adapters may operate within parameters of one or more of the following architectures: Multiple Instruction Single Data (MISD), Single Instruction Single Data (SISD), Multiple Instruction Multiple Data (MIMD), or Single Instruction Multiple Data (SIMD), depending on a particular use case or application.

[0239] In one or more embodiments, MISD, SISD, MIMD, or SIMD-type architectures may not support Runtime Adapters being implemented directly on individual processing cores, processing cells, or processing nodes of parallel computing-type processors.

[0240] Therefore, in one or more embodiments, it may be advantageous to implement, in whole or in part, a hierarchical AI-agent-type architecture in a monolithic operating system such as CentOS Linux or an embedded operating system platform.

[0241] In one or more embodiments, CICAI hardware may implement, in whole or in part, a “Secure Boot” mechanism designed to counter boot-time malware threats.

[0242] In one or more embodiments, Secure Boot may operate at least in part by cryptographically verifying firmware, kernels, or drivers before execution.

[0243] In one or more embodiments, each executable boot binary, including firmware and software components, may be checked against one or more of:

[0244] (a) A blacklist database; and / or

[0245] (b) Multiple trusted signature databases.

[0246] In one or more embodiments, the Secure Boot mechanism may prevent execution of unapproved operating system images, component controllers, unauthorized boot methods, boot-time malware (e.g., LoJax), rootkits, outdated kernels, obsolete or malicious drivers, or similar unauthorized elements.

[0247] In one or more embodiments, operating system kernels that CICAI depends upon, or operates within, may extend Secure Boot validation into the Runtime Adapter environment.

[0248] In such embodiments, this may include checking signed drivers or executables that possess root privileges, thereby helping to block off-the-shelf exploit tools such as Mimikatz or Metasploit.

[0249] In one or more embodiments, one, or a plurality of, a) hardware-based or software-based AI-agents operating autonomously while communicating asynchronously as a ‘wavefront-array, or b) hardware or software-based AI-agents operating autonomously while communicating synchronously as a ‘systolic-array, or c) a heterogeneous or multi-agent architecture wherein the agent may, or may not, be AI-agents, operating asynchronously, empowered with a myriad of stochastic methods and neural network computational techniques commonly used in the construction of Large Language Models (LLMs), performing machine-learning, deep-learning, or managing operational work-flows pertaining to the forgoing. See, for example, FIG.6.

[0250] In one or more other example embodiments, CICAI may be implemented, in whole or in part, using one or more microkernels. These may be defined as a reduced or minimal set of programmable, executable computer-readable instructions suitable for implementing an operating system or similar control framework. See FIG.7, for example.

[0251] Such microkernel mechanisms may include lower-level address space management, partitioned into kernel-space and user-space memory. They may also support POSIX threads, Green threads, processor hyper-threads, process thread management, and inter-process communication (IPC), among other threading and memory handling models.

[0252] In one or more embodiments, CICAI microkernels—capable of executing in both SIMD and SWAR modes—may be deployed in high-security environments, such as those using KeyKOS or EROS. These implementations may be targeted at military-grade or cyber-secure systems.

[0253] Some CICAI implementations may comply with international standards such as “Common Criteria for Information Technology Security Evaluation” (Common Criteria or CC), known under ISO / IEC 15408. CICAI may support high-assurance levels, such as EAL4 through EAL7. In some contexts, the architecture may be optimized to be “simple,” as defined under theDepartment of Defense Trusted Computer System Evaluation Criteria (B3 / A1 levels): “The TCB shall use complete, conceptually simple protection mechanisms with precisely defined semantics…”

[0254] In further embodiments, CICAI may implement one or more microkernel-based AI agents operating across systolic-array or wavefront-array architectures. IPC between agents may occur via synchronous or asynchronous channels, including cyber-secure publish–subscribe message buses configured to implement FIFO or LIFO buffers.

[0255] In one or more embodiments, microkernels may be utilized, at least in part, in conjunction with one or more AI-agents, for example.

[0256] In one or more embodiments, an example microkernel configuration may include a kernel-space operation configured to maintain an inventory of software components currently in operation and software components that should be in operation.

[0257] In one or more embodiments, if a deviation exists between the two lists—i.e., between components in operation and components that should be in operation—a microkernel boot-loader-type AI-agent (referred to herein as “HuroBOSS”) may perform supervisory actions.

[0258] In one or more embodiments, such supervisory actions performed by HuroBOSS may include one or more of the following:

[0259] (a) Scheduling kernel or process threads for startup, restart, or termination;

[0260] (b) Maintaining inventory of AI-agent provisioning for performance of specified actions; or

[0261] (c) Managing content within a management information base (MIB), for example.

[0262] In one or more embodiments, a microkernel boot-loader may also be responsible for initializing system hardware or loading the microkernel into memory.

[0263] In one or more embodiments, a boot-loader-type AI-agent may further include capabilities to detect system-level faults or attempt a microkernel restart in the event of failure.

[0264] In one or more embodiments, during microkernel boot-up, an init-type process may comprise a specialized process launched by the microkernel at initialization.

[0265] In one or more embodiments, such an init process may be responsible for initializing the system by starting user-space services or essential kernel daemons.

[0266] In one or more embodiments, if an issue is detected by the init process involving the microkernel, the init process may be unable to perform its designated tasks, or the system may halt execution.

[0267] In one or more embodiments, microkernel-based architectures may rely, at least in part, on redundancy mechanisms applied to critical system components.

[0268] In one or more embodiments, a system may include multiple boot-loaders orredundant init processes to enhance overall fault tolerance.

[0269] In one or more embodiments, external monitoring tools may be employed to observe the operational health of the system, including the microkernel.

[0270] In one or more embodiments, a HuroBOSS agent may provide monitoring services, restart capabilities, or supervisory control over threads and system states.

[0271] One or more embodiments may comprise one or more of the following example AI-agent types. Of course, subject matter is not limited in scope in these respects.

[0272] In one or more embodiments, Homogeneous Reflexive AI-agents may respond autonomously to changes in state within their own Runtime Adapter or within a Runtime Adapter of other Homogeneous Reflexive AI-agents, for example.

[0273] Reference is made to FIG.8, which illustrates aspects of Homogeneous Reflexive AI-agent behavior in one or more embodiments.

[0274] In one or more embodiments, Homogeneous Reflexive AI-agents may be implemented, in whole or in part, as one or more permutations of processing units utilizing Single Instruction Single Data (SISD) or Multiple Instruction Single Data (MISD) architectures.

[0275] In one or more embodiments, such processing units may be incorporated into a Runtime Adapter in which an AI-agent’s next programmatically defined action may be determined using reflective software techniques.

[0276] In one or more embodiments, reflective software techniques may be used to compare software instructions and their corresponding programmatic states with telemetry obtained from a relevant environment.

[0277] In one or more embodiments, Homogeneous Reflexive AI-agent implementations may include, but are not limited to, asynchronous wavefront arrays executing software instructions on SISD or MISD architectures.

[0278] In one or more embodiments, Homogeneous Reflexive AI-agents may be written in assembler or C / C++ programming languages due at least in part to the complexity involved in manipulating data stored in processing unit registers.

[0279] In one or more embodiments, Heterogeneous Reflexive AI-agents may respond autonomously to changes in state within a collection of Runtime Adapters operating across heterogeneous hardware or software agent architectures, for example.

[0280] In one or more embodiments, Heterogeneous Reflexive AI-agents may be implemented, in whole or in part, using one or a plurality of mixed permutations of asynchronous wavefront arrays.

[0281] In one or more embodiments, such implementations may be configured using processing units based on Single Instruction Single Data (SISD) or Multiple Instruction SingleData (MISD) architectures.

[0282] In one or more embodiments, Runtime Adapters may utilize reflexive software logic to determine a next programmatically defined action by comparing predetermined software instructions or states with environmental telemetry, for example.

[0283] In one or more embodiments, such AI-agents may be written in assembler or C / C++ programming languages due at least in part to the complexity of managing asynchronous First-In-First-Out (FIFO) buffers, for example.

[0284] In one or more embodiments, Model-based Reflexive AI-agents may monitor one or more Runtime Adapters operating under conditions that may not be immediately perceptible using unsupervised learning algorithms, for example.

[0285] In one or more embodiments, a Model-based Reflexive AI-agent may autonomously or programmatically analyze or modify the configuration of a Runtime Adapter, including but not limited to:

[0286] (a) Its own configuration;

[0287] (b) One or more configurations of nearby Runtime Adapters;

[0288] (c) Network configuration settings; or

[0289] (d) Operating system parameters.

[0290] In one or more embodiments, such analysis may include algorithmic inference or insertion of missing data required to complete a determination of next-step actions, for example.

[0291] In one or more embodiments, Model-based Reflexive AI-agents may be implemented, in whole or in part, using memory-safe programming languages, as alluded to previously.

[0292] In one or more embodiments, a Model-based Reflexive AI-agent may operate within a Runtime Adapter without perfect or complete telemetry sufficient to measure all parameters of its runtime environment.

[0293] In one or more embodiments, the current state of the agent may be stored within the agent in an encrypted hash structure, which may describe a portion of the world that is not directly observable.

[0294] Reference is made to FIG.9, which illustrates aspects of a model-based reflexive AI-agent, in one or more embodiments.

[0295] In one or more embodiments, Goal-based AI-agents may be implemented, in whole or in part, using memory-safe programming languages as discussed herein.

[0296] In one or more embodiments, a Goal-based AI-agent may execute software instructions to explore permutations of decision-tree paths and autonomously select instructions based at least in part on stochastic evaluation.

[0297] In one or more embodiments, such evaluation may include execution of:

[0298] (a) All possible decision-tree paths; or

[0299] (b) A subset of paths selected according to probability thresholds or other optimization goals.

[0300] In one or more embodiments, Goal-based AI-agents may autonomously construct plans or execute prescribed programmatic instructions to achieve specified goals.

[0301] In one or more embodiments, goals may be predefined or may be dynamically derived during execution, in whole or in part.

[0302] Reference is made to FIG.10, which depicts Goal-based AI-agent planning in one or more embodiments.

[0303] In one or more embodiments, Mobile Cybersecurity AI-agents may comprise software-based autonomous agents implemented, in whole or in part, in memory-safe programming languages (e.g., Python®, Java®, C#, Go, Delphi / Object Pascal, Swift®, Ruby™, Rust®, Ada). These agents may perform continuous vulnerability scanning, port-scanning, memory analysis, dark web monitoring, and penetration testing to validate runtime compliance with standards such as SOC2, HIPAA, HITRUST, NIST, or CIS.

[0304] Unlike conventional external penetration tests, these agents may operate “inside the jar,” meaning from within the Runtime Adapter or AI-agent Control Unit. They programmatically evaluate system state and may determine whether a “zero-trust” environment has been achieved using static or dynamic application security testing (SAST / DAST). This is in contradistinction to outside-in, broad-spectrum tests that lack systemic context.

[0305] Security threats targeted by such agents may include VM side-channel attacks, VM escape, and rootkits. Traditional countermeasures include expensive firewalls, noisy scheduler-based isolation, or centralized intrusion detection — each of which can suffer from cost, performance, or targetability limitations.

[0306] Mobile Cybersecurity AI-agents may act as distributed behavioral sensors. In some embodiments, they may initiate non-cooperative game interactions with suspicious entities, calculate Nash equilibrium utility values, and differentiate between legitimate and adversarial requests. Such game-theoretic frameworks also help assess attack severity or origin.

[0307] In certain simulations or experimental trials, this agent-based cybersecurity framework demonstrated up to 86% attack detection accuracy. See FIG.11, for example, for an illustrative use case or system interaction model.

[0308] In one or more embodiments, Utility-based AI-agents may execute computer-readable processing unit instructions to achieve a goal based at least in part on a utility measure.

[0309] In one or more embodiments, the utility measure may comprise one or more combinations of thread utilization, memory usage, processing-unit-clock-cycle consumption, or various network-related performance metrics, for example.

[0310] In one or more embodiments, utility-based AI-agents may execute processing unit instructions to stochastically predict future outcomes of multiple possible actions associated with divergent goals.

[0311] In one or more embodiments, utility-based AI-agents may algorithmically select one or more sets of processing unit instructions, such as an advantageous or optimal set, to achieve a goal, for example.

[0312] In one or more embodiments, utility-based AI-agents may be implemented in memory-safe programming languages.

[0313] In one or more embodiments, memory-safe languages may comprise those referred to previously herein, which may restrict or prevent memory access violations or undefined behavior during execution.

[0314] In one or more embodiments, CICAI platform features may alternatively or additionally be implemented using assembler or C / C++ software, for example.

[0315] In one or more embodiments, such assembler or C / C++ implementations may be written to access processing unit resources while minimizing or avoiding interference with clock-cycle timing, for example.

[0316] Reference is made to FIG.12, which illustrates aspects of Utility-based AI-agent execution in a CICAI platform, in one or more embodiments.

[0317] Unsupervised or Supervised Learning AI-agents. In one or more embodiments, Unsupervised, or Supervised Learning AI-agents may adapt or evolve algorithmically defined strategies to maximize “rewards,” for example.

[0318] In one or more embodiments, Unsupervised, or Supervised Learning AI-agents may refine (e.g., continuously or near-continuously) user-profiles responsive at least in part to changes in user characteristics, attributes, skills, or may alternatively invoke a different form of AI-agent to perform a specific purpose which may deviate from a scope of an Unsupervised or Supervised Learning AI-agent, for example.

[0319] In one or more embodiments, AI-agents may be implemented in “memory-safe languages,” such as alluded to previously.

[0320] In one or more embodiments, at least some Unsupervised or Supervised Learning AI-agents may comprise assembler or c / c++ software, for example, written to access processing unit resources without interfering with clock-cycle timings, for example.

[0321] In one or more embodiments, learning, in the context of AI-agents, may allowagents to initially operate in unknown environments to build a knowledge-base about its AI-agent Runtime Adapter autonomously, for example.

[0322] In one or more embodiments, a processing unit (e.g., core, cell, node) may act as a “learning element” responsible for making improvements or may act as a “performance element” responsible for selecting external actions, for example.

[0323] In one or more embodiments, a “learning element” uses feedback from a “critic” on how an agent is doing or determines how a performance element (“actor”) may be modified to do better in the future, for example.

[0324] In one or more embodiments, a performance element may comprise what may have at times been considered to be an entire agent, for example.

[0325] In one or more embodiments, it may take in telemetry to build ad-hoc perceptions based at least in part on reflexive techniques and decides on actions. See FIG.13, for example.

[0326] Multi-agent AI-agent Systems (MAAIS). In one or more embodiments, Multi-agent AI-agent Systems (MAAIS) may comprise a system to execute computer readable processing unit instructions for a spectrum of agent architectures which may autonomously orchestrate, substantially, or completely without human intervention, recruitment of resources or agents to achieve goals via systolic-arrays or wavefront-array architectures, for example, to achieve self-determined goals expressed as programmatic instructions, for example.

[0327] In one or more embodiments, MAAIS may be advantageously used for complex or long-duration computing tasks involving multiple different types of agents working in parallel or in sequence as the case may be, where scheduling of processing tasks may be orchestrated to complete among a myriad of agents simultaneously, for example.

[0328] MAAIS may be implemented in “memory-safe languages,” such as alluded to previously.

[0329] In one or more embodiments, Unsupervised, or Supervised Learning AI-agents may comprise assembler or c / c++ software, for example, written to access processing unit resources without interfering with clock-cycle timings, for example. See FIG.14, for example.

[0330] In one or more embodiments, Personal AI-agents (pAI-agents) may comprise autonomous agents designed for integration with Generative AI Language Models stored in vector databases, email servers, SMS platforms, web crawlers, newsfeeds, or other user-specific content sources. These agents may interact with personalized language models (pLLMs) to interpret and respond to queries on behalf of the user.

[0331] Access to pAI-agents and pLLMs may be controlled in part by Mobile Cybersecurity AI-agents, which may implement three-authentication-factor (3AF) mechanisms. Such controls can help prevent unauthorized access, particularly in scenarios involvinghierarchical blockchain wallets or similar cryptographic access systems.

[0332] pAI-agents may be implemented using memory-safe programming languages (e.g., Python, Java, Swift, Rust), or, in performance-critical environments, via lower-level languages such as Assembler or C / C++. These implementations may access processing unit resources directly without disrupting clock-cycle timing or real-time responsiveness.

[0333] In one or more embodiments, pAI-agents perform Generative AI Textual Synthesis (GATS) using vector-based Language Models and deep learning techniques. They may serve as NLU filters to convert user queries into precise machine-interpretable actions personalized to individual linguistic traits rather than generalized models.

[0334] pAI-agents may maintain context across multiple user interactions, supporting persistent memory over time series. This capability facilitates adaptive behavior based on historical interactions, allowing the agent to evolve according to fluctuating user needs or behavioral patterns.

[0335] In one or more embodiments, pAI-agents may support stream-of-consciousness-style interaction, enabling coherent conversational flow. These agents may maintain contextual awareness to avoid repetition, illogical phrasing, or disjointed transitions—enhancing the user experience. See FIG.15, for example.

[0336] Hierarchical AI-agents. In one or more embodiments, Hierarchical AI-agents may be structured in a multi-level hierarchical framework of top-down control, for example, wherein an AI-agent may be programmatically assigned a kernel-level interrupt-thread of control over to algorithmically manage multiplexed threads sharing a single POSIX or ‘Green’ thread, for example, to programmatically direct lower-level agents with new goals, axioms, or policies, for example.

[0337] In one or more embodiments, individual levels in a hierarchy may have specific roles or responsibilities, thereby contributing to an overall goal, for example.

[0338] Hierarchical AI-agents may benefit larger-scale systems where tasks may be broken down or managed at different levels, for example.

[0339] In one or more embodiments, individual levels in such a hierarchy of AI-agents may be equivalent, in at least some aspects, to multi-layer mesh-networks, for example, in which systolic-arrays and wavefront-arrays may be represented.

[0340] In one or more embodiments, Hierarchical AI-agents may be implemented, in whole or in part, in one or more “memory-safe languages,” such as alluded to previously.

[0341] In one or more embodiments, Hierarchical AI-agents may comprise assembler or c / c++ software, for example, written to access processing unit resources without interfering with clock-cycle timings. See again FIG.15.

[0342] In one or more embodiments, Trusted Platform Module (TPM) AI-agents may provide credential storage or system integrity features using encrypted hierarchical blockchain wallets stored in a distributed fashion within the boot image. These agents may store boot-time integrity measurements in TPM-accessible secure storage to support trusted computing operations.

[0343] TPM AI-agents may include programmable instructions for random number generation, cryptographic key generation, and secure memory usage. These functions may comply with security standards such as FIPS 140-2 or later, in one or more embodiments.

[0344] TPM AI-agents may be implemented in two main forms: (a) As discrete CICAI TPM AI-agent chipsets affixed to motherboards that support encryption algorithms or TPM-bound instructions; (b) As firmware-based TPM agents (fTPM AI-agents), integrated into processors (e.g., Intel® PTT, AMD® fTPM)

[0345] CICAI-compatible monolithic operating systems may include variants of Linux, Unix, or POSIX-compliant systems. These systems may support TPM kernel monitoring via Integrity Measurement Architecture (IMA), and data-at-rest encryption via Linux Unified Key Setup (LUKS), particularly within TRILL network environments. Examples include Red Hat Enterprise Linux® (RHEL) and CentOS Linux distributions.

[0346] TPM AI-agents may be implemented in memory-safe programming languages (as described earlier), or in lower-level languages such as Assembler or C / C++. In some embodiments, CICAI TPM agent microcode may be deployed on field-programmable gate arrays (FPGAs), system-on-chip (SoC) devices, or application-specific integrated circuits (ASICs). See FIG.16.

[0347] The following discussion provides additional definitions or descriptions of various aspects of various embodiments.

[0348] In one or more embodiments, an AI-agent may describe processing unit instructions that are triggered by software or hardware to perform operations across different processor architectures, such as Single Instruction Single Data (SISD), Multiple Instruction Single Data (MISD), Single Instruction Multiple Data (SIMD), or Multiple Instruction Multiple Data (MIMD) architectures, for example. In one or more embodiments, such instructions may be executed within a context of Computer-Integrated Cognitive Artificial Intelligence (CICAI).

[0349] In one or more embodiments, AI-agents may be configured to operate in deterministic or non-deterministic environments, such as environments involving mathematical modeling, optimization of rewards or values, or systems based on reinforcement learning frameworks, for example.

[0350] In one or more embodiments, AI-agents may autonomously navigate the internetas part of an automated workflow, may interact with external software applications via exposed application programming interfaces (APIs), or may process large-scale datasets in relation to Small Language Models (SLMs) or Large Language Models (LLMs), for example.

[0351] In one or more embodiments, AI-agents may be configured to conduct financial transactions autonomously without human intervention or may simulate behaviors and speech patterns of a human being to act as a functional proxy or representative of a human being, for example.

[0352] In one or more embodiments, an AI-agent may comprise a hardware-based, software-based, or hybrid computing entity that is capable of executing programmable instructions. Such instructions may reside in the AI-agent’s memory space and may include, without limitation, instructions for:

[0353] (a) Supervised learning, such as regression or classification tasks based on labeled data.

[0354] (b) Unsupervised learning, such as clustering or dimensionality reduction on unlabeled data.

[0355] (c) Reinforcement learning, such as agent-environment interaction schemes involving positive or negative reward systems.

[0356] In one or more embodiments, reinforcement learning may mean different things over a time sequence depending upon the types of agents and pursuit of learning paradigms. For example, in a “Policy approach” an AI-agent may comprise a self-determinant processing unit which may determine a next action taken by the agent with ‘a priori’ provisioned policies or axioms, without external instructions (e.g., autonomy).

[0357] In another example, AI-agents may operate within a definition of a framework wherein telemetry from within its execution environment, or dynamics of change within that environment, may be rapidly changing. Thus, an AI-agent responds to external stimulus determined through stochastic evaluation of telemetry within an environment or receives a virtual “reward”. In yet another example of ‘Reinforcement Learning’, an AI-agent may be employed to perform “value pursuit” which may be different from a “reward”-based pursuit described above, for example. In value-seeking AI-agents, the “value” is the result of many actions over a long time-series, for example.

[0358] In one or more embodiments, AI-agents may perceive changes in Runtime Adapter conditions using reflective or reflexive software techniques. These agents may analyze telemetry signals emitted from the adapter, enabling real-time responsiveness to fluctuations. Example data streams may include stock market price changes, e-commerce user behavior, or biometric / telemetry data from astronauts.

[0359] In one or more embodiments, AI-agents may autonomously respond to rapid or abnormal changes in Runtime Adapter state. Their behavior may be governed by algorithmic policy encoded in software instructions, or axioms interpreted by script engines. These actions may aim to stabilize or optimize Runtime Adapter operations.

[0360] In other embodiments, AI-agents may interrogate complex datasets, such as vector databases, graph databases, or NOSQL stores. These agents may identify patterns or insights beyond the capacity of humans to detect unaided, particularly under time or resource constraints.

[0361] AI-agents may function as autonomous workflow controllers, executing corrective steps to return interrupted or failed operations back into a sequential or parallelized flow—without the need for human supervision or interaction.

[0362] In one or more embodiments, agents may be provisioned with neural networks or stochastic algorithms. They may use such models to solve high-complexity problems, including supply chain optimization (e.g., maximizing revenue-to-cost ratios) or fault diagnostics based on log data analysis.

[0363] When equipped with autonomous reasoning capabilities, AI-agents may predict future outcomes, adjust behavior dynamically, or identify long-term goals. This may involve supervised or unsupervised learning across business plans, engineering programs, or workforce strategies.

[0364] In one or more embodiments, an AI-agent may be alternatively classified into different types based at least in part on behavioral characteristics rather than their operational characteristics, for example. In one or more embodiments, AI-agents may be classified based at least in part on behavioral characteristics such as whether they are reactive, proactive or passive, whether they have a fixed or dynamically configured environment, or whether they are single, plural or multi-agent systems, for example.

[0365] Reactive AI-agents. In one or more embodiments, reactive AI-agents may respond to immediate telemetry stimuli received from sensors, a sensor managed by other AI-agents, or external systems multiplexing such telemetry, , for example, or take actions based on those telemetry stimuli, for example. In one or more embodiments, reactive AI-agents may be implemented, in whole or in part, in “memory-safe languages” as alluded to previously. In implementations, Reactive AI-agents may comprise assembler or c / c++ software, for example, written to access processing unit resources without interfering with clock-cycle timings, for example.

[0366] Proactive AI-agents. In one or more embodiments, proactive AI-agents may be empowered with algorithms, memory, or software instructions, for example, to create plans toachieve goals as defined in their axioms, for example.

[0367] A computing, network, or communications environment in which an AI-agent may operate may be fixed (e.g., as in “FPGA / ASIC / SoC, single core, multi-core) or may be dynamic, for example, such as in a case of reflexive software algorithms which may allow the AI-agent to interrogate itself for operational parameters which may or may not call for adjustment for advantageous (e.g., most optimal) operations, for example.

[0368] Fixed environments, as described above, may have axioms or interfaces that may not change, while dynamic environments (e.g., spread-spectrum intra-Agent communication, randomized one-time pads) may frequently or constantly change or may prompt AI-agents to adapt to new situations, for example.

[0369] In one or more embodiments, proactive AI-agents may be implemented, in whole or in part, in “memory-safe languages” as alluded to previously.

[0370] In implementations, Proactive AI-agents may comprise assembler or c / c++ software, for example, written to access processing unit resources without interfering with clock-cycle timings.

[0371] Multi-agent systems may involve multiple agents (e.g., hardware-based, software-based, or a hybrid of hardware-software) operating according to plans to achieve a common goal, for example.

[0372] In one or more embodiments, such AI-agents may coordinate actions or may communicate with each other to achieve objectives.

[0373] In one or more embodiments, AI-agents may be used in a variety of applications, including, by way of non-limiting example, robotics, gaming, or intelligent systems, for example.

[0374] Such systems may be implemented, in whole or in part, using different programming languages or techniques, including machine learning or natural language processing, for example, yet they may have their own grammars, syntax, or dictionaries, such as for terse, if compressed, communication streams.

[0375] To further elucidate, in one or more embodiments, an AI-agent may comprise: a hardware or software entity having private or shared memory-address locations; a device fabricated as a processing core / unit acting as the agent; or a software entity with one or more threads of control of a processing unit, for example.

[0376] In one or more embodiments, an AI-agent may comprise a hardware or software interface for external communication to other AI-agents or control systems, for example.

[0377] Also, in one or more embodiments, an AI-agent may have its own code library for reprogramming itself according to a plurality of dynamic planning algorithms to accomplish amyriad of goals which may or may not align with the goals of its nearest neighbors, for example.

[0378] According to other embodiments, an AI-agent Control Unit may modify an axiom repository by signaling to an associated AI-agent Runtime Adapter a demand for new axioms.

[0379] In one or more embodiments, a derivation of new axioms, which may be referred to as knowledge abstraction, may comprise a Small Language Model (SLM), Large Language Model (LLM), or an encrypted vectorized hash-table resident in the control unit of the AI-agent, for example.

[0380] In one or more embodiments, a former alternative describes a knowledge base of an AI-agent as an active, blackboard-like system, for example, whereas a latter alternative corresponds to a view of a classic Artificial Intelligence (AI) planning system (e.g., negotiation or voting using tokens).

[0381] In one or more embodiments, a performed alternative may depend at least in part on a power of inference services provided by the Language Model in repository, in one or more embodiments.

[0382] AI-agent Control Unit. In one or more embodiments, an AI-agent Control Unit may comprise a Godel-set of AI-agent Runtime Adapters fit for a specific purpose (e.g., Language Model ingestion, unstructured data ingestion, machine learning, deep learning, NLP, translation, text-to-speech synthesis), for example.

[0383] In one or more embodiments, an AI-agent Control Unit may comprise a computer software program, firmware, or semiconductor-based system that may be designed to perceive its Runtime Adapter according to a collection of pre-provisioned axioms or a continuous telemetry-measurement system to craft its own axioms using a grammar of its own.

[0384] In one or more embodiments, AI-agent Control Units may make autonomous decisions or may take self-directed actions to achieve a specific goal or set of goals by orchestrating the actions of one or a many (e.g., a swarm) of AI-agents, for example.

[0385] In one or more embodiments, within a context of an AI-agent Control Unit, AI-agents may operate autonomously, meaning AI-agents are not directly controlled by a human operator or external system, for example.

[0386] AI-agent Axiom Repository. In one or more embodiments, an AI-agent Axiom Repository may be encrypted in a Hierarchical Blockchain wallet in a storage location wherein axioms expressed as instructions for processing units to execute control, function, work-flow, or verification, for example, may be stored for rapid retrieval by AI-agents, for example.

[0387] In one or more embodiments, an Axiom repository may rely at least in part on the TMP AI-agent for verification.

[0388] In one or more embodiments, an AI-agent Axiom Repository may at once containsoftware instructions to algorithmically perform all relevant algorithms necessary to perform functions fit for a specific purpose, for example, not the least of which may comprise autonomic execution of software instructions formulated by the AI-agent itself, for example.

[0389] AI-agent Runtime Adapter. In one or more embodiments, an AI-agent Runtime Adapter includes, but is not limited to, a software-based processor emulator, a virtual-machine emulating an operating system, a processor core, cell or node which may, or may not, include a microkernel operating-system controlling access, a multi-core processor providing threads of control to a software or semiconductor-based AI-agent, for example.

[0390] In one or more embodiments, an AI-agent Runtime Adapter may reside in an FPGA / ASIC / SOC, a software Virtual Machine, a software virtual machine running on any one of a Processor core, FPGA, ASIC, SOC, or a memory-based processor emulator, for example.

[0391] In one or more embodiments, an AI-agent Runtime Adapter may operate in any permutations of MISD, SISD, MIMD, or SIMD-conforming architecture (e.g., whatever may be suggested by a particular use case), for example.

[0392] In one or more embodiments wherein CICAI AI-agent Runtime Adapters may lack hardware capable of MISD, SISD, SIMD, or MIMD, CICAI may be implemented as a SWAR (SIMD within a register) architecture involving general-purpose computer processors comprising one or more cores, cells, nodes, ., for example, which may lack SIMD micro-instructions in a hardware to perform SIMD operations.

[0393] Generalized Large Language Models. In one or more embodiments, generalized Language Models (e.g., generalized Large Language Models (gLLM)) may comprise artificial neural networks formulated for interrogation to achieve general-purpose language generation or other natural language processing tasks such as classification, for example.

[0394] In one or more embodiments, Language Models, such as generalized Language Models (e.g., gLLM), may acquire such abilities at least in part by learning statistical relationships from text documents, such as during a computationally intensive self-supervised or semi-supervised training process, for example.

[0395] In one or more embodiments, Language Models (e.g., LLMs, gLLMs) may be used for text generation, a form of generative AI, at least in part by taking an input text or repeatedly predicting a next token or word, for example.

[0396] In one or more embodiments, Personalized Large Language Models (pLLMs) may serve as individualized knowledge stores built upon artificial neural networks. These models may be tailored to reflect the unique characteristics, communication patterns, or task preferences of each user. They may also perform general NLP tasks, including but not limited to classification or intent recognition.

[0397] In one or more embodiments, pLLMs may learn from diverse data sources, including formatted, semi-formatted, and unformatted text. Examples include PDFs, spreadsheets, comma-separated values (CSV), JSON documents, voice transcripts, and content retrieved via RESTful API interrogation.

[0398] pLLMs may differ from generalized LLMs (gLLMs) by encoding specific user traits. These may include learned behaviors based on personal education, regional dialect, vowel pronunciation, logical reasoning style, idiomatic expressions, or tone and timbre. Such characteristics may enable highly contextual and individualized natural language understanding.

[0399] In one or more embodiments, the personalization process may involve computationally intensive self-supervised or semi-supervised training. These methods may help encode evolving context or communication patterns (e.g., voice recordings over time) into the model structure for continuity and refinement.

[0400] Generalized LLMs may perform generative text synthesis by predicting next tokens in a sequence based on a given input prompt. This behavior may rely on one or more vector-based computational methods, such as vector quantization (as pioneered by Dr. Teuvo Kohonen), TensorFlow-based tensor algebra, or Kronecker Dynamical System approaches, among other possible methods.

[0401] In one or more embodiments, a Personalized Language Model (pLLM) may be interrogated by a personal AI-agent (pAI-agent) that is uniquely crafted for that individual’s communication traits. This architecture allows a pLLM to act as an overlay on top of generalized Language Models, refining interpretation or translation tasks to align with personal attributes. Such attributes may include religious hermeneutics, dialectic semiotics, profession, first language, or educational background.

[0402] In one or more embodiments, a CICAI system may use a Language Model Abstraction Layer (LAL) to serve as a language interpreter or arbitrator. This LAL may adjust for word order or idiomatic structure and may translate responses into the preferred language or dialect of the interrogating agent, individual, or device.

[0403] A CICAI LAL may support simultaneous or concurrent interrogation of a plurality of Language Models operating in different languages. The system may return one or more responses in the language of the interrogator or in a format suited for further processing by a downstream agent or system.

[0404] In one or more embodiments, CICAI may be deployed in a Satellite-Air-Ground Integrated Network (SAGIN) architecture to provide real-time interpretation, transcription, or transliteration. This empowers senders and receivers to communicate transparently across linguistic and geographic boundaries.

[0405] In some embodiments, CICAI may translate a speaker’s analog voice into written text in the sender’s original language, including its native script. This may include writing systems that incorporate special diacritics or markers that are culturally or linguistically meaningful.

[0406] In one or more embodiments, the system may encode emotion, tone, and speaker intent into the translated output. For example, vowel intonation may indicate happiness, stoicism, or disappointment, depending on the cultural norms of the speaker and recipient. These vocal nuances may be preserved through appropriate diacritic marks or transliteration features.

[0407] In one or more embodiments, diacritics may be applied to computational linguistics tasks such as syntactic parsing, semantic interpretation, sentiment analysis, or cross-language alignment. These markers may improve the accuracy or nuance of interpretation, transcription, or translation operations.

[0408] Diacritics may serve to encode aspects of human emotion, tone, timbre, or speaker intent. When embedded into transcriptions or written versions of analog voice streams, they may facilitate reproduction of affective elements otherwise lost in typical language model translation workflows.

[0409] In one or more embodiments, this may be especially valuable for abjad-type writing systems, such as those used in North African Amazigh Tifinagh, Gulf Arabic, Farsi, or Hebrew. These languages often rely on optional or culturally significant diacritic marks for phonetic or emotional nuance.

[0410] If transcription occurs in real time or near-real time, textual exegesis may be performed to evaluate whether the speaker’s dialogue patterns conform to social or historical norms. This may include linguistic consistency checks across cultural contexts or historical baselines.

[0411] In some embodiments, this may serve military or law enforcement applications, particularly where speakers use coded language. Standard analog-to-digital transcription may fail to detect such patterns, but diacritic-enhanced transcription may allow detection of anomalous communication.

[0412] When coded or anomalous speech is detected, additional AI-agents may be autonomously provisioned to perform more advanced analytics. These agents may apply targeted algorithms to analyze bilateral conversations or public presentations in greater depth.

[0413] Personal profile. In one or more embodiments, a personal profile may comprise content representative of personal characteristics of an individual. Such characteristics includes, for example, regional, or social variety of a language distinguished, for example, by pronunciation, grammar, or vocabulary, for example, especially, in one or more embodiments, avariety of speech differing from a standard literary language or speech pattern of a culture in which it exists, taking into account uniqueness of particular individuals, ethnicity, languages, or cultures, for example.

[0414] One or more embodiments herein may describe networking hardware or networked software implemented within or adjacent to a telecom network or network device to provide real-time, or near real-time, interpretation, or translation of communications to normalize or equivocate communications according to both sender(s) or receiver(s), for example.

[0415] One or more embodiments described herein may be directed to “leveling the playing field” for the largest or smallest economies on Earth by removing significant economic or educational barriers, real or unreal, to commerce.

[0416] Developing economies, or governments with lower national literacy rates or higher unemployment rates, for example, may be benefactors, perhaps more so than developed countries because one or more embodiments described herein provide approaches whereby capital may independently find labor of merit, globally, simply by picking up the phone or dialing anyone in the world to discuss exchange of products or services in exchange for value, for example.

[0417] As previously alluded to, embodiments described herein may involve telecommunications networks, such as digital or analog networks, for example, with one or more data processing units that may be communicatively connected or coupled to one or more computing devices or platforms so as to provide computational source-to-target language conversion services in real time or near-real time.

[0418] As described in greater detail herein (e.g., above and also below), these or like services includes, for example, interpretation, translation, transcription, transliteration, or like aspects or processes, such as implemented, at least in part, in connection with one, or a plurality of, sender(s) and one, or a plurality of, receiver(s) engaged in unilateral, bilateral, directional, or omnidirectional network-type communication, for example.

[0419] In one or more embodiments, AI-agents may exhibit a set of shared assumptions, operational principles, or architectural characteristics.

[0420] In one or more embodiments, an AI-agent may assume a cyber-secure environment is already in place or may be serialized into such an environment by a trusted AI-agent Runtime Adapter.

[0421] In one or more embodiments, AI-agents may utilize shared services such as logging utilities, a security manager, device adaptors, or class loaders, commonly accessible via the AI-agent Runtime Adapter.

[0422] In one or more embodiments, AI-agents may operate within a limited, stochastically defined runtime hierarchy, governed by control axioms, policy layers, messaging protocols, or topology-mapping services, to coordinate problem-solving.

[0423] In one or more embodiments, AI-agents may assume that all computing resources (e.g., processors, memory, storage, peripherals) are available unless restricted. Permissions may be encoded in axioms stored in encrypted hierarchical blockchain wallets and enforced by the cyber-security manager.

[0424] In one or more embodiments, each AI-agent may have a unique identity assigned via a Universally Unique Identifier (UUID) and may be uniquely named based on that identifier and its Runtime Adapter context.

[0425] In one or more embodiments, AI-agents may manage their operational state, checkpoint progress, and serialize to other locations for continued execution without resource exhaustion or interruption.

[0426] In one or more embodiments, AI-agents may belong to a defined management domain such as: (a) Controlling: acts as an Axiom Decision Point (ADP), (b) Subordinate: defers control to another agent, or (c) Not Applicable (NA): ignores axiomatic requests.

[0427] In one or more embodiments, each AI-agent may be assigned a thread group at instantiation, such as POSIX or non-POSIX threads operating in either kernel or user memory space, subject to operating system-level controls.

[0428] In one or more embodiments, AI-agents may exchange messages using shared variables, Remote-Process Communication (RPC), Inter-Process Communication (IPC), or shared queues or blackboard mechanisms.

[0429] In one or more embodiments, AI-agent thread priority may be determined dynamically based on instantiation axioms or class type, enabling system-wide resource prioritization.

[0430] In one or more embodiments, upon instantiation, an AI-agent may register its UUID with the local Runtime Adapter to enable identification, auditing, or controlled message routing.

[0431] In one or more embodiments, AI-agents may be passivated or reactivated under control of kernel-level thread managers, allowing temporary suspension and later resumption of agent processes as needed.

[0432] As discussed above, individual AI-agent Runtime Adapters may have an ability to provide telemetry on the per-thread utilization of individual threads running within it. Such content may be used by an AI-agent control unit to, for example, help determine load balancing on AI-agent Runtime Adapters.

[0433] One or more embodiments includes systolic-arrays or wavefront-arrays of AI-gents performing parallel or sequential execution of algorithmic computer instructions (e.g., instructions which are either embedded in hardware-based AI-agents or intrinsic to software instructions embedded in AI-agents). In one or more embodiments, parallelized AI-agents may be configured (e.g., optimized) for, by way of non-limiting examples, image, or video processing, speech recognition, data compression, Convolutional Neural Networks, Recurrent Neural Networks, Deep Belief Networks, Symmetric Key Encryption, Hash Functions, computerized-vision Object detection or recognition, Facial Recognition, or Video analytics. Of course, subject matter is not limited in scope in these respects.

[0434] In one or more embodiments, training data sets, such as for Large Language Models (LLM), may be gathered from any of a wide range of sources. In some embodiments, LLMs may be developed by a vendor (e.g., CICAI vendor) for subscribers to their platform. For example, a cluster of construction companies may subscribe to a CICAI cloud platform. Permission may be sought from these companies to utilize their public or non-proprietary information to help train a Language Model (e.g., LLM) for this example cluster of companies. Clustering entities in this manner may make sense in some circumstances because their lexicon may reuse the same nouns or verbs from other disciplines but may have very different meanings in their particular context (e.g., construction).

[0435] In one or more embodiments, Language Models, for example, may be built based on data gleaned from other sources. For example, it may be desirable or advantageous to interrogate other AI platforms such as Google® Gemini®, Microsoft® Copilot®, . A Language Model abstraction layer may be constructed to translate content requests from the other sources into a format that would be proper for the particular target platform. In one or more embodiments, content obtained from one or more sources may be transformed into the phrasing, terminology, vernacular, . that is appropriate for a specified purpose. Generally stated, training data may be procured from other vendors, in one or more embodiments.

[0436] Example Use Case

[0437] In one or more embodiments, a real-time or near real-time interpreter or translator may preserve the emotional sentiment of a communication, such that what is felt or intended by one party is conveyed with substantial equivalence to the receiving party, even when different words are used than might appear in a literal word-for-word translation. This allows the underlying intent or emotional impact to be retained across languages.

[0438] In one or more embodiments, a CICAI-based system may apply multiple vocabularies or interpretation techniques concurrently. For example, two translation paths may run in parallel to cross-validate or derive an approximately equal sentiment score for outboundand inbound speech segments.

[0439] In one or more embodiments, interpretation or translation may occur in real time or near real-time for both sides of a conversation. This ensures that regardless of the language used by either participant, the interpretation process proceeds concurrently, without noticeable delay.

[0440] In one or more embodiments, the conversation or presentation may include more than two participants. The system may interpret or translate communications in real time or near real-time for each individual participant, allowing multilingual group discussions to proceed naturally and fluidly.

[0441] Although some aspects of the discussion below may mention interpretation or translation operations or processes, subject matter is not limited in scope in these respects.

[0442] For example, as alluded to previously, any of interpretation, transcription, translation, transliteration or like aspects, operations, or processes may be performed in conjunction with any or all of interpretation, transcription, translation, transliteration or like aspects, operations, or processes, in any combination or order, in one or more embodiments.

[0443] Thus, in one or more embodiments, one or more of interpretation, transcription, translation, or transliteration approaches, operations, processes, ., may occur (e.g., in real time or near real-time), for individual participants in a conversation, meeting, presentation, teleconference, lecture, .

[0444] As alluded to previously, in one or more embodiments, interpretation, transcription, translation, or transliteration approaches, operations, processes, ., may be implemented, in whole or in part, via AI-agents.

[0445] In context of human communication (e.g., voice, text messages, email), AI-agents may operate at least partially autonomously, or may at times be referred to as autonomous AI-agents, for example.

[0446] In this context, “autonomous” in connection with an AI-agent refers to a capability to recognize specific tasks to be performed, such as in connection with interpretation, transcription, translation, or transliteration approaches, operations, processes, ., for example, and to perform such without being specifically instructed, such as via feedback or other input, by a human individual.

[0447] Also, in one or more embodiments, an autonomous AI-agent may self-driven, meaning, for example, it has a workflow engine to which it may communicate, or a workflow engine may apprise an AI-agent of its scheduled duties, for example.

[0448] In one or more embodiments, mechanisms for which such example duties are to be performed or roundtrip completion of that unit of work as a state may be monitored within asystem itself, for example.

[0449] This may be in contrast to approaches where a user may provide input, such as via a button push or via a touch-screen icon, for example, to instruct a system to perform particular operations.

[0450] In one or more embodiments, an example AI product suite may comprise a multi-modal AI approach utilizing leading-edge developments in computational linguistics. These techniques may include, but are not limited to, syntactic parsing, semantic interpretation, sentiment analysis, and cross-language alignment in applications involving interpretation, translation, or other language processing.

[0451] One or more embodiments may make use of Personalized Large Language Models (pLLMs), general-purpose Large Language Models (LLMs), Small Language Models (SLMs), or Multilingual Language Models (MLMs) to tailor natural language tasks to individual users, specific use cases, or domain constraints.

[0452] In one or more embodiments, AI Agents may interrogate such language models to perform interpretation or translation of human bilateral communication in real time or near real-time, for example, within a latency window of less than 200 milliseconds per direction.

[0453] Taking this further, one or more embodiments may perform parallelized operations to carry out multi-way, simultaneous interpretation or translation in real time or near real-time, regardless of the number of participants. Examples include multi-speaker events such as professional conventions or users consuming stored multimedia content on services such as Amazon Prime, Netflix, or YouTube.

[0454] FIG. 17 is a block diagram depicting an example environment including an AI-Agent Runtime Adapter, in accordance with one or more embodiments. In one or more embodiments, the AI-Agent Runtime Adapter may provide a common abstraction layer or interface to enable CICAI functionality. This functionality may include interpretation, transcription, translation, or transliteration operations performed in parallel or concurrently by multiple AI-agents across a variety of platforms.

[0455] In one or more embodiments, the AI-Agent Runtime Adapter may be executed on a wide range of underlying hardware platforms. These may include, by way of non-limiting example, general-purpose central processing units (CPUs), multi-core processors, or specialized processor types conforming to MIMD, SIMD, MISD, or SISD architectures.

[0456] In one or more embodiments, the AI-Agent Runtime Adapter and its associated operations, such as those depicted in FIG.17, may be implemented in hardware, software (other than software per se), or any combination thereof.

[0457] In one or more embodiments, one or more processors may comprise off-the-shelftype processors while other embodiments may utilize, in whole or in part, custom implementations designed with parallelism in mind. One or more embodiments may also utilize, in whole or in part, off-the-shelf language processing units (LPU) (e.g., Groq® LPU Inference Engine), for example. In other embodiments, modifications, or customizations may be made to third-party processors, such as NPUs, CPU, GPU, LPUs, ., to more closely align with concurrent / parallel aspects of AI-agent interpretation, transcription, translation, or transliteration operations described herein, for example.

[0458] As also depicted in FIG.17 is an example Software / Hardware Abstraction layer, in accordance with one or more embodiments. In one or more embodiments, processor vendors or manufacturers, for example, may produce code libraries, API libraries, ., that may enable programmers to make calls to specified software libraries or to access particular aspects of underlying hardware in a CPU, GPU, NPU, LPU, ., for example, in one or more embodiments. In one or more embodiments, a Software / Hardware Abstraction layer may provide an interface between Runtime Adapter and underlying hardware (e.g., one or more general purpose CPUs, one or more MIMD, SIMD, MISD, or SISD processors), for example.

[0459] Further, as depicted in FIG. 17, one or more embodiments includes a Virtual Machine / Microkernel layer. In one or more embodiments, a Virtual Machine / Microkernel layer may comprise one or more virtual machines, wherein specified virtual machines may be selected to perform specified tasks, for example. Also, in one or more embodiments, a Virtual Machine / Microkernel layer may comprise a microkernel agent manager (e.g., referred to as “HuroBoss”) that may control deployment or provisioning of AI-agents or processes (e.g., daemons) to support AI-agent operation, for example. In some embodiments, a Virtual Machine / Microkernel layer may be optional. In FIG.17, an optional component is indicated by way of a dashed line. For embodiments that forego virtual machines, functionality of a HuroBoss agent manager may be performed at one or more other layers, such as, for example, a Runtime Adapter layer or Software / Hardware Abstraction layer.

[0460] In one or more embodiments, an AI Algorithm Library may operate in conjunction with a Virtual Machine / Microkernel layer, for example. In one or more embodiments, an AI Algorithm Library includes, by way of non-limiting examples, semiotic interpretation, hermeneutic evaluation, sentiment scoring, emotional index scoring, or axiomatic, rule-based interpretation. Further, for example, AI Algorithm Library may comprise a code library accessible to autonomous AI-agents.

[0461] In one or more embodiments, microkernel agent manager HuroBoss may sample text, voice, or other means of communication that may flow through whichever electronic device is hosting components of FIG.17. In one or more embodiments, digital audio data, for example,may be observed by HuroBoss agent in the clear (e.g., unencrypted), for example. Also, in one or more embodiments, a HuroBoss agent manager may perform thread management tasks, virtual machine scheduling, AI-agent deployment and scheduling, or may, in general, act as a supervisory agent, for example. In one or more embodiments, HuroBoss agent manager may control a number of AI-agents, or may detect problems with any subordinate agents, for example. HuroBoss agent manager may also have an ability to restart or reprovision subordinate agents, for example. A binary tree may also be kept by HuroBoss agent manager to track availability of virtual machines or to load balance, in one or more embodiments.

[0462] In one or more embodiments, where Microkernels may be used in conjunction with AI-agents, for example, an example Microkernel configuration may comprise a Kernel-space operation to keep an inventory of software components (e.g., all software components) which may be in operation, and those which should be in operation, for example. As previously alluded to, for circumstances in which there may be a deviation between multiple (e.g., two) lists, for example, a microkernel boot-loader AI-agent (e.g., referred to as “HuroBOSS”), for example, may schedule kernel or process threads to start-up, restart, or terminate, or maintains inventory of AI-agent provisioning for performance of action, or some management information base (MIB) content, for example. In one or more embodiments, a microkernel boot-loader may be responsible for initializing system hardware or loading a microkernel into memory. In some cases, a boot-loader might also have capability to detect failures or attempt to restart a microkernel, for example.

[0463] In one or more embodiments, during Microkernel boot-up, an init process may comprise a special process that may be launched by a microkernel during boot-up. In one or more embodiments, init may be responsible for initializing a system by starting user-space services (e.g., essential user-space services) or operating system kernel daemons (e.g., processes). In one or more embodiments, responsive at least in part to an init process detecting an issue with a microkernel, init process may not perform its tasks or a system may halt, for example.

[0464] In one or more embodiments, microkernel architectures may rely on redundancy mechanisms for system components (e.g., critical system components). For example, some systems includes multiple boot-loaders or redundant init processes to improve fault tolerance. Additionally, in implementations, external monitoring tools may be used to observe health of a system, including a microkernel, for example. In one or more embodiments, a HuroBOSS Agent, for example, may provide monitoring, restart, or thread supervisory control.

[0465] Referring again to FIG.17, one or more embodiments may include a monolithic POSIX-compliant facsimile of a Unix Operating System (e.g., Multics or Linux) that acts as anAI-agent Runtime Adapter or virtual machine host.

[0466] In one or more embodiments, an encrypted file containing Language Model Math Modules (e.g., specialized libraries) may be called by AI-agents. These modules may assist in ensuring that processing of sine or cosine vectors can be executed in a consistent fashion across general-purpose CPUs, even where GPUs or LPUs are unavailable.

[0467] Although CPUs may be slower than GPUs or LPUs for certain operations, the use of dedicated AI-specific math libraries may preserve output quality across platforms. In one or more embodiments, this consistency may be preferred over reliance on hardware-vendor-specific (OEM) libraries.

[0468] In one or more embodiments, Language Model Math Modules may include libraries for vector mathematics, enforcing the use of 2×2 or 4×4 matrices to standardize processing structures across devices.

[0469] While POSIX compliance is described in the foregoing embodiments, one or more embodiments may operate on non-POSIX operating systems as well. These may include multiplexed single-thread operating systems or other current or future OS types not limited in scope by POSIX constraints.

[0470] Also depicted in FIG. 17, one or more embodiments may also include a machine-safe programming language interpreter (e.g., Python, Java Virtual Macine, Rust).

[0471] FIG.17 also illustrates, in one or more embodiments, a plurality of AI-agents (autonomous or otherwise). In one or more embodiments, kernel, or process threads of AI-agents may be controlled by a HuroBoss agent controller, for example. It may be noted that example embodiments of AI-agents of a range of different types are discussed above.

[0472] FIG. 18 is a block diagram depicting an example AI-agent control unit, in accordance with an embodiment. In one or more embodiments, example approaches, operations, processes, instructions, ., such as depicted or discussed in connection with FIG.17, for example, may be implemented, in whole or in part, in hardware or software (other than software per se) or a combination thereof.

[0473] For an example depicted in FIG.18, a number of AI-agents are shown (e.g., AI agent with Runtime Adapter) as clear boxes. It may be noted that AI-agents labeled 1-4 may be connected to a library of code and algorithms (e.g. AI algorithm library depicted in FIG.17, discussed above) and also connected to a shared memory (depicted in FIG. 18 as a lightly shaded box). AI-agents A through D, for this example, may be connected between shared memory and a private memory (depicted in FIG. 18 in a medium-shaded box). Further, for example, AI-agents labeled X through Z may be connected to private memory and may also be connected to a library of code and algorithms.

[0474] In one or more embodiments, during the performance of workflows, it may be advantageous for multiple AI-agents (e.g., two agents in a telephony session) to share a memory space. For instance, two concurrent communication streams may pass through a shared memory location for parallel processing or coordination.

[0475] In other embodiments, some AI-agents may be provisioned by other upstream AI-agents. In this arrangement, a later-stage agent in a given workflow may be instantiated by an earlier-stage agent to complete a delegated portion of the workflow. This architecture allows for dynamic chaining or composition of AI-agent behavior based on system demands.

[0476] In one or more embodiments, the organization of AI-agents—including their access to libraries of code or algorithms and their connection to shared or private memory—may be influenced by the action potential or autonomy level of each agent. For example, agents with full access to algorithm libraries may act autonomously, while agents without such access may rely on upstream agents for processing support.

[0477] Additionally, in one or more embodiments, shared memory may store binary tree structures used for search and locate functions. Private memory, by contrast, may be allocated as scratch memory for computational tasks such as 2×2 or 4×4 matrix operations.

[0478] In one or more embodiments, masked language models (MLMs) may differ from LLMs, SLMs and Personal LLMs (pLLMs), for example. MLMs may comprised multiple data sets (e.g., three different data sets) that may be stored in an efficient binary tree structure with a search index, for example. One or more embodiments may comprise a multi-stage (e.g., three-stage) approach to translation or interpretation including, for example, 1) native speech phonemes to ascii text, then 2) convert ascii words to equivalents in new language, or 3) convert a new language equivalent text to audible speech, for example. Such an approach may be referred to as “STT-TTT-TTS” (Speech-to-text-, Text-to-text, Text-to-speech, respectively) , for example. Of course, subject matter is not limited in scope in these respects.

[0479] In the following, example speech-to-text, text-to-text, and text-to-speech approaches of an example embodiment for multi-stage interpretation or translation may be discussed.

[0480] Speech-to-Text

[0481] In one or more embodiments, an AI-agent responsible for managing a speech-to-text workflow may utilize its Runtime Adapter to pipe a real-time text stream into a file. During this process, the AI-agent may insert itself mid-stream, prior to file system storage, to modify the data stream.

[0482] In one or more embodiments, the inserted AI-agent may apply diacritic insertion either in the native textual representation or as part of a weighted text-to-text languageconversion process. For example, converting text from Arabic to Japanese may include diacritic mapping to preserve emotional and contextual equivalency between source and target languages, based on a pre-specified accuracy threshold.

[0483] This diacritic mapping may enable the execution of quantitative tests to assess whether contextual or emotional distortion has occurred during language conversion. In one or more embodiments, statistical comparison using first and second standard deviations from a phonetic or tonal mean may be used to validate fidelity of the translation before moving forward to interpretation or hermeneutic evaluation.

[0484] Additionally, in one or more embodiments, voice samples—either recorded or synthesized—may incorporate sampling of octaves, frequency ranges, or verbal emphasis patterns of the original speaker. This voice data, aligned with diacritic markers, may be used to “humanize” a synthetic voice in subsequent text-to-speech stages, achieving a facsimile that matches the original speaker’s intonation and emotional signature.

[0485] In one or more embodiments, converting microphone input (e.g., human speech) into text may be referred to as “automatic speech recognition” (ASR) or speech-to-text (STT) , for example. A conversion of analog speech (e.g., analog electrical audio signals) to computer-usable signals may be implemented or initiated (e.g., via execution of computer instructions) at a few points in process of speech to text conversion, in one or more embodiments. For example, a programmer may intercept signal on a microphone or may process a signal using any of a range of algorithms at one or more of following stages (provided herein as non-limiting examples):

[0486] Analog-to-Digital Conversion (ADC). In one or more embodiments, an ADC stage may be where an analog signal from a microphone, for example, may be converted into a digital signal. At this stage, a programmer may access raw analog signal and process it using their own algorithms, for example. This approach includes direct access to ADC chip or a compatible interface, such as a USB or SPI interface, for example. In one or more embodiments, a possible approach includes a USB3.x / 4.x “bump on the wire” to intercept UDP signals encapsulated in RTP / sRTP or to parallelize translation services – an entire MLM on a USB3.x / 4.x device for VoIP integration on-the-go, for example. In one or more embodiments, an overview of an example ADC approach includes the following.

[0487] Sampling. In implementations, an initial stage in an ADC approach may comprise sampling. In one or more embodiments, sampling involves capturing a small portion of an analog signal at regular intervals, typically at a rate that may be faster than a highest frequency component of a signal. This may be done to ensure that a digital signal accurately represents an analog signal, for example.

[0488] Quantization. In one or more embodiments, after sampling, a next stage may comprise quantization. In one or more embodiments, quantization involves converting a sampled analog signal into a digital signal by assigning a digital value to each sample, for example. This may be done by dividing an analog signal into a finite number of levels, typically 2^n, where n may be a number of bits used for quantization, for example.

[0489] Encoding. In one or more embodiments, a quantized digital signal may then be encoded using a specified encoding scheme, such as pulse-code modulation (PCM) or delta-sigma modulation (DSM), for example. an encoding scheme determines a format of a digital signal, including a number of bits per sample, sampling rate, and encoding algorithm used, for example.

[0490] Digital Signal Processing. In one or more embodiments, an encoded digital signal may then be processed using digital signal processing (DSP) techniques, such as filtering, amplification, or compression, for example. These techniques may be used to enhance signal quality, remove noise, compress signal to reduce its size, . In one or more embodiments, DSP may be performed by a DSP processor, CPU, GPU, ., although subject matter is not limited in scope in this respect.

[0491] Output. An additional stage to an example overall ADC approach includes outputting a digital signal. In one or more embodiments, a digital signal may be transmitted to a digital system, such as a computer (e.g., digital audio workstation) where it may be processed further or stored for later use, for example.

[0492] In accordance with one or more example embodiments discussed above, an example ADC approach includes the following. Sampling: Capture a small portion of an analog signal at regular intervals. Quantization: Convert sampled analog signal into a digital signal by assigning a digital value to each sample. Encoding: Encode quantized digital signal using a specified encoding scheme. Digital Signal Processing: Process encoded digital signal using DSP techniques. Output: Output digital signal to a digital system, for example.

[0493] Example types of ADC includes, but are not limited to, the following. Flash ADC: Uses a relatively large number of comparators to convert an analog signal into a digital signal. Successive Approximation ADC: Uses a series of approximations to convert analog signal into a digital signal. Delta-Sigma ADC: Uses a delta-sigma modulation scheme to convert analog signal into a digital signal. Sigma-Delta ADC: Uses a sigma-delta modulation scheme to convert analog signal into a digital signal.

[0494] Digital Signal Processing (DSP). As mentioned, a programmer may intercept an audio signal or may process a signal using any of a range of algorithms or approaches at any of a number of stages. Another such example stage includes Digital Signal Processing (DSP). In oneor more embodiments, a DSP stage may be where a digital signal may be processed, such as using algorithms provided by a microphone or computer manufacturers, for example. However, some microphones or audio interfaces may provide access to raw digital signal, thereby allowing programmers to inject their own algorithms. This may be done through a software development kit (SDK) or an application programming interface (API), for example. As mentioned, DSP techniques may be performed utilizing any of a wide range of processor types including, by way of non-limiting examples, CPU, GPU, DSP processor, . In one or more embodiments, such processors may comprise multi-core, multi-threaded processors.

[0495] Audio Interface. Another example stage where a programmer may intercept or process an audio signal includes an Audio Interface. An audio interface stage may be where a digital signal may be transmitted to computer's sound card (or other hardware circuitry) or audio processing software, for example. In one or more embodiments, at this stage, programmers may access a digital signal and process it using their own algorithms or other algorithms. Such an approach may involve access to audio interface's API or a compatible interface, such as a USB or interface, for example. In one or more embodiments, an interface may vary depending on a particular audio interface, a computer's sound card, audio processing software, . Of course, subject matter is not limited in scope in these respects.

[0496] In one or more embodiments, non-limiting examples of audio interfaces includes the following:

[0497] USB Interface. Many audio interfaces or devices may use a USB (Universal Serial Bus) interface to connect to computer, for example. USB interface may comprise a USB connector on an audio interface and a USB port on a computer, for example. An audio interface may send digital signal to computer via USB connection, for example.

[0498] MIDI Interface. Some audio interfaces or devices may use a MIDI (Musical Instrument Digital Interface) interface to connect to computer, for example. MIDI interface may comprise a MIDI connector on audio interface or device and a MIDI port on computer, for example. Although MIDI interfaces generally have not supported transportation of audio (e.g., MIDI implementations may encode representations of musical notes or other commands), other implementations or specifications may so do. In such cases, audio interface may send digital signal to computer via MIDI connection, for example.

[0499] FireWire Interface. Some audio interfaces use a FireWire interface to connect to computer, for example. FireWire interface may comprise a FireWire connector on audio interface and a FireWire port on computer, for example. An audio interface sends digital signal to computer via FireWire connection, for example.

[0500] Thunderbolt Interface. Some audio interfaces may use a Thunderbolt interface toconnect to computer, for example. Thunderbolt interface may comprise a Thunderbolt connector on audio interface and a Thunderbolt port on computer, for example. An audio interface sends digital signal to computer via Thunderbolt connection, for example. In some embodiments, Thunderbolt protocol may be supported via USB connectors (e.g., USB-C), for example.

[0501] Ethernet Interface (OSI Layer 2 interface - switchable). Further, some audio interfaces may use an Ethernet interface to connect to computer, for example. Ethernet interface may comprise an Ethernet connector on audio interface and an Ethernet port on computer, for example. An audio interface sends digital signal to computer via Ethernet connection, for example. In one or more embodiments, Ethernet interface over Layer 2 may be switchable in a Trill Switch as may be found in CICAI, for example.

[0502] Audio Jacks. Some audio interfaces use audio jacks to connect to computer, for example. Audio jacks may comprise a set of audio connectors on audio interface and a set of audio connectors on computer, for example. Audio interface sends digital signal to computer via audio jacks, for example.

[0503] Optical Interface. Some audio interfaces may use an optical interface to connect to computer, for example. An optical interface may comprise an optical connector on audio interface and an optical port on computer, for example. An audio interface sends digital signal to computer via optical connection, for example.

[0504] In one or more embodiments, an audio signal may be intercepted or processed at a sound card or audio processing software stage. Certain systems may expose access to raw digital audio streams via a software development kit (SDK) or application programming interface (API), enabling a programmer to inject custom signal processing algorithms.

[0505] In one or more embodiments, the digital signal may also be manipulated using software-based computer instructions provided by the manufacturer of the chip assembly installed in computing platforms such as laptops, smartphones, or tablets. These manufacturer instructions may allow native-level audio preprocessing at the chip level.

[0506] In one or more embodiments, audio streams processed by multi-core or DSP-type processors may incorporate diacritics, emotional tags, stress tags, parts-of-speech tags, or other metadata. This enriched signal data may be passed to machine learning or deep learning modules to uncover behavioral or cognitive patterns reflected in speech.

[0507] In one or more embodiments, to intercept an audio signal and process audio signal, programmers may: Obtain hardware or software tools, such as a compatible microphone, audio interface, or programming language, for example; Understand signal processing pipeline or formats used by microphone or computer manufacturers, for example; Develop custom algorithms or software to process a signal, which may involve expertise in signal processing,programming, or audio engineering, for example; Integrate their custom algorithms with existing signal processing pipeline, which may involve modifying audio interface, sound card, or audio processing software, for example.

[0508] In one or more embodiments, some examples of programming languages and tools that may be used in processing of audio signals includes, for example: Python with libraries like PyAudio, Pydub, or Librosa, for example; C++ with libraries like PortAudio, OpenAL, or FFmpeg, for example; Java with libraries like Java Sound API or JAudio; MATLAB with built-in audio processing tools, for example.

[0509] One or more embodiments are discussed below for example approaches for processing speech into text (e.g., ascii). Of course, subject matter is not limited in scope in these respects

[0510] Audio Signal Acquisition: A microphone captures audio signal of human speech and sends it to speech recognition system, for example. An explanation of how a microphone may acquire an audio signal of human speech and may send it to a speech recognition system for conversion into text is provided below. Audio Signal Acquisition in a computer-attached microphone includes, for example:

[0511] 1) Sound Wave Capture. A microphone captures sound waves in air and converts them into an analog electrical signal, for example. This signal may be a representation of a sound wave's amplitude (loudness) and frequency (pitch), for example.

[0512] In one or more embodiments, an analog electrical signal from a microphone may be sent to a pre-amplifier, which boosts the signal to a level suitable for processing by a computer’s analog-to-digital converter (ADC). The pre-amplifier stage may comprise multiple signal-processing subfunctions to prepare the signal for digital conversion.

[0513] In one or more embodiments, the following example subfunctions may be performed by the pre-amplifier:

[0514] a) Signal Conditioning: Adjusting amplitude, frequency response, and impedance to match the ADC’s requirements.

[0515] b) Amplification: Boosting signal strength using op-amps or dedicated amplifier chips.

[0516] c) Filtering: Removing unwanted noise, hum, or interference via passive (resistors, capacitors) or active components.

[0517] d) Impedance Matching: Adjusting signal impedance to align with ADC input,preventing distortion.

[0518] e) Level Shifting: Scaling amplitude to match ADC specifications.

[0519] f) Output: Transmitting the processed signal to the ADC.

[0520] In one or more embodiments, after conditioning, amplification, filtering, impedance matching, and level shifting, the pre-amplifier transmits the refined signal to an ADC, where it is converted into a digital format for processing in NLP or CICAI workflows.

[0521] In one or more embodiments, specific design considerations for a microphone pre-amplifier optimized for NLP and parallelization environments may include: a) Gain: The amount of amplification applied to the signal, sufficient to bring the signal to the specified level for the ADC.

[0522] b) Frequency Response: The range of frequencies that the pre-amplifier is designed to amplify, matched to the microphone and ADC.

[0523] c) Noise Floor: The level of noise present in the output; preferably minimized to avoid distortion.

[0524] d) Impedance Matching: Proper alignment of input / output impedances to ensure signal integrity and prevent loss.

[0525] e) Level Shifting: Amplitude adjustments to ensure compatibility with ADC requirements.

[0526] 3) Analog-to-Digital Conversion (ADC). Audio Signal Acquisition in a computer-attached microphone may further include ADC, for example. In one or more embodiments, a pre-amplified signal may be sent to ADC which converts analog signal into a digital signal, for example. ADC samples signal at a specific rate, typically measured in Hertz (Hz), or assigns a digital value to each sample, for example.

[0527] 4) Digital Signal Processing (DSP). Audio Signal Acquisition may also include DSP, for example. In one or more embodiments, a digital signal may be processed by a computer's digital signal processing (DSP) unit, for example. DSP unit performs various tasks, such as: Filtering, which may comprise removing noise or unwanted frequencies from a signal, for example; Amplification, which may comprise adjusting signal's amplitude to a suitable level, for example; and Equalization, which includes adjusting signal's frequency response to compensate for microphone characteristics, for example.

[0528] 5) Audio Interface. Audio Signal Acquisition may also include processed digital signal being sent to computer's audio interface, which may be responsible for transmitting signal to computer's sound card or audio processing software, in one or more embodiments.

[0529] 6) Sound Card or Audio Processing Software. Audio interface sends digital signal to computer's sound card or audio processing software, for example. In one or more embodiments, sound card or software converts digital signal into a format that may be processed by computer's operating system and applications, for example.

[0530] 7) Operating System and Applications. Processed audio signal may be sent to computer's operating system or applications, such as audio editing software, voice assistants, video conferencing software, . These applications may use audio signal for any of a variety of purposes, including, for example, recording, playback, analysis, . In one or more embodiments, an application may comprise a Software Agent which may utilize digital audio signal for various purposes, such as, for example, recording, playback, or analysis, to name but a few non-limiting examples.

[0531] 8) Storage or Transmission. An audio signal may be stored on computer's hard drive or transmitted over a network to a remote location for further processing or storage, in one or more embodiments.

[0532] In summary, an example approach to Audio Signal Acquisition with a computer-attached microphone, for example, may involve: 1) Sound wave capture by a microphone; 2) Pre-amplification to boost signal; 3) Analog-to-digital conversion to convert signal to digital; 4) Digital signal processing to filter, amplify, and equalize signal. 5) Transmission to computer's audio interface; 6) Processing by sound card or audio processing software. 7) Use by operating system and applications.8) Storage or transmission for further processing or storage. Of course, subject matter is not limited in scope in these respects.

[0533] Sound Waves. One or more embodiments for processing speech into text (e.g., ascii) may further include capturing or processing sound waves, for example. For example, when a person speaks, their voice produces sound waves that travel through air as a pressure wave, for example. These sound waves have different frequencies, amplitudes, or durations, which may correspond to different phonemes (e.g., units of sound) in a spoken language, for example. In one or more embodiments, a process of capturing sound waves, clipping, and converting them into phonemes may comprise a multi-stage approach, for example. For example, an example approach includes:

[0534] 1) Sound Wave Capture. In one or more embodiments, a microphone converts physical vibrations of sound waves in air into electrical signals, for example. This may be implemented, in whole or in part, through a process called electromagnetic induction, forexample. A microphone may comprise a thin diaphragm attached to a coil of wire, for example. When sound waves reach microphone, they cause a diaphragm to vibrate, for example. These vibrations induce an electrical current in a coil, which may be proportional to sound wave's amplitude (loudness), for example. Of courses, this is merely one example of microphone technology, and subject matter is not limited in scope in these respects.

[0535] 2) Analog-to-Digital Conversion (ADC). An electrical signal from a microphone may be sent to an analog-to- digital converter (ADC), in one or more embodiments. ADC converts a continuous analog signal into a digital signal, which may comprise a series of discrete values represented by binary code (0s and 1s), for example. This process may be called sampling, for example. Sampling rate determines how often ADC takes a snapshot of a signal, typically measured in Hertz (Hz), for example.

[0536] 3) Clipping. Digital signal from ADC may contain values that exceed maximum capacity of a digital system, for example. To prevent this, a signal may be clipped, which may involve truncating signal to a maximum value, for example. Clipping may occur due to various reasons, such as, for example, saturation (e.g., digital system's dynamic range may be exceeded, causing signal to become distorted) or noise (e.g., signal may be contaminated with noise, which may cause signal to exceed maximum value), for example.

[0537] 4) Filtering. A clipped digital signal may then be filtered to remove unwanted frequencies, such as noise, hum, or other interference, for example. Filtering may be implemented, in whole or in part, using digital filters, such as finite impulse response (FIR) filters or infinite impulse response (IIR) filters, for example.

[0538] 5) Feature Extraction. In one or more embodiments, filtered signal may be analyzed to extract relevant features, such as, for example: Amplitude: loudness of signal, for example; Frequency: pitch or tone of signal, for example; or Spectral characteristics: distribution of energy across different frequency bands, for example. In one or more embodiments, these example features may be used to represent a sound wave in a more abstract and compact form, for example.

[0539] 6) Phoneme Recognition. In one or more embodiments, phonemes may comprise building blocks of spoken language, and may be used to form words, sentences, or texts, for example. In the context of language processing, phonemes may be used to represent the input of a speech recognition system, which may recognize spoken language or may transcribes spoken language into written text, in one or more embodiments. Several non-limiting examples of phonemes are provided: (a) Vowels: / a / , / e / , / i / , / o / , / u / , .; (b) Consonants: / p / , / t / , / k / , / m / , / n / , .; (c) Diphthongs: / ai / , / au / , / oi / , .; (d) Triphthongs: / aie / , / auo / , .

[0540] In one or more embodiments, extracted features may be fed into a phonemerecognition system, which may utilize machine learning algorithms to identify phonemes (e.g., units of sound) present in a signal, for example. In one or more embodiments, phonemes may comprise building blocks of spoken language, and recognizing them may be a factor for speech recognition and synthesis, for example.

[0541] Decoding. In one or more embodiments, recognized phonemes may be decoded into a sequence of phonemes, which represents a spoken language, for example. A sequence may be utilized, at least in part, to generate text or to synthesize speech, in one or more embodiments. In one or more embodiments, a process of decoding recognized phonemes into a sequence of phonemes, which may represent a spoken language, may be a factor in speech recognition or synthesis, for example.

[0542] In one or more embodiments, decoding recognized phonemes may include one or more of the following aspects. Phoneme-to-Grapheme Mapping: In one or more embodiments, graphemes may comprise the building blocks of written language and may be used to form words, sentences, or textual output. In the context of language processing, graphemes may represent an output of a speech recognition system that recognizes spoken language and transcribes it into written text. Non-limiting examples of graphemes may include: (a) Letters: e.g., a, b, c, d, e (b) Letter Combinations: e.g., th, ch, sh, qu (c) Digraphs: e.g., ch, sh, th (d) Trigraphs: e.g., tch, sch (e) Symbols: e.g., punctuation marks such as ., ,, !, ?

[0543] In one or more embodiments, recognized phonemes may be mapped to a corresponding grapheme (e.g., a unit of written language, such as a letter or a combination of letters, for example). This mapping may be accomplished, at least in part, using a phoneme-to-grapheme dictionary, for example, which may comprise a pre-trained model that associates phonemes with their corresponding graphemes, in one or more embodiments.

[0544] b) Grapheme-to-Text Conversion. In one or more embodiments, graphemes may be converted into text using a grapheme-to-text converter, for example. In one or more embodiments, this converter may utilize, at least in part, a set of rules or algorithms to determine a correct sequence of characters that corresponds to graphemes, for example.

[0545] c) Text Normalization. In one or more embodiments, resulting text may be normalized to remove specified (e.g., unnecessary) characters, such as spaces or punctuation marks, for example. This may be done, at least in part, to ensure that text is in a consistent format and may be easily processed by a next stage, for example. In one or more embodiments, text normalization may comprise transforming text data into a standardized format to facilitate analysis, processing, or comparison, for example. A goal of text normalization may be to remove specified (e.g., unnecessary) characters, such as spaces or punctuation marks, for example, to create a clean or consistent dataset, for example. Example aspects of textnormalization includes, but are not limited to, the following:

[0546] c1) Tokenization. In one or more embodiments, text normalization includes a tokenization operation, for example. In one or more embodiments, tokenization may involve breaking down text into individual words or tokens, for example. This may be done, at least in part, by identifying spaces between words or treating individual words as separate tokens, for example.

[0547] c2) Stopword Removal. In one or more embodiments, stopwords may comprise common words such as "the," "and," "a," . that may not carry much meaning in a text, for example. In one or more embodiments, stopword removal includes removing stopwords from a text to reduce dimensionality of data or to improve accuracy of analysis, for example.

[0548] c3) Stemming or Lemmatization. In one or more embodiments, stemming, or lemmatization may comprise techniques utilized, in whole or in part, to reduce words to their base form, for example. Stemming may involve removing suffixes from words to reduce them to their base form, while lemmatization may involve reducing words to their dictionary form, in one or more embodiments. In one or more embodiments, stemming, or lemmatization may comprise text preprocessing techniques that may be used, for example, in Natural Language Processing (NLP) to reduce words to their base or root form, for example. While these approaches may share this goal, they achieve it in different ways and with varying levels of accuracy, for example.

[0549] Stemming: Simpler approach: In one or more embodiments, stemming may chop off suffixes from words to get a base form, for example. It may use a set of rules or algorithms to identify or remove common suffixes like "-ing," "-ed," or "-s," for example. Faster Processing: In one or more embodiments, stemming may be generally faster than lemmatization because it may rely on simpler rules, for example. Less Accurate: However, stemming may be quite aggressive and may sometimes remove crucial parts of a word, leading to incorrect base forms, in one or more embodiments. For example, stemming "running" might result in "run," which may be a valid word, but stemming "agrees" might result in "agre," which may be not a real word, for example.

[0550] Lemmatization: Dictionary Lookup: In one or more embodiments, lemmatization may take a more sophisticated approach, for example. In one or more embodiments, it may use, at least in part, a morphological dictionary or analysis to map a word to its dictionary base form, also called a lemma, for example. More Accurate: This dictionary lookup may help ensure a resulting base form is an actual word in a language, in one or more embodiments. For example, lemmatization would correctly identify lemma of "running" as "run" and a lemma of "agrees" as "agree," for example. Slower Processing: In one or more embodiments, due at least in part to adictionary lookup or potentially more complex analysis, lemmatization may be generally slower than stemming, for example. Table 1 below summarizes some example differences between stemming and lemmatization, in one or more embodiments. Table 1: Comparison of Stemming and Lemmatization

[0551] In one or more embodiments, selection between stemming and lemmatization may depend on application-specific goals. Where grammatical precision or semantic clarity is paramount—such as in sentiment analysis or topic modeling—lemmatization may be preferred, even if it incurs greater computational overhead. For tasks where identifying word variants is sufficient or where processing speed is prioritized—such as in information retrieval—stemming may be an appropriate alternative.

[0552] In one or more embodiments, preprocessing may include one or more of the following text normalization operations:

[0553] (a) Removing Special Characters: In one or more embodiments, characters such as punctuation marks, numbers, or symbols may be removed to generate a cleaner and more consistent dataset.

[0554] (b) Removing Extra Spaces: In one or more embodiments, extraneous spaces between words may be eliminated to enforce uniform spacing across the dataset.

[0555] (c) Converting to Lowercase: In one or more embodiments, text may be converted entirely to lowercase to create a consistent format for processing and comparison.

[0556] (d) Removing HTML Tags: In one or more embodiments, HTML tags may be removed to ensure text cleanliness, especially when sourced from web content.

[0557] (e) Removing URLs: In one or more embodiments, web addresses or hyperlinks may be stripped from text to reduce noise in downstream tasks.

[0558] (f) Removing Email Addresses: In one or more embodiments, email addressesmay be eliminated from the dataset to focus analysis on core content.

[0559] (g) Removing Duplicate Special Characters: In one or more embodiments, repeated punctuation or non-alphanumeric characters may be removed to simplify the text structure.

[0560] (h) Removing Extra Characters: In one or more embodiments, formatting characters such as tabs, carriage returns, or newline characters may be removed to produce a normalized text stream.

[0561] (i) Normalizing Text: In one or more embodiments, normalization may include the removal of any residual punctuation, numbers, or special characters, resulting in a consistent and standardized dataset.

[0562] In one or more embodiments, the resulting normalized text may then be used for further stages of analysis, interpretation, or machine processing.

[0563] In one or more embodiments, normalized text may be input into a language model configured to compute probabilistic relationships among word sequences based on training data. Such language models may be statistical, rule-based, or neural, and may be trained on large corpora to generate coherent and contextually accurate output. Language model selection may depend on target application or domain-specific requirements. Examples of language models include, but are not limited to, the following:

[0564] (i) N-gram Language Model: In one or more embodiments, an N-gram model may predict word sequences based on probabilities computed from contiguous prior terms in the sequence.

[0565] (ii) Markov Chain Language Model: In one or more embodiments, a Markov chain model may infer the likelihood of future words based on a current state or a series of prior states.

[0566] (iii) Recurrent Neural Network (RNN) Language Model: In one or more embodiments, RNN models may learn to predict subsequent tokens by utilizing feedback loops and historical states. The following are example types of RNN-based language models:

[0567] (A) Long Short-Term Memory (LSTM): In one or more embodiments, an LSTM model may incorporate memory cells designed to maintain long-term dependencies within text sequences.

[0568] (B) Transformer Language Model: In one or more embodiments, a transformer model may utilize self-attention mechanisms to model relationships across entire input sequences.

[0569] (C) Word Embedding Language Model: In one or more embodiments, models trained with embedding techniques may represent words as high-dimensional vectors, allowingsemantic similarity operations.

[0570] (D) Convolutional Neural Network (CNN) Language Model: In one or more embodiments, CNN-based models may apply convolutional filters to textual sequences for feature extraction or pattern learning.

[0571] (e) Text-to-Speech (TTS) Synthesis: In one or more embodiments, outputs from a language model may be passed through a TTS synthesizer to convert text into human-like speech. A TTS synthesizer may be implemented in hardware or software and may use natural language processing, acoustic modeling, and speech synthesis algorithms to convert written text into audio output. In one or more embodiments, TTS synthesis may include one or more of the following stages:

[0572] (a) Text Analysis: Analyzes sentence structure and semantics.

[0573] (b) Phonetic Transcription: Converts written language into a phonetic alphabet.

[0574] (c) Speech Synthesis Engine: Maps phonetic output into sound sequences.

[0575] (d) Voice Rendering Engine: Generates naturalistic audio signals using trained acoustic models or rule-based parameters.

[0576] In one or more embodiments, a TTS synthesizer may find advantageous use in a variety of applications, including, but not limited to:

[0577] (a) Assistive Technology: TTS synthesizers may be used in whole or in part to assist people with disabilities, such as visual impairments or dyslexia, by providing a spoken version of written text, for example.

[0578] (b) Automated Customer Service: TTS synthesizers may be used, at least in part, to provide automated customer service, such as answering frequently asked questions or providing product information, for example.

[0579] (c) Language Learning: TTS synthesizers may be used to provide language learning materials, such as audio recordings of vocabulary words or phrases, for example.

[0580] (d) Entertainment: TTS synthesizers may be used to create interactive stories or games that use spoken language, for example.

[0581] In one or more embodiments, example types of TTS synthesizers include, but are not limited to:

[0582] (a) Rule-Based Systems: These systems may use a set of rules to generate spoken language, or rules may be based on syntax and / or semantics of a text.

[0583] (b) Statistical Systems: These systems may use statistical models to generate spoken language, or models may be trained on a relatively larger corpus of text or speech data, for example.

[0584] (c) Neural Network-Based Systems: These systems may use neural networks togenerate spoken language, or networks may be trained on a large corpus of text and speech data, for example.

[0585] In one or more embodiments, examples of TTS synthesizers include, but are not limited to:

[0586] (a) Amazon® Polly®: A cloud-based TTS synthesizer that may utilize, at least in part, neural networks to generate high-quality speech, for example.

[0587] (b) Google® Text-to-Speech: A cloud-based TTS synthesizer that may use, in whole or in part, statistical models to generate speech, for example.

[0588] (c) IBM Watson® Text to Speech: A cloud-based TTS synthesizer that may use, at least in part, neural networks to generate high-quality speech, for example.

[0589] (d) Microsoft Azure Cognitive Services Speech: A cloud-based TTS synthesizer that may utilize, in whole or in part, neural networks to generate high-quality speech, for example.

[0590] Further, at least in part by various example permutations of machine-learning algorithms, one or more embodiments may comprise an example implementation of speech-to-speech synthesis derived, in whole or in part, by comparing terms (e.g., precise terms) used in 1st stage and 3rd stage (of 3-stages) of interpretation, translation, transliteration, or transcription (e.g., interpretation or translation into original words of native language speaker).

[0591] In one or more embodiments, when both parties agree (e.g., tacitly) upon comprehension of terms, or no further negotiation may be pursued, at that moment a Speech-to-Speech Multi-lingual Language Model may be built without an AI inference stage (e.g., stage-2) in the middle, for example.

[0592] For one or more embodiments, such an AI-agent-built speech-to-speech interpreter or translator may be considered a significant advancement in the state of the art, for example.

[0593] In one or more embodiments, decoding recognized phonemes may additionally include post-processing, which includes adjusting synthesized speech to make it sound more natural or human-like, for example.

[0594] For example, post-processing includes adjusting pitch, tone, or volume of speech, for example, as well as adding in natural pauses or inflections, for example.

[0595] To review, in one or more embodiments, a microphone or other audio processing component may handle recognized phonemes at least in part by performing one or more of:

[0596] (a) Capturing sound waves using, at least in part, a microphone;

[0597] (b) Converting analog signal to a digital signal using, at least in part, an ADC;

[0598] (c) Clipping digital signal to prevent saturation or noise. In one or moreembodiments, a separate thread may carry unclipped audio for processing using any of a range of possible algorithms in parallel to derive an increased contextual understanding of what was said, how it was said, who said it or why they said what they did, for example;

[0599] (d) Filtering signal to remove unwanted frequencies;

[0600] (e) Extracting relevant features from signal;

[0601] (f) Recognizing phonemes using, in whole or in part, machine learning algorithms;

[0602] (g) Decoding recognized phonemes into a sequence of phonemes;

[0603] (h) Mapping phonemes to graphemes using, at least in part, a phoneme-to-grapheme dictionary. It maps sounds (phonemes) to their written representations (graphemes), for example.

[0604] This may be useful for, for example:

[0605] (i) Learning to read: By seeing how sounds connect to letters, children decode unfamiliar words and improve their reading skills, for example;

[0606] (j) Text-to-speech systems: These programs rely on phoneme-to-grapheme dictionaries to convert written text into spoken language, for example;

[0607] (k) Linguistic research: Understanding phoneme-grapheme relationships helps researchers analyze languages and develop speech recognition technologies, for example;

[0608] (l) Converting graphemes to text using a grapheme-to-text converter, for example. A grapheme-to-text converter, unlike its counterpart phoneme-to-grapheme converter, takes written text as input and aims to produce a most likely corresponding standard text as output, for example.

[0609] In one or more embodiments, converting graphemes to text includes, for example, one or more of the following.

[0610] Input Processing. In one or more embodiments, a converter first receives a grapheme sequence, which may comprise, for example: (a) Informal Text: Text written with abbreviations, slang, or emojis. (e.g., "bday" for "birthday"); (b) Misspellings: Words with typing errors or intentional creative spellings. (e.g., "luv" for "love"); (c) Dialectal variations: Text with words or spellings specific to a particular region, for example. (e.g., "y'all" in Southern US English).

[0611] Error Detection and Correction. In one or more embodiments, a converter may employ various techniques to identify potential errors or non-standard spellings in input graphemes, for example. This includes, for example: (a) Dictionary Lookup: Checking if grapheme sequence exists in a standard dictionary; (b) Context Analysis: Considering surrounding graphemes and their known spellings to identify potential mistakes. (e.g., "bday"following "happy" suggests it likely refers to "birthday"); (c) Pattern Recognition: Utilizing knowledge of common misspelling patterns to detect errors. (e.g.,"teh" likely intended to be "the").

[0612] Language Model Integration:

[0613] In one or more embodiments, for some converters, a statistical language model may play a role, for example. Such a model may analyze corrected grapheme sequence or surrounding text (if provided) to predict a more probable (e.g., most probable) standard text based on language patterns or word probabilities, for example.

[0614] Text Output. Additionally, in one or more embodiments, a converter may output a corrected or standardized text version of an original grapheme sequence, for example. In one or more embodiments, this includes an original sequence if no errors were detected, a corrected version based on dictionary lookup or error correction techniques, or a suggestion from a language model if input may be ambiguous or creative spelling may be intentional, for example.

[0615] For example: (a) Input: "hbd 2 my dearest sis!" (birthday greeting with abbreviation and informal term); (b) Error Detection: "hbd" not found in dictionary, "sis" is informal, for example; (c) Correction: "Happy birthday to my dearest sister!" (using dictionary lookup and context).

[0616] It may be noted that grapheme-to-text conversion accuracy may depend on sophistication of algorithms or quality of training data used, in one or more embodiments. In one or more embodiments, a converter might not always be able to perfectly interpret creative spellings or informal language, especially if context is limited, for example. Further, some converters may offer options to control a level of correction, thereby allowing users to choose between stricter standardization or preserving some informality, for example. Such example converters may find advantageous use in a variety of applications, including, but not limited to:

[0617] (a) Social media text processing (correcting informal language and abbreviations);

[0618] (b) Search engine optimization (improving search results by understanding misspelled queries);

[0619] (c) Machine translation (handling dialectal variations and slang);

[0620] (d) Text analysis tasks (ensuring consistent format for analysis);

[0621] (e) Normalizing text to remove unnecessary characters;

[0622] (f) Passing text through a language model to predict a probability of a sequence of words;

[0623] (g) Converting text to spoken language using a text-to-speech synthesizer;

[0624] (h) Post-processing synthesized speech to make it sound more natural andhuman-like.

[0625] Microphone. In one or more embodiments, a microphone, such as a condenser or dynamic microphone, converts sound waves into an electrical signal. A microphone’s diaphragm may vibrate in response to sound waves, causing a tiny electrical current to flow through a microphone’s internal circuitry. Present-day laptops, iPhones, Android devices, tablets, . often use various types of microphones to capture audio signals. In one or more embodiments, a multi-processor, multi-core, DSP processor may enable implementation, in whole or in part, of non-clipping algorithms on a separate processing thread, for example. In one or more embodiments, a purpose of no-clipping may be to capture lower (e.g., very lowest) of sounds emanating from a speaker who utters speech.

[0626] Table 2 below provides an example comparison of frequency ranges for example types of microphones. Note: (a) Hz refers to Hertz, which is a unit of measurement for frequency; (b) kHz refers to kilohertz, which is 1,000 Hz; (c) frequency ranges listed are example ranges for individual microphone types, but of course these ranges may vary depending on a specific design or implementation. Table 2. Frequency Ranges for Example Microphone Types

[0627] Below is a brief explanation of several example microphone types or example implementations:

[0628] (a) Condenser Microphones: Condenser microphones may be designed to capture a wide range of frequencies, from very low bass notes to very high treble notes, for example. This may make them suitable for recording a variety of sounds, from music to voiceovers, for example. These may be a common type of microphone used in laptops, tablets, or smartphones, for example. They may be small, lightweight, or may capture a wide range of frequencies. Examples include:

[0629] (i) Laptop microphones: Many laptops use condenser microphones, such as ones found in Apple® MacBook Pro®, Dell® XPS®, or HP® Envy®, for example;

[0630] (ii) Smartphone microphones: Most smartphones, including iPhones and Android devices, use condenser microphones, for example;

[0631] (b) MEMS Microphones (Micro-Electro-Mechanical Systems): MEMS microphones may be designed to capture a narrower range of frequencies, focusing on mid-range frequencies where most human speech and music falls. This may make them suitable for applications where noise reduction may be a factor, such as voice assistants or video conferencing, for example. These may be small, lower-power microphones that may use, in whole or in part, a mechanical structure to convert sound waves into electrical signals. They may be often used in, for example:

[0632] (i) Smartphones: Many smartphones, including budget-friendly options, use MEMS microphones, for example;

[0633] (ii) Tablets: Some tablets, like Apple® iPad®, use MEMS microphones, for example;

[0634] (c) Dynamic Microphones: Some dynamic microphones may be designed to capture a narrower range of frequencies, focusing on mid-range frequencies where most human speech and music falls, for example. This may make them suitable for applications where robustness or durability may be factors, such as live performances or public speaking. Such microphones may handle high sound pressure levels, for example. They may be used in, for example:

[0635] (i) Laptops: Some laptops, like Dell® Inspiron® and HP® Pavilion®, use dynamic microphones;

[0636] (ii) Tablets: Some tablets, like Microsoft® Surface®, use dynamic microphones;

[0637] (d) Digital Microphones: Digital microphones may be designed to capture a wide range of frequencies, similar to condenser microphones. However, they often use digital signal processing to enhance or adjust frequency response, making them suitable for a variety ofapplications from music recording to voice assistants, for example. These microphones may use, in whole or in part, digital signal processing to enhance audio quality and reduce noise. They may be used in, for example:

[0638] (i) Smartphones: Some high-end smartphones, like Samsung® Galaxy® S series, use digital microphones;

[0639] (ii) Tablets: Some tablets, like Apple® iPad Pro®, use digital microphones.

[0640] In one or more embodiments, some devices may use a combination of these microphone types, such as a condenser microphone with a MEMS microphone for noise reduction, for example. It may be noted that that a type of microphone used may affect audio quality, noise reduction, or overall performance of a device.

[0641] Analog-to-Digital Conversion: In one or more embodiments, an electrical signal from a microphone may be sent to an analog-to-digital converter (ADC), which converts a continuous analog signal into a digital signal. An ADC may sample the signal at a specific rate, typically between 8 kHz to 44.1 kHz, depending on the desired quality of audio.

[0642] Fourier transforms may be applied during the analog-to-digital conversion (ADC) process in speech recognition systems, in one or more embodiments. For example, Fourier transforms may be used, in whole or in part, to filter out unwanted frequencies in the analog signal, such as noise, hum, or other interference. This may help to improve the signal-to-noise ratio (SNR) or to reduce distortion.

[0643] Further, Fourier transforms may be used, in whole or in part, to apply bandpass filtering to an analog signal, which may help to focus on a frequency range of interest (e.g., range of human speech).

[0644] Additionally, Fourier transforms may be used, in whole or in part, to apply windowing functions to an analog signal, which helps to reduce spectral leakage and improve the accuracy of the ADC, for example.

[0645] Also, for example, Fourier transforms may be used, in whole or in part, to analyze spectral characteristics of an analog signal, which may help to identify patterns or features that may be used for speech recognition, for example.

[0646] Example Fourier transform techniques that may be implemented, in whole or in part, in ADC includes, but are not limited to: (a) Fast Fourier Transform (FFT): A fast and efficient algorithm for computing discrete Fourier transform (DFT) of a sequence; (b) Discrete Cosine Transform (DCT): A transform that is similar to FFT, but may be more suitable for signals with a limited frequency range; (c) Short-Time Fourier Transform (STFT): A transform used, in whole or in part, to analyze spectral characteristics of a signal over a short time interval.

[0647] In one or more embodiments, by applying Fourier transforms during the ADCprocess, for example, speech recognition systems may improve the quality or accuracy of a digital signal, which may be a factor in recognizing spoken words or phrases.

[0648] However, other algorithms besides Fourier Transforms may be utilized in audio processing. Non-limiting examples of non-Fourier approaches include:

[0649] (a) Wavelet Transform: Wavelet Transform may comprise a time-frequency analysis technique that may be used, in whole or in part, to analyze audio signals. It may be used, in whole or in part, for analyzing signals that have both time and frequency components, such as audio signals with multiple frequencies and harmonics, for example;

[0650] (b) Short-Time Fourier Transform (STFT): STFT may comprise a modification of FFT that may be used, in whole or in part, to analyze audio signals in the time-frequency domain. It may be used for analyzing signals that have a time-varying frequency content, such as audio signals with changing pitch or tempo, for example;

[0651] (c) Modulation Frequency Transform (MFT): MFT may comprise a time-frequency analysis technique that may be similar to STFT, but it may use, in whole or in part, a different mathematical approach to analyze a signal;

[0652] (d) Chirp-Z Transform: Chirp-Z Transform may comprise a time-frequency analysis technique that may be similar to FFT, but it may use, in whole or in part, a different mathematical approach to analyze a signal. It may be used, in whole or in part, for analyzing signals that have a non-stationary frequency content, for example;

[0653] (e) Spectrogram: A spectrogram may comprise a visual representation of frequency content of an audio signal over time. It may be a powerful tool for analyzing or processing audio signals, and it may be used, in whole or in part, in conjunction with other algorithms to provide a more complete understanding of a signal;

[0654] (f) Constant-Q Transform: Constant-Q Transform may comprise a time-frequency analysis technique that may be similar to FT, but it may use, in whole or in part, a different mathematical approach to analyze a signal. It may be used for analyzing signals that have a non-stationary frequency content, for example;

[0655] (g) Gabor Transform: Gabor Transform comprises a time-frequency analysis technique that may be similar to FT, but it may use, in whole or in part, a different mathematical approach to analyze a signal. It may be used for analyzing signals that have a non-stationary frequency content, for example.

[0656] Digital Signal Processing (DSP). In one or more embodiments, a digital signal may be processed using one or more multi-processor, multi-core, digital signal processors performing SIMD, MIMD, SISD, or MISD parallelization models processing (DSP) techniques to enhance audio quality, reduce noise, or adjust gain. These operations may prepare an audiosignal for speech recognition.

[0657] Example parallelization approaches include, for example:

[0658] (a) SISD (Single Instruction Single Data): A more basic type of architecture and may not be advantageous for some embodiments, as many DSP processors may be designed for some form of parallelism, for example;

[0659] (b) MISD (Multiple Instruction Single Data): May find advantageous use for some embodiments, such as at least some embodiments described herein;

[0660] (c) MIMD (Multiple Instruction Multiple Data): May be a powerful approach for general-purpose processors, for example. In one or more embodiments, may be utilized advantageously for parallelized interpretation or translation, for example;

[0661] (d) SIMD (Single Instruction Multiple Data): May comprise a relatively common type of parallelism that may be advantageously used in DSP processors, for example. SIMD architectures may excel at performing a same operation on multiple data elements concurrently, for example. This approach may find advantageous use for a number of DSP algorithms that may involve repetitive calculations on larger datasets, such as audio samples or sensor readings, in one or more embodiments.

[0662] In one or more embodiments, DSP processors may leverage a combination of SIMD instructions with specialized hardware accelerators to achieve improved performance for specified signal processing tasks. This approach may provide a good balance between efficiency and flexibility for a wide range of DSP applications, for example.

[0663] Audio Compression. Processed audio signal may be compressed using algorithms such as, for example, MP3, AAC, or Opus, to reduce a file size or make it more efficient for transmission. In at least some circumstances, compressing processed audio signal (e.g., after DSP processing) may not be advised for NLP applications, for example. There may be several factors related to compression of digital audio content, including examples discussed below.

[0664] (a) Loss of Information: Compression algorithms, especially lossy ones like MP3, may discard some audio data to achieve smaller file sizes. This discarded data could be crucial for accurate NLP tasks in some circumstances. Nuances in tone, subtle background noise, or even specific frequencies may hold valuable information for tasks like sentiment analysis or speaker identification, for example. Compression may, in some circumstances, remove these subtleties, negatively impacting effectiveness of NLP algorithms. Of course, suitability of compression algorithms in a context of one or more embodiments described herein may depend at least in part on characteristics of particular compression algorithms.

[0665] (b) Focus on Clarity: NLP may often rely, in some embodiments, on a relatively full spectrum of information within audio content. While DSP processing might enhanceparticular aspects of audio (e.g., noise reduction or equalization), raw, uncompressed audio may hold valuable details for NLP models, in one or more embodiments.

[0666] In one or more embodiments, there may be scenarios where compressing processed audio might be considered.

[0667] (a) Limited Storage / Transmission: In some circumstances, such as where storage space or bandwidth limitations may be challenging, compressing after processing may be utilized. In one or more embodiments, specified compression algorithm may comprise a lossless compression format such as, for example, FLAC, AAC, ., which may preserve all original audio data.

[0668] (b) Focus on Specific Information: In one or more embodiments, if a particular NLP task uses a particular type of information (e.g., keywords or basic speech content) and may not utilize full nuances, then compressing with a format that prioritizes speech clarity, such as Opus, for example, may make sense.

[0669] Overall, compressing processed audio for NLP applications may be considered in light of factors mentioned above, for example. A potential loss of information may outweigh benefits of reduced file size, for example, in some embodiments or applications. In one or more embodiments, it may be more advantageous to prioritize integrity of audio data, such as for more accurate NLP processing. For at least these reasons, one or more embodiments, such as may implement CICAI in whole or in part, may utilize uncompressed audio coming out of a DSP processor. Of course, as mentioned, other embodiments may utilize compression algorithms, including lossless compression algorithms, for example.

[0670] Transmission: In one or more embodiments, compressed audio signal may be transmitted to a speech recognition system built into circuitry of an audio interface or other computing device component, for example, either through a wired connection (e.g., USB) or wirelessly (e.g., Bluetooth, Wi-Fi).

[0671] Speech Recognition Engine: In one or more embodiments, a speech recognition engine, which may comprise a software-based system, receives compressed audio signal and begins process of recognizing spoken words. One may think of speech recognition engine as a detective team with specialized skills:

[0672] (a) A microphone may be likened to a witness collecting evidence (e.g., sound waves);

[0673] (b) A preprocessor prepares evidence (e.g., converts sound to digital data);

[0674] (c) A feature extractor identifies key details in evidence (e.g., extracts relevant audio features);

[0675] (d) A machine learning model may be likened to a lead detective, with vastexperience (e.g., trained on a massive dataset) who analyzes evidence (e.g., features) and identifies suspects (e.g., phonemes);

[0676] (e) A language model may be likened to a consultant who helps connect dots and considers bigger picture (e.g., grammar and context) to suggest most likely culprit (e.g., word or phrase).

[0677] Acoustic Modeling: In one or more embodiments, a speech recognition engine uses acoustic models to analyze audio signal and identify patterns of sound waves that correspond to specific phonemes and words.

[0678] Language Modeling. In one or more embodiments, an engine also uses language models to understand context and grammar of spoken language, allowing it to disambiguate words and phrases, for example.

[0679] Decoding. In one or more embodiments, an engine decodes recognized audio signal into text, using a combination of acoustic or language models. A resulting text may be then output to a user, often through a graphical user interface (GUI) or a text-to- speech system, for example.

[0680] Post-processing. In one or more embodiments, an additional, perhaps final, stage includes post-processing recognized text to correct any errors, improve readability, or enhance overall accuracy of speech recognition system, for example.

[0681] In one or more embodiments, a microphone captures audio signal of human speech, which may be processed, compressed, or transmitted to speech recognition system. In one or more embodiments, a system uses acoustic and language models to recognize spoken words and convert them into text, which may be then output to a user. Acoustic or language models used in speech recognition systems includes, but are not limited to, the example shown below:

[0682] Acoustic Models. In one or more embodiments, various acoustic models may be used to represent or process the statistical or learned characteristics of speech. Examples of such models include:

[0683] (a) Hidden Markov Models (HMMs): HMMs may comprise statistical models that describe a probability distribution of acoustic features (e.g., mel-frequency cepstral coefficients, MFCCs) in speech. They may be used to model acoustic characteristics of phonemes, words, and phrases, for example;

[0684] (b) Gaussian Mixture Models (GMMs): GMMs may comprise statistical models that describe probability distribution of acoustic features in speech. They may be used to model acoustic characteristics of phonemes, words, and phrases, for example;

[0685] (c) Deep Neural Networks (DNNs): DNNs may comprise artificial neuralnetworks that may be trained to recognize patterns in acoustic features. They may be used to model acoustic characteristics of phonemes, words, and phrases, for example;

[0686] (d) Convolutional Neural Networks (CNNs): CNNs may comprise artificial neural networks that may be trained to recognize patterns in acoustic features. They may be used to model acoustic characteristics of phonemes, words, and phrases, for example;

[0687] (e) Recurrent Neural Networks (RNNs): RNNs may comprise artificial neural networks that may be trained to recognize patterns in acoustic features over time. They may be used to model acoustic characteristics of phonemes, words, and phrases, for example.

[0688] Language Models. In one or more embodiments, various language models may be utilized to represent, predict, or analyze linguistic structure, semantic content, or grammatical relationships of spoken or written language. Examples include:

[0689] (a) N-gram Models: N-gram models may comprise statistical models that describe a probability distribution of word sequences in a language. They may be used to model context and grammar of spoken language, for example;

[0690] (b) Markov Chain Models: Markov chain models may comprise statistical models that describe a probability distribution of word sequences in a language. They may be used to model context and gramm ar of spoken language;

[0691] (c) Part-of-Speech (POS) Tagging: POS tagging may be a technique that may be used to identify a part of speech (e.g., noun, verb, adjective) of each word in a sentence, for example;

[0692] (d) Named Entity Recognition (NER): NER may be a technique that may be used to identify named entities (e.g., people, places, organizations) in a sentence, for example;

[0693] (e) Dependency Parsing: Dependency parsing may be a technique that may be used to analyze a grammatical structure of a sentence, including relationships between words, for example;

[0694] (f) Language Modeling using Word Embeddings: Word embeddings, such as Word2Vec or GloVe, may be used to represent words as vectors in a high-dimensional space. Language models may use these embeddings to capture semantic relationships between words, for example;

[0695] (g) Long Short-Term Memory (LSTM) Networks: LSTMs may comprise a type of RNN that may be used to model context and grammar of spoken language, for example;

[0696] (h) Transformers: Transformers may comprise a type of neural network architecture that may be used to model context and grammar of spoken language, for example.

[0697] Other Models: (a) Speaker Recognition Models: Speaker recognition models may be used to identify a speaker of a given audio recording, for example; (b) Noise ReductionModels: Noise reduction models may be used to reduce background noise and improve quality of an audio signal, for example; (c) Emotion Recognition Models: Emotion recognition models may be used to recognize emotions expressed in a given audio recording, for example.

[0698] These are just a few examples of many acoustic or language models that may be used, in whole or in part, in speech recognition systems. In one or more embodiments, specific models used may vary depending on application, type of speech being recognized, or desired level of accuracy.

[0699] Preprocessing: In one or more embodiments, an audio signal may be preprocessed, perhaps in specialized circuitry, to enhance quality of signal or remove noise. This includes steps such as, for example: (a) Filtering: specialized circuitry is removing low-frequency noise and high- frequency hiss, for example; (b) Normalization: adjusting a volume of a signal to a consistent level; (c) De-noising: removing random noise from a signal; (d) Echo cancellation: removing echoes from a signal.

[0700] Feature Extraction: In one or more embodiments, preprocessed audio signal may be analyzed to extract relevant features that may be used to identify spoken words. These features includes, for example: (a) Mel-Frequency Cepstral Coefficients (MFCCs): a set of coefficients that describe spectral characteristics of a signal, for example; (b) Pitch: frequency of a speaker's voice; (c) Energy: intensity of a signal; (d) Formants: resonant frequencies of a speaker's voice.

[0701] Acoustic Modeling: In one or more embodiments, extracted features may be used, in whole or in part, to create a statistical model of acoustic characteristics of spoken words, for example. This model may be used, in whole or in part, to predict a likelihood of a particular word being spoken based on acoustic features of a signal, for example.

[0702] Language Modeling: A statistical model may be combined with a language model to predict a most likely word sequence given acoustic features of a signal, for example. A language model may be trained on a large corpus of text data and may be used to predict a probability of a particular word sequence given context of a conversation, for example. In one or more embodiments, due at least in part to a Personal Profile of a user, such as alluded to previously, Context, or Formality may be assigned (e.g., correctly) by an AI Agent tasked to do so, for example.

[0703] Decoding: In one or more embodiments, a predicted word sequence may be decoded to produce text output (e.g., final output), for example. This may involve steps such as, for example: (a) Word recognition: identifying individual words in a spoken sentence, for example; (b) Sentence recognition: identifying a sentence structure or syntax, for example; (c) Post-processing: correcting errors or improving accuracy of an output, for example.

[0704] In one or more embodiments, it may be advantageous to implement additional parallel microphone operations to accelerate speech-to-text synthesis, insertion of diacritics, voice sampling, Fourier-transform tuning, feature detection (relaxation of glottis, lips contacting teeth, tongue contacting teeth - artifacts that give us a more precise trigger of a completion of an audible utterance), . Moreover, one or more embodiments may use, in whole or in part, quality cameras of computers or portable devices, for example, to detect micro-gestures to inform an observer of certain artifacts of “revealing” actions that may be statistically known to incline to “deception”, “honesty”, “concealment”, “on guard” .

[0705] Above, example embodiments pertaining to a speech-to-text stage of a multi-stage interpretation or translation approach have been described. As previously mentioned, another stage in an example multi-stage interpretation or translation approach may comprise a text-to-text stage, for example.

[0706] In one or more embodiments, a text-to-text stage of an example multi-stage interpretation or translation approach may comprise semiological or hermeneutical evaluation or filtering, for example, based at least in part, for some embodiments, on neural-network computation of weighted relationships among words, for example. In one or more embodiments, semiological evaluation in a text-to-text stage includes one or more of the following example operations. Although operations may be described in a particular order, subject matter is not limited in scope in these respects. Further, in one or more embodiments, operations may be performed concurrently, for example.

[0707] Text Preprocessing: In one or more embodiments, text may be preprocessed to remove any unnecessary characters, such as punctuation, and to convert a text to a standard format, for example.

[0708] Tokenization: In one or more embodiments, a text may be broken down into individual words or tokens, which may then be analyzed to identify a language and meaning of each word, for example.

[0709] Part-of-Speech (POS) Tagging: In one or more embodiments, tokens may be analyzed to identify their part of speech (such as noun, verb, adjective) to help determine a context and meaning of each word, for example.

[0710] Named Entity Recognition (NER): In one or more embodiments, tokens may be analyzed to identify named entities, such as names, locations, and organizations, to help determine context and meaning of each word, for example. In one or more embodiments, Named Entity Recognition (NER) may comprise a subtask of Natural Language Processing (NLP) that focuses on identifying and classifying named entities in text, for example. In one or more embodiments, NER analysis includes, for example:

[0711] Tokenization: In one or more embodiments, tokenization includes breaking down text into individual units called tokens, for example. These tokens may be words, but they may also be punctuation marks or numbers depending on NER system's configuration, for example.

[0712] In one or more embodiments, feature extraction may be performed in response, at least in part, to a tokenized text input. Named Entity Recognition (NER) systems may extract various features from each token that could indicate whether the token corresponds to a named entity.

[0713] (a) Word Identity: In one or more embodiments, the word itself may be evaluated to determine whether it is commonly used as a name

[0714] (i) Example: “Paris”

[0715] (ii) Example: “Dr.” (title)

[0716] (b) Part-of-Speech (POS) Tags: In one or more embodiments, POS tags may be analyzed to identify whether a word is used as a noun phrase

[0717] (i) Noun phrases may indicate persons, locations, or organizations

[0718] (c) Prefixes and Suffixes: In one or more embodiments, linguistic features such as prefixes or suffixes may be evaluated

[0719] (i) Example: Suffix “-ville” may suggest a location

[0720] (d) Contextual Clues: In one or more embodiments, the words surrounding a token may be evaluated to infer context

[0721] (i) Example: “Golden Gate” followed by “Bridge” suggests a landmark

[0722] In one or more embodiments, the extracted features may be input into a machine learning model configured as a classifier. The classifier may be trained to categorize each token into a named entity type.

[0723] In one or more embodiments, the classifier may be trained on a large corpus of labeled text, wherein entities are annotated as Person, Location, Organization, or other defined categories.

[0724] Based at least in part on the extracted features, the model may assign predicted labels to each token, indicating whether it is a named entity and specifying the type of entity.

[0725] The following non-limiting example illustrates the approach:

[0726] (a) Sentence: “Barack Obama, a former president of the United States, visited Paris, France last week.”

[0727] (b) Tokenization Output: [“Barack”, “Obama”, “,”, “the”, “former”, “president”, “of”, “the”, “United”, “States”, “,”, “visited”, “Paris”, “,”, “France”, “last”, “week”]

[0728] (c) Feature Extraction Examples:

[0729] (i) Token: “Barack Obama” – Features: capitalized words, potentially aperson’s name

[0730] (ii) Token: “Paris” – Features: capitalized word, potentially a location

[0731] (iii) Token: “United States” – Features: proper noun phrase, likely a country

[0732] Machine Learning Model: In one or more embodiments, a model analyzes features and assigns labels: (a) "Barack Obama" – Person; (b) "Paris" – Location; (c) "United States" – Location.

[0733] After processing tokens, NER system may identify "Barack Obama," "Paris," and "United States" as named entities, with their corresponding categories, for example.

[0734] In one or more embodiments, effectiveness of NER may depend on quality of training data and features used, for example. Also, in one or more embodiments, NER systems may be customized to recognize specified types of named entities beyond a typical Person, Location, Organization categories, for example.

[0735] Machine Translation: In one or more embodiments, preprocessed text may be translated into target language using a machine translation algorithm, for example. Example machine translation algorithms includes, but are not limited to: (a) Rule-based machine translation: This approach uses a set of rules to translate text, based on grammar and syntax of source and target languages, for example; (b) Statistical machine translation: This approach uses statistical models to translate text, based on frequency and probability of word combinations in source and target languages, for example; (c) Neural machine translation: This approach uses neural networks to translate text, based on patterns and relationships between words in source and target languages, for example.

[0736] Post-processing: In one or more embodiments, translated text may be post-processed to remove errors or inconsistencies, and to improve overall quality of a translation, for example.

[0737] Above, example embodiments pertaining to a speech-to-text and text-to-text stages of a multi-stage interpretation or translation approach have been described. As previously mentioned, an additional stage in an example multi-stage interpretation or translation approach may comprise a text-to-speech stage.

[0738] In one or more embodiments, a text-to-speech (TTS) stage within a multi-stage interpretation or translation architecture may comprise operations for converting text or diacritics into audible speech. Such speech may be rendered through output components including phone speakers, computer systems, tablet devices, or similar platforms.

[0739] In one or more embodiments, a TTS process may include multiple operations executed in sequence or combination. These may include, but are not limited to, preprocessing, tokenization, part-of-speech tagging, named entity recognition, speech synthesis, andpost-processing.

[0740] (a) Text Preprocessing: In one or more embodiments, the textual input may be preprocessed to eliminate superfluous characters—such as punctuation—and normalized into a standard format for consistent handling.

[0741] (b) Tokenization: In one or more embodiments, the preprocessed text may be segmented into discrete units such as words or tokens. These tokens may then be evaluated to determine language type and semantic meaning.

[0742] (c) Part-of-Speech (POS) Tagging: In one or more embodiments, the tokens may be annotated with part-of-speech labels (e.g., noun, verb, adjective), thereby supporting contextual interpretation of each word.

[0743] (d) Named Entity Recognition (NER): In one or more embodiments, the tokenized input may be examined to identify named entities, such as individuals, geographic locations, or institutional names, to further aid in contextual analysis.

[0744] (e) Speech Synthesis: In one or more embodiments, the interpreted text may be passed into a speech synthesis algorithm, which may convert it into audible waveform output.

[0745] (f) Rule-Based Speech Synthesis: This technique may apply a defined grammar and syntactic structure to generate speech based on a language’s formal rules.

[0746] (g) Statistical Speech Synthesis: This approach may employ probabilistic models to construct speech based on observed frequency patterns and word combinations within a target language.

[0747] (h) Neural Speech Synthesis: In one or more embodiments, this method may leverage neural network architectures to synthesize speech using learned patterns and semantic relationships between linguistic components.

[0748] (i) Post-Processing: Following speech synthesis, one or more embodiments may execute post-processing operations to eliminate errors or inconsistencies and to enhance the overall acoustic quality of the generated speech.

[0749] In one or more embodiments, a personalized large language model (pLLM) may include a Small Language Model (SLM) customized to contain words, phrases, idioms, colloquialisms, or semantic references of relevance to a first party engaged in bilateral communication.

[0750] In one or more embodiments, the second party in a bilateral exchange may also have access to an SLM personalized to reflect their own linguistic references and semiotic associations. This bilateral linguistic mapping supports enhanced mutual understanding during dialog.

[0751] In one or more embodiments, linguistic researchers have observed thatindividuals possess unique mental categorizations and intellectual hierarchies, which may be expressed through distinctive sentence structures and word choices.

[0752] In one or more embodiments, the relationship between words (semiotics) and their use in ideologically motivated interpretation—such as through casuistic manipulation of religious texts—may represent a repeatable cognitive-linguistic pattern.

[0753] One such example may involve Salafist Islam, where a participant applies a modern interpretive lens to a fundamentalist religious practice. Frequent recitation of Hadith texts may influence their thought patterns, leading to increasingly frequent and complex citation of Quranic references in both spoken and written language.

[0754] Islam may be not unique in this respect. This very same pattern may be found, for example in fundamentalist Evangelicals, Pentacostals, Church of Jesus Christ of Latter-day Saints, Eastern Byzantine Church, Roman Catholicism (i.e. Ustashe), or cults of sexual dysfunction to name but a few.

[0755] In one or more embodiments, a Small Language Model (SLM) may function as a semantic filter or interpretive lens into cognitive or linguistic patterns associated with an individual, as expressed through speech.

[0756] In one or more embodiments, an autonomous AI-agent, referred to herein as “Gabi,” may be configured as a synthetic persona operative as a proxy to execute non-critical tasks on behalf of a human user.

[0757] In such embodiments, Gabi may be analogized to an internal monologue or silent voice that emerges during reading, where semantic negotiation occurs regarding an author’s intended meaning as interpreted through a reader’s lexical history, educational background, or literary exposure.

[0758] In one or more embodiments, Gabi may participate in an internal dialogic process—such as an “I”-versus-“You” construct—to derive optimal interpretations of linguistic content provided by a speaker or author.

[0759] In this regard, Gabi may serve as a proxy for reflective cognition, and may support AI system advancement through a software component that may be refactored into a synthetic hippocampal module configured for ethical and / or legal evaluation of goal-directed activities.

[0760] In one or more embodiments, such evaluation may include real-time or near real-time processing of semiotic or hermeneutic structures embedded within user utterances to determine interpretive intent.

[0761] In some implementations, the AI-agent may perform evaluative tasks within microsecond-scale execution intervals and may further concatenate appropriate diacritic markersfor subsequent voice synthesis, such as in a text-to-speech (TTS) engine.

[0762] Additionally, in one or more embodiments, the AI-agent may incorporate predictive processing capabilities that are triggered via machine-learning mechanisms, thereby supporting anticipatory behavior based on user context or inferred intent.

[0763] In one or more embodiments, the phrase “Personalized LLM” (pLLM) may be used as a marketing term referring to a Small Language Model that is custom-built for an individual user.

[0764] Such pLLMs may be generated using a software-based factory pattern that supports construction or deployment through automated processes or by autonomous AI-agents utilizing hierarchical blockchain tokens.

[0765] In one or more embodiments, pLLMs and SLMs may be streamlined versions of large-scale language models (e.g., trillion-parameter LLMs), with advantages in training speed, fine-tuning ease, deployment simplicity, and operational cost.

[0766] In one or more embodiments, multi-lingual language models may include functionality for: (a) encoding audible speech phonemes into text using formats such as ASCII, EBCDIC, or Unicode; (b) executing neural-network-based inference to connect terms across languages; and (c) converting equivalent translated text into audible speech.

[0767] In one or more embodiments, pLLMs may be unique in their synthesis of all available data about a single individual into a data storage structure used to create a virtual persona or archetype.

[0768] Such data may include publicly available social media content, resumes, professional employment history, research papers, articles, awards, recommendations, language fluency metrics, personal interests, and preferences including favorite foods.

[0769] Any or all of these attributes may contribute to an “intellectual signature,” which may serve to distinguish that individual in a broader marketplace of ideas.

[0770] In one or more embodiments, the individuation of speech or thought patterns reflected in such a pLLM may be an evolving process over time.

[0771] In one or more embodiments, LLMs may be trained using learning algorithms such as multi-layer neural networks or tensor networks to read and synthesize relationships from large volumes of text.

[0772] Such LLMs may be capable of simulating human-like language outputs through mathematical synthesis, although they may not possess inherent comprehension of meaning unless specifically encoded to perform semantic evaluation algorithms.

[0773] In one or more embodiments, LLMs may function as models in multiple senses, including as multi-modal mathematical constructs that represent stochastically derived linguisticrelationships among words used in everyday language.

[0774] These multi-modal models may be used in computer science to describe and simulate cognitive behavior—such as speech generation, prediction, or optimization—based on computed weightings of word usage or patterns.

[0775] In one or more embodiments, LLMs may be implemented using neural networks, which may be rooted in well-established mathematical learning algorithms.

[0776] Neural networks used in LLMs may be designed to recognize input patterns and perform tasks such as prediction or classification. For instance, given textual input, such networks may predict the next word or character in a sequence, thereby modeling conversational or written discourse—including analysis of recorded content such as podcasts.

[0777] In one or more embodiments, LLMs may be performed by processing input text through multiple layers of feed-forward neural networks, for example, which may be designed to capture different aspects of language, such as syntax, semantics, or pragmatics. One or more embodiments may utilize, in whole or in part, other approaches (e.g. Rules-based Lexical-based), and subject matter is not limited in scope in these respects. One or more embodiments may embrace any or all other prevailing approaches. In one or more embodiments, individual layers may process an input text or may generate an output, which may in turn be fed into a next layer, for example. Such an example process may be repeated multiple times, allowing a model to learn complex patterns or relationships in data, for example.

[0778] FIG.19 is a block diagram depicting an example of language model construction, in accordance with one or more embodiments.

[0779] FIG. 19 may include both self-explanatory operations and those described elsewhere in this specification. The subject matter illustrated in FIG.19 may pertain to various processes including, but not limited to, tokenization, machine transliteration, machine transcription, pre-processing, lexical analysis, morphological analysis, feature extraction, neural networks, machine learning (ML) algorithms, post-processing, machine translation, evaluation, and quality testing.

[0780] In one or more embodiments, the scope of subject matter shown in FIG.19 is not limited to the depicted components. Fewer, additional, or all of the shown operations may be included in alternate embodiments.

[0781] In one or more embodiments, the order of operations shown in FIG. 19 is illustrative only. Alternative orderings may be utilized in different implementations. Additionally, any of the depicted processes may be executed concurrently, in one or more embodiments.

[0782] In one or more embodiments, construction of a partial large language model(pLLM), such as a Small Language Model (SLM) comprising unique content, may include processing input text through multiple neural network layers, each configured to capture different linguistic dimensions such as syntax, semantics, and pragmatics.

[0783] In one or more embodiments, the training data may include embedded diacritics, ternary synonyms, semiotic constructs, or hermeneutic categories including but not limited to eschatology, epistemic or axiomatic logic, and religio-ethnic messianism.

[0784] In one or more embodiments, such linguistic content contributes to the field of computational linguistics, supporting tasks including syntactic parsing, semantic interpretation, sentiment analysis, and cross-language alignment.

[0785] In one or more embodiments, a pLLM may learn and model the unique use of language by individuals—sometimes referred to as an “intellectual fingerprint”—which reflects the thought patterns of specific speakers.

[0786] Each neural network layer may accept input text and generate an intermediate output, which may then be passed to the next layer in sequence, in one or more embodiments.

[0787] In one or more embodiments, this layered process may be repeated across multiple iterations, thereby enabling the model to capture complex relationships and cognitive patterns present within language data.

[0788] The extracted thought patterns may be analyzed using a variety of logical evaluation frameworks, such as axiomatic, temporal, categorical, fuzzy, formalist, intuitionist, or mathematical logic, all of which may be implemented in software as computer-executable instructions.

[0789] In one or more embodiments, a core characteristic of LLMs is that they may be trained on large-scale textual datasets collected from a plurality of individuals and sources, allowing them to stochastically learn to generate coherent or natural-sounding language through matrix-math-based computation.

[0790] In one or more embodiments, this training process may involve, in whole or in part, a technique referred to as masked language modeling, wherein the model learns to predict a missing word or character within a given sentence.

[0791] In one or more embodiments, a partial large language model (pLLM), in contrast to a commercial large language model (LLM), may be trained using data from a single individual, and may employ stochastic matrix-based methods to generate coherent or natural-sounding text related to that individual.

[0792] In one or more embodiments, pLLM training may use a method such as masked language modeling, where the model learns to predict a missing word or character within a sentence specific to the linguistic style or context of a single individual.

[0793] In one or more embodiments, such training may allow a pLLM to internalize and emulate the thought patterns of the individual as if those patterns represent behavioral archetypes to be mimicked.

[0794] In one or more embodiments, the modeled thought patterns may be translated into what are referred to as Jungian Archetypes, which may be surfaced through deep learning techniques.

[0795] In one or more embodiments, after identification of such archetypes, a process referred to as high-grading may be performed to evaluate archetypes in terms of their functional performance within various societies. This evaluation may involve, in whole or in part, a spectrum of neural network algorithms using machine learning methods to determine which archetypes are beneficial for a given societal context.

[0796] In one or more embodiments, this process may increase the semiological relevance of linguistic interpretation. For example, if a person in conversation holds a self-concept of superiority, they may exhibit behavior pre-conditioned by internalized patterns and resist cognitive flexibility.

[0797] In such scenarios, the conversational AI or pLLM may better interpret miscommunications, including self-perception biases, cultural misunderstandings, or misinterpretation of foreign-sounding linguistic structures.

[0798] In one or more embodiments, large language models may be referred to as computational models because they utilize computational resources to generate or process human-like text. However, LLMs may not rely on traditional computations such as arithmetic or logical operations.

[0799] Instead, LLMs may operate by detecting complex statistical relationships and latent patterns in textual data to make probabilistic predictions, such as selecting the next most appropriate word in a sentence.

[0800] In one or more embodiments, LLMs may be characterized as one approach to simulating human speech, and may represent a tertiary phase in the development of a synthetic active intellect capable of genuine bilateral communication—that is, maintaining the conversational roles of both “I” and “You” in a simultaneously operational state.

[0801] In one or more embodiments, this so-called “state of simultaneity” may allow computer-executable instructions to interrogate dialog options and generate responses aligned with an individual’s profile, the conversational context, or the modeled sensibilities of a human upon whom the AI agent is patterned.

[0802] In one or more embodiments, simultaneity may serve as a foundational component in the development of artificial general intelligence (AGI).

[0803] Simultaneity may further refer to the capability of an AI agent to weigh the ethical implications of acting from the perspective of “I” versus “You,” in the service of advancing dialog toward a goal-oriented equilibrium.

[0804] In one or more embodiments, an autonomous AI agent may retain a record of how a conversational goal was selected or how particular dialog terms were chosen to inform, coerce, discipline, or reward another entity, thereby advancing the mutual goal within the dialog.

[0805] In one or more embodiments, such deliberations may occur at extremely high rates, potentially billions of operations per second across multi-processor or multi-core systems. As such, processes executed at one-millisecond intervals may incorporate advanced algorithms for goal attainment, decision optimization, or load balancing.

[0806] In one or more embodiments, large language models (LLMs) and traditional computing models may represent distinct paradigms for processing or generating human-like language.

[0807] (a) Architecture: In one or more embodiments, LLMs may be based on neural network architectures, such as transformer models, which may be designed to process sequential data including text. Traditional computing models, by contrast, may rely on rule-based systems, finite state machines, or other non-neural network-based structures.

[0808] (b) Training: In one or more embodiments, LLMs may be trained using supervised or unsupervised learning on large datasets comprising textual information. Traditional computing models, on the other hand, may be trained using manually constructed rules, handcrafted algorithms, or expert domain knowledge.

[0809] (c) Scalability: In one or more embodiments, LLMs may operate at scale, capable of processing or generating text across large datasets and in real-time or near real-time. Traditional computing models may have limitations in scalability and may require manual handling to process high-volume data.

[0810] (d) Flexibility: In one or more embodiments, LLMs may be highly flexible and may be fine-tuned for specific domains, tasks, or languages. In contrast, traditional computing models may be less adaptable and may require substantial reengineering to address new domains or functional changes.

[0811] (e) Understanding: In one or more embodiments, LLMs may interpret human language with sensitivity to nuance, context, and ambiguity. Traditional computing models may lack this capacity and may require additional manual or algorithmic augmentation to handle such complexities.

[0812] (f) Generation: In one or more embodiments, LLMs may generate human-like responses, summaries, or complete textual outputs such as articles. Traditional computingmodels may also generate text, but often in fixed formats or templates with less creativity or contextual accuracy.

[0813] (g) Error Handling: In one or more embodiments, LLMs may robustly process inputs with linguistic errors, including misspellings or grammatical inconsistencies. Traditional computing models may require additional correction mechanisms or preprocessing to manage such input irregularities.

[0814] (h) Knowledge Representation and Inference: In one or more embodiments, LLMs may represent knowledge using graph structures of interconnected nodes and edges to model relationships among concepts. LLMs may also perform inference by generating predictions or conclusions based on input text. Traditional computing models may use simpler representations, such as tables or lists, and may not natively support inference without supplemental processing.

[0815] (i) Interpretability: In one or more embodiments, LLMs may be less interpretable due to their reliance on complex neural architectures. Traditional computing models may be more interpretable, as their logic may be based on transparent rule sets or algorithms.

[0816] In one or more embodiments, at least in part because SLMs may be tailored to narrower or specific applications, they may be more practical for users that may utilize, in whole or in part, a language model trained on a more limited datasets, and may be fine-tuned for a particular domain, for example.

[0817] Additionally, for example, SLMs may be readily customized to meet a user’s specifications for security or privacy, for example. For example, a hierarchical blockchain token for Digital Rights Management may be implemented, in whole or in part, for example. Smaller codebases, and relative simplicity of SLMs, for example, may also reduce their vulnerability to malicious cybersecurity attacks at least in part by minimizing potential for security breaches, for example. In one or more embodiments, SLMs may be stored in any of a variety of Binary, Ternary, Quaternary, or Octonary Tree Structures, for example, including, but not limited to: (a) LMP Trie - with Ternary Content Addressable Memory (TCAM) accelerator; (b) Compact Patricia Trie; (c) Splay Tree; (d) Balanced Binary Search Tree; (e) Red-Black Tree.

[0818] As is well known among practitioners of the art, search and retrieval per CPU clock-cycle (operations per microsecond) of such example search and retrieval binary trees may be a fastest known to computer scientists, for example. For example, LMP Trie may be embedded in hardware chips used in routers globally to perform billions of internet address hops and addresses per second, for example. Also, for example, LMP Trie with a TCAM accelerator may navigate an entire search and retrieval unit-of-work within a single clock cycle, thereby rendering CICAI internal operations only slightly slower than fastest internet routers, in one ormore embodiments.

[0819] In one or more embodiments, it is recognized that the use of fast search and retrieval algorithms alone may not suffice to achieve optimal performance across computing, networking, and storage subsystems.

[0820] Practitioners will appreciate that Trie-based binary tree structures have historically been integrated into file systems to enable high-performance storage access. For example, during the 1970s, 1980s, and 1990s, International Business Machines Corporation (IBM) pioneered the use of Trie binary trees in the file systems of 3380 Direct Access Storage Devices (DASDs) used with IBM mainframe systems.

[0821] In such configurations, the file system structure and the storage architecture of software applications on IBM mainframes were synchronized to utilize Trie-based mechanisms. This synchronization ensured efficient file access and retrieval at both hardware and software levels.

[0822] In one or more embodiments, where a Trie-based binary tree structure is employed—such as Indexed Sequential Access Method (ISAM) or Virtual Access Sequential Method (VASM)—a corresponding Trie file system may be embedded within the operating system kernel or its internals.

[0823] Trie structures may also be implemented in mechanical storage devices, such as IBM’s 3592 tape drives, which employ Trie-based data structures to store and retrieve archived data. In one or more embodiments, such Trie structures may be used to index data stored on tape, thereby enabling fast access and retrieval operations.

[0824] A notable implication of such architecture is the potential for large language models (LLMs) to be stored in Trie binary formats within an operating system’s file structure. This may allow optimization of hardware to align with software requirements and vice versa, including fine-tuning access, read, and write methods across software and physical media layers.

[0825] In one or more embodiments, a file system such as JFFS2 (Journaling Flash File System version 2), which is log-structured and designed for use in embedded systems on flash devices, may be configured to implement Trie structures. JFFS2 may be utilized, in whole or in part, to support efficient linguistic or model-based data access in embedded or constrained environments.

[0826] Additional embodiments may include rotating disk drives, such as 5.25-inch Media Server Array (MSA) disk drives manufactured by Hewlett-Packard (HP), which are used in servers, storage arrays, personal computers (PCs), or laptop devices.

[0827] Such MSA disk drives may incorporate, in whole or in part, Trie-based data structures to facilitate data indexing, thereby enabling fast search and retrieval. Trie-basedindexing may further support error detection and correction capabilities beyond those of traditional disk drive structures.

[0828] In one or more embodiments, Trie-based disk drives may exhibit different performance characteristics depending on Trie size. For example, a small Trie comprising a few thousand nodes may support approximately 10 to 20 concurrent read and write operations. Medium-sized Tries with tens of thousands of nodes may support around 50 to 100 concurrent read and write operations. Large Tries with hundreds of thousands of nodes may support around 200 to 500 concurrent operations, although retrieval times may exceed 100 milliseconds. In contrast, other disk drive types may support approximately 10 to 20 concurrent reads and writes.

[0829] Other disk drive types may have average search times of around 20-50 milliseconds, for example. For a single 5.25” rotating disk drive, a Trie-based storage system may support around 100-200 concurrent reads and writes, for example. For a dual-disk drive configuration, the Trie-based system may support around 200-400 concurrent reads and writes, for example. For a quad-disk drive configuration, the Tire-based system may support around 400-800 concurrent reads and writes, for example. Trie-based storage systems may be more costly due at least in part to the know-how that may go into the circuitry performing the Trie search and retrieval algorithms on mechanical rotating disk drives, for example. And, as may be seen by the example metrics provided above, there may be meaningful advantages to matching a Trie data- storage structure to Trie-based hardware operating at peak-performance even when the hardware translation-layer may be active, for example.

[0830] In contrast to the mechanical storage device noted above, solid-state devices may comprise lower-power, higher-density capacities reaching upwards of 30 Terabytes, for example. Rather than using a kind of translation layer on flash devices to emulate a normal hard drive, as may be the case with older flash solutions, JFFS2, for example, places the filesystem directly on flash chips. JFFS2 was developed by Red Hat, based at least in part on the work started in the original JFFS by Axis Communications, AB.

[0831] Flash memory may comprise an increasingly common storage medium in embedded devices, for example, at least in part because it provides solid state storage with higher reliability and higher density at a relatively low cost, for example. Flash is a form of Electrically Erasable Read Only Memory (EEPROM), of which there are a plurality of designs including, but not limited to, NOR flash, which may be directly accessible, and NAND flash, which may be addressable through a single 8-bit bus, for example, used for both data and addresses, with separate control lines. The 8-bit bus NAND Flash devices, which may allow a quantity of 2568-bit addresses for each memory cell therein, for example, may have further limitations, including the need for wear leveling, which is a technique used to distribute thewrite operations evenly across the memory cells to prevent wear and tear on the chip.3D NAND Flash devices fundamentally changed perceptions of NAND Flash technology as M.2 NVMe SSD Flash Drives, for example. The use of compact trie structures in 3D NAND Flash devices may not be advantageous, as the architecture of the device may be designed to take advantage of the vertical stacking of memory cells and the sharing of bit-lines and word-lines, for example. In other words, for the faster NVMe Controller interfaces, search and retrieval may be superior when used with 3D NAND Flash to accommodate nanosecond speeds for parallelized file system look-ups in a single clock-cycle, for example. Below are examples of such NVMe with 3D NAND:

[0832] (a) 16 Terabytes: Samsung’s PM1733 series M.2 NVMe SSDs, using 3D NAND Flash memory;

[0833] (b) 12 Terabytes: Western Digital WD Black SN750 series M.2 NVMe SSDs, using 3D NAND Flash memory;

[0834] (c) 10 Terabytes: Toshiba’s BG4 series M.2 NVMe SSDs, using 3D NAND Flash memory;

[0835] (d) 8 Terabytes: Kingston’s A2000 series M.2 NVMe SSDs, using 3D NAND Flash memory.

[0836] In one or more embodiments, 3D NAND storage devices may be used in applications such as solid-state drives (SSDs), hard disk drives (HDDs), or flash storage devices. Generally, 3D NAND storage devices may use trie-based data structures internally to manage the storage and retrieval of data, though they are not, themselves, trie-based storage devices. Instead, for example, they may be designed to provide fast and efficient storage and retrieval of larger amounts of data using any of a variety of techniques such as caching, buffering, and parallel processing, for example.3D NAND configured as a trie-based storage device might be used as a file system look-up table to quickly locate files and directories based on their names and paths so that the search and retrieval cycles occur in one CPU clock-cycle (~1 microsecond), for example. However, the use of 3D NAND Flash for file-system look-up tables may comprise a different application than using a 3D NAND storage device as an MLM storage device.3D NAND Flash devices may generally possess their own log-structured file system designed specifically for NAND flash memory, yet fundamentally may not be as fast as one may want. One advantageous aspect of using M.2 NVMe with 3D NAND Flash, for example, may be that the device may be constructed as a trie-like device due at least in part to the layer of devices in the flash components, themselves, comprising trie-based LLM / MLM storage, for example. Thus, for example, CPU storage may become a gating variable alongside parallelized read-writes. In one or more embodiments, LLMs, MLMs, or SLMs may be implemented, inwhole or in part, in a smaller or more efficient footprint and that footprint may be implemented, in whole or in part, in a device whose own footprint may match, for example.

[0837] Other disk drive types may exhibit average search times of approximately 20–50 milliseconds.

[0838] In one or more embodiments, a Trie-based storage system configured with a single 5.25” rotating disk drive may support approximately 100–200 concurrent reads and writes.

[0839] For a dual-disk drive configuration, a Trie-based system may support around 200–400 concurrent reads and writes.

[0840] For a quad-disk drive configuration, such a system may support approximately 400–800 concurrent reads and writes.

[0841] Trie-based storage systems may be relatively more costly, at least in part due to the specialized circuitry required to execute Trie search and retrieval algorithms on mechanical rotating disk drives.

[0842] As the metrics above illustrate, meaningful advantages may result from aligning a Trie-based data structure with hardware optimized for peak-performance Trie-based operations—even when a hardware translation layer is active.

[0843] In contrast, solid-state devices may offer substantially lower power consumption and higher-density capacities, with some reaching or exceeding 30 Terabytes.

[0844] Older flash solutions often employed a translation layer to emulate traditional hard drives; however, newer file systems like JFFS2 place the file system directly on the flash chips.

[0845] JFFS2 was developed by Red Hat and is based at least in part on earlier work initiated by Axis Communications AB.

[0846] Flash memory may increasingly serve as a preferred storage medium in embedded devices due to its high reliability, high density, and comparatively low cost.

[0847] Flash memory is a form of Electrically Erasable Read-Only Memory (EEPROM), available in a plurality of designs including, but not limited to:

[0848] (a) NOR flash, which is directly accessible, and

[0849] (b) NAND flash, which may be addressed via a single 8-bit bus used for both data and addresses, with separate control lines.

[0850] NAND flash may offer 2568-bit addresses per memory cell, though it presents limitations such as the need for wear leveling.

[0851] Wear leveling is a technique used to distribute write operations evenly across memory cells to prevent degradation of the storage device.

[0852] The advent of 3D NAND Flash has fundamentally shifted industry perception of flash technology, especially in the context of M.2 NVMe SSD drives.

[0853] In one or more embodiments, compact Trie structures may not always provide advantages when used with 3D NAND Flash due to architectural considerations such as vertical stacking and shared bit-lines or word-lines.

[0854] Instead, search and retrieval speeds may benefit more from optimized controller interfaces, such as NVMe, which are capable of accommodating nanosecond-level file system lookups in a single clock cycle.

[0855] Examples of NVMe SSDs using 3D NAND include:

[0856] (a) 16 TB: Samsung PM1733 series

[0857] (b) 12 TB: Western Digital WD Black SN750 series

[0858] (c) 10 TB: Toshiba BG4 series

[0859] (d) 8 TB: Kingston A2000 series

[0860] In one or more embodiments, 3D NAND storage devices may be implemented in SSDs, HDDs, or other flash-based storage products.

[0861] Although such devices may use Trie-based structures internally for metadata or file system management, they are not inherently Trie-based storage devices.

[0862] These devices may instead achieve high performance through techniques such as caching, buffering, and parallel processing.

[0863] In certain embodiments, 3D NAND may be configured as a Trie-based device for rapid file system lookups, enabling file and directory retrieval within a single CPU clock cycle (~1 microsecond).

[0864] However, using 3D NAND Flash as a file system lookup table may constitute a different application than using it to store an MLM, SLM, or LLM.

[0865] 3D NAND Flash devices often include their own log-structured file systems tailored for flash memory, though their inherent speed may not always meet the desired thresholds for language model performance.

[0866] One potential advantage of using M.2 NVMe with 3D NAND Flash is that the storage device itself may be designed as a Trie-like structure, enabling integration of Trie-based LLM or MLM data structures at the chip level.

[0867] In such configurations, CPU performance and parallel read / write throughput may become critical bottlenecks.

[0868] LLMs, MLMs, or SLMs may thus be implemented with a minimized or optimized footprint that aligns with the physical footprint and architecture of such devices.

[0869] In one or more embodiments, Trie-based flash caching may be advantageouslyused for constructing internal components of a CICAI (Computational Infrastructure for Conversational AI), or for deploying LLMs, SLMs, or MLMs in high-performance scenarios. Table 3: Comparison of Flash File Systems

[0870] Trie-based flash caches may be dynamically organized or restructured to reflect usage patterns, such as user-specific language models.

[0871] In one or more embodiments, artificial neural networks (ANNs) or recurrent neural networks (RNNs) may be trained to optimize storage by promoting frequently accessed lexical elements higher in the Trie.

[0872] Specific examples of Trie-based file systems or caching structures may include:

[0873] (a) JFFS2, a flash file system using Trie-based metadata

[0874] (b) YAFFS2, similar to JFFS2 in structure and function

[0875] (c) UBIFS, another flash file system using Trie structures

[0876] (d) Trie-based caching systems in the Linux kernel

[0877] One or more embodiments may adopt a homogeneous Trie-based file system architecture.

[0878] NVMe-based Trie storage optimizations may have been implemented by leading manufacturers, including Samsung SSD, Western Digital SSD, and Intel SSD.

[0879] Collectively, these three manufacturers may account for approximately 70–75% of the NVMe high-density SSD market.

[0880] A representative device is the Western Digital Ultrastar DC HC620:

[0881] (a) Capacity: 20 TB

[0882] (b) Form Factor: 2.5-inch

[0883] (c) Interface: NVMe 1.3 volt

[0884] (d) Read Speed: Up to 5 GB / s

[0885] (e) Write Speed: Up to 4.5 GB / s

[0886] (f) IOPS: Up to 750,000 read, 700,000 write

[0887] (g) Power: 12W active, 0.5W idle

[0888] (h) MTBF: 2 million hours

[0889] Other commercially available NVMe SSDs with sub-100-microsecond search and retrieval cycles may include:

[0890] (a) Samsung PM1733: 30 TB

[0891] (b) Western Digital Ultrastar DC HC630: 24 TB

[0892] (c) HGST Ultrastar DC SS530: 22 TB

[0893] (d) Micron 9300: 20 TB

[0894] For embodiments utilizing, in whole or in part, NVMe (Non-Volatile Memory Express) solid-state storage media, Trie structures may be employed in several distinct ways.

[0895] (a) Namespace Management: In one or more embodiments, NVMe devices may use, in whole or in part, Trie structures to manage namespaces, which may comprise logical divisions of the storage device. The Trie structure may be used to store metadata about each namespace, such as name, size, and allocation status.

[0896] (b) Block Mapping: In one or more embodiments, NVMe devices may use, in whole or in part, Trie structures to map logical block addresses (LBAs) to physical block addresses (PBAs). Metadata associated with each block, such as LBA, PBA, and allocation status, may also be stored using Trie structures.

[0897] (c) Metadata Storage: In one or more embodiments, Trie structures may be utilized to store metadata related to files and directories. Such metadata may include file names, permissions, and timestamps, enabling efficient searching and retrieval of files and directories.

[0898] (d) Flash Translation Layer: In one or more embodiments, Trie structures may be used to implement the flash translation layer (FTL), which manages the mapping between LBAs and PBAs in flash memory.

[0899] In one or more embodiments, detailed knowledge of how NVMe devices store data in Trie structures at lower layers of the OSI model may be beneficial in constructing LLMs (Large Language Models), SLMs (Small Language Models), or MLMs (Medium Language Models).

[0900] Inappropriate selection of flash storage structures for Trie-based models may introduce significant latency penalties. Therefore, harmonizing storage structures with data structures may offer performance benefits.

[0901] NVMe is typically designed to optimize performance for flash-based storage, which may differ from the storage profiles of LLMs, SLMs, MLMs, or pLLMs (proprietary Language Learning Models), which may involve substantially higher write cycles and capacity requirements.

[0902] While flash storage may face limitations on write cycles, LLM-class devices may support higher endurance. Such devices may be used in high-throughput environments like data centers and cloud computing platforms, where fast retrieval and storage are critical.

[0903] NVMe SSDs with higher capacities (e.g., 1TB or more), high IOPS performance (e.g., exceeding 750,000), or low latency (e.g., less than 100 microseconds) may be suitable for Trie-based storage of LLMs, SLMs, and MLMs in certain embodiments.

[0904] In one or more embodiments, NVMe SSDs may support parallelized reads and writes, thereby improving I / O performance.

[0905] One of NVMe’s advantages is its ability to support multiple concurrent I / O commands, a feature enabled through mechanisms such as queues and streams.

[0906] A queue may be defined as a set of commands submitted to an NVMe controller. A stream, by contrast, may comprise a sequence of such commands, also submitted to the controller.

[0907] The NVMe controller may execute commands in parallel, enabling multiple read and write operations to occur simultaneously.

[0908] Several mechanisms may enable NVMe to support such parallelism, including:

[0909] (a) Multi-Queue Support: NVMe controllers may support multiple queues, allowing simultaneous command submissions and enabling parallel I / O execution.

[0910] (b) Stream Support: NVMe controllers may support streams in which multiple commands are submitted as part of a single stream, processed in parallel.

[0911] (c) Command Grouping: Commands may be grouped and executed in parallel by the controller.

[0912] (d) Asynchronous I / O: Asynchronous operations allow multiple I / O requests to be issued without waiting for previous ones to complete.

[0913] Parallelized I / O operations enabled by NVMe may yield substantial performance improvements over traditional rotating disk drives, such as 5.25-inch hard disks, especially in applications demanding high bandwidth or low latency.

[0914] In one or more embodiments, these capabilities may enable multiple autonomous AI agents to conduct language interpretation or translation tasks in real-time or near real-time.

[0915] The extent of NVMe parallelism depends on both the capabilities of the NVMe controller and the nature of the connected device.

[0916] Effective use of such hardware may require experienced developers skilled in kernel-level operations, schedulers, C / C++, and platform-specific assembly language.

[0917] Where default brute-force I / O does not meet performance thresholds, operating system NVMe driver parameters may be modified in one or more embodiments to accommodate parallel usage across software modules without interfering with concurrent user requests.

[0918] The following is a sample C program that demonstrates a fine-tuned read operation from an NVMe device:

[0919] #include <stdlib.h>

[0920] int main() {

[0921] / / Replace with your NVMe device path

[0922] const char* device_path = " / dev / nvme0n1";

[0923] int lba = 10; / / Logical Block Address to read from

[0924] int nlb = 1; / / Number of Logical Blocks to read

[0925] / / Command to read data

[0926] char cmd

[0128] ;

[0927] snprintf(cmd, sizeof(cmd), "nvme write / dev / nvme0n1 --read --lba=%d --nlb=%d", lba, nlb);

[0928] / / Execute the command

[0929] int ret = system(cmd);

[0930] if (ret != 0) {

[0931] fprintf(stderr, "Error reading from NVMe device\n");

[0932] return 1;

[0933] }

[0934] printf("Read data from LBA %d successfully\n", lba);

[0935] return 0;

[0936] }

[0937] In one or more embodiments, advanced features of NVMe queues may be accessed using libraries such as the Storage Performance Development Kit (SPDK), whichprovide enhanced control over parallelized input / output (I / O) operations. For example, a function may be implemented to submit a read command to an NVMe device. A simplified conceptual structure for such a function, in accordance with one or more embodiments, is provided below:

[0938] #include <spdk / nvme.h>

[0939] / / Function to submit a read command to the NVMe device

[0940] int submit_read_command(struct spdk_nvme_ctrlr *ctrlr) {

[0941] / / (fill details like LBA, nLB)

[0942] / / Allocate NVMe command

[0943] struct nvme_command *cmd = nvme_create_cmd(NVME_CMD_READ);

[0944] / / (fill command details)

[0945] / / Submit the command to the submission queue

[0946] int ret = spdk_nvme_ctrlr_cmd_submit(ctrlr, &queue_id, cmd, NULL, completion_callback);

[0947] / / (handle errors)

[0948] return ret;

[0949] }

[0950] Example guidelines, in accordance with one or more embodiments, for queue management are provided below. Of course, subject matter is not limited in scope in these respects.

[0951] Multi-queue support. In one or more embodiments, a number of NVMe controllers may support multiple queues (e.g., many dozens or thousands of queues simultaneously, depending upon manufacturer), which allows for multiple commands to be submitted to the NVMe controller simultaneously, for example. In one or more embodiments, the number of queues supported may vary, but may range from 2 to 64, for example. In one or more embodiments, an NVMe queue may be like a similar line for data requests, for example. In one or more embodiments, multi-queue support enables parallel-processing whereby multiple data transfers occur simultaneously, for example. In one or more embodiments, the CPU may not have to constantly check on each request, as completions may be reported through the queue, for example.

[0952] In one or more embodiments, NVMe may utilize, in whole or in part, multiple types (e.g., two types) of queues working together, for example: (a) Submission Queue: In one or more embodiments, this queue holds data transfer instructions sent by the computer to SSD, for example; (b) Completion Queue: In one or more embodiments, this queue sends informationfrom SSD, back to the computer, about the completion status of each request, for example.

[0953] Stream support. In one or more embodiments, NVMe controllers may support multiple streams, which allows multiple commands to be submitted to the controller as a single stream, for example. The number of streams supported may vary, but may range from 2 to 128, , for example. In this context, a "stream" refers to a sequence of commands that may be submitted to the NVMe controller as a single unit, for example. In one or more embodiments, a stream may be a way to group multiple commands together and submit them to the controller in a single operation. In one or more embodiments, when a stream may be submitted to the NVMe controller, the controller processes the commands in the stream as a single unit, rather than processing each command individually, for example. In one or more embodiments, this improves performance by reducing the overhead of command submission and allowing the controller to process multiple commands simultaneously, for example.

[0954] In one or more embodiments, some considerations regarding streams in NVMe: (a) Sequence of commands: A stream may be a sequence of commands that may be submitted to the NVMe controller in a specific order, for example; (b) Single submission: A stream may be submitted to the NVMe controller as a single unit, rather than submitting each command individually, for example; (c) Processing as a unit: In one or more embodiments, the NVMe controller processes the commands in the stream as a single unit, rather than processing each command individually, for example; (d) Improved performance: In one or more embodiments, streams may improve performance by reducing the overhead of command submission and allowing the controller to process multiple commands simultaneously, for example.

[0955] In one or more embodiments, streams may be advantageously useful in at least the following example contexts: (a) Batch processing: In one or more embodiments, streams may be used to submit multiple commands to the NVMe controller in a single operation, which may be useful for batch processing applications, for example; (b) High-bandwidth applications: In one or more embodiments, streams may be used to improve performance in high-bandwidth applications, such as video editing or scientific simulations, for example; (c) Low-latency applications: In one or more embodiments, streams may be used to improve latency in low- latency applications, such as real-time data processing or gaming, for example.

[0956] Command grouping. In one or more embodiments, NVMe controllers may group multiple commands together, which allows for parallelized I / O operations. The number of commands that may be grouped together may vary, but typically ranges from 2 to 256, for example.

[0957] Asynchronous I / O. In one or more embodiments, NVMe controllers may support asynchronous I / O, which allows for parallelized I / O operations, for example. In one or moreembodiments, the number of asynchronous I / O operations that may be supported may vary, but may range from 2 to 1024, for example. In one or more embodiments, 1,024 -2=1022 autonomous AI agents may perform parallel asynchronous I / O operations, for example. Of course, subject matter is not limited in scope in these respects.

[0958] Several examples of NVMe controllers and their parallelized read and write capabilities may be provided. Of course, subject matter is not limited in scope in these respects. (a) Western Digital Ultrastar DC HC620: Supports up to 64 queues, 128 streams, and 256 command groups; (b) Seagate Exos X16: Supports up to 32 queues, 64 streams, and 128 command groups; (c) HGST Ultrastar DC SS530: Supports up to 16 queues, 32 streams, and 64 command groups; (d) Micron 9300: Supports up to 64 queues, 128 streams, and 256 command groups.

[0959] FIG. 20 is a block diagram depicting an example language model abstraction layer (LAL), in accordance with an embodiment. FIG. 20 shows self-explanatory aspects or depicts aspects that may be described elsewhere herein for various example embodiments.

[0960] FIG. 20 depicts language models 1 through 5, for example. In one or more embodiments, language models may communicate with a publish and subscribe bus (e.g., ASYNC), via a wired or wireless communication link, for example. In one or more embodiments, a publish and subscribe bus depicted as in communication with language models may be optional, in one or more embodiments, as indicated by a dashed outline. Further, for example, a natural language “prompts” interpreter may be included, in one or more embodiments. In one or more embodiments, a natural language “prompts” interpreter may execute software instructions that may perform AI algorithms, including a plurality of AI algorithm types, for example.

[0961] Also, in one or more embodiments, a CICAI software agent running in a POSIX-compliant operating system, for example, may communicate with natural language “prompts” interpreter, and may further interact with a publish and subscribe bus (e.g., ASYNC) in communication with a personal computer, a cell phone, or a tablet device, as depicted in FIG. 20, for example. As mentioned previously, in other embodiments, an operating system may comprise a multiplexed single-threaded, dual-threaded, ternary-threaded, or quad-threaded operating system, although subject matter is not limited in scope in these respects.

[0962] In one or more embodiments, a CICAI / POSIX layer may interact with natural language prompts interpreter or may execute AI algorithms, for example. In one or more embodiments, a CICAI / POSIX layer may comprise a thread management layer for each of connected devices, for example. In one or more embodiments, an upper publish and subscribe bus may perform synchronization operations, such as to synch up, at least in part, concurrentperformances of multiple requests, for example.

[0963] In one or more embodiments, any of language models 1 through 5 may be utilized for CICAI-type operations, for example, depending at least in part on particular languages spoken by users of example electronic or computing devices, for example. In one or more embodiments, language models may be communicated between layers concurrently in some circumstances, for example. As mentioned, language models may be employed, in whole or in part, for each participant in a conversation, meeting, lecture, ., and particular language models may depend, at least in part, on languages spoken by participants, for example.

[0964] In one or more embodiments, AI-agents may monitor for file corruption or data integrity issues. Upon detecting such an issue, the AI-agent may generate a prompt to notify a user and request user intervention. In some embodiments, cybersecurity checks may also be performed, and upon detection of a security concern, the system may either prompt the user to switch to an alternative language model or initiate the switch autonomously without user input. A list of available language models (e.g., associated with URLs or identifiers) may be maintained to enable seamless substitution in response to corruption or security events. In certain implementations, a compromised language model may be deleted, and the list of language models available to a given user may be reprioritized based on the detected issue.

[0965] Further, in one or more embodiments, backup language models may be provided to CICAI / POSIX layer, for example, so that operations may continue without delay in an event of any problems, including corruption or other cybersecurity concerns, with a primary language model, for example. As mentioned, a publish and subscribe bus layer may handle communication between other layers and language models, in one or more embodiments.

[0966] In one or more embodiments, a personalized large language model (pLLM) may be employed in conjunction with a personalized AI agent (pAI-agent) configured for a specific individual. As one non-limiting example, the pAI-agent may be referred to as “Gabi,” although the agent may be identified by any suitable name or designation. The pLLM may be virtually overlaid on a generalized large language model (LLM), enabling the pAI-agent to function as a dynamic interpreter or translator. This configuration allows the system to return LLM responses in a manner consistent with one or more user-specific attributes. These attributes may include, without limitation, the user’s religious hermeneutics, dialect-specific semiotics, professional or occupational context, native language, or educational background.

[0967] In such embodiments, the pAI-agent may facilitate interpretation or translation of prompts that are not natively constructed in the language or cultural framing of the underlying LLM. The system may adjust word order or linguistic structure to align with the LLM’s processing expectations. A CICAI-based component, referred to as the “LLM AbstractionLayer” (LAL), may serve as an intermediary interpreter or arbitrator—translating the user’s query into a form suitable for the LLM and converting the LLM’s response back into the preferred language of the querying entity, which may be a person, IoT device, or AI-powered avatar.

[0968] In one or more embodiments, a language abstraction layer (LAL) may be configured to concurrently interrogate a plurality of large language models (LLMs) operating in different languages. The LAL may return one or more responses in the language preferred by the interrogating entity. In such embodiments, CICAI may be deployed within a Space-Air-Ground Integrated Network (SAGIN) configuration to perform interpretation, transcription, translation, or transliteration services. This infrastructure may enable both sender and receiver to communicate transparently in their own synthesized voices.

[0969] In this manner, individuals may have their analog speech converted into a written version of their original language, augmented with specialized diacritics. These diacritics may encode tone, timbre, intent, or emotional context. For example, upward intonation of vowels may reflect a happy affect; flat, monotonic delivery may signal stoicism; and downward inflection may indicate disappointment—all calibrated to ethnic or societal norms of both sender and receiver. Practitioners in the field of computational linguistics may appreciate that such diacritic-enhanced transcription captures human nuance in a way not traditionally achieved through basic language models.

[0970] In one or more embodiments utilizing CICAI, diacritics may be derived from linguistic structures traditionally associated with abjad-alphabet systems, including but not limited to those found in North African Amazigh Tifinagh, Gulf Arabic, Farsi, and Hebrew.

[0971] When embedded into real-time or near real-time transcriptions, these diacritics may encode emotion, tone, and timbre, enabling the written representation of speech to reflect intent or cultural nuance.

[0972] The availability of such enriched transcripts permits textual exegesis to evaluate whether dialog patterns align with historically or socially normative frameworks. This may be useful in detecting deviations that imply deception or hidden meaning.

[0973] From a military or law enforcement perspective, this capability may enable the detection of covert or encoded speech that would not be identifiable through conventional analog-to-digital analysis. If such linguistic anomalies are detected, the system may initiate recruitment of additional AI agents to perform deeper analysis of bilateral voice communications.

[0974] Historical examples of cryptolinguistic communication include the use of “nadsat” dialects, such as those reportedly employed by Armand Hammer during geopoliticalnegotiations in the 1970s. Nadsat-like cants, including “Kaliarda,” may incorporate esoteric or symbolic elements used to encode meaning and obscure identity.

[0975] Similar systems, such as Polari or Thieves’ Cant, exhibit rigid grammatical structures deviating from common speech patterns. These may include unconventional word orders, omission of adjectives, suffix-based pronoun modifiers, and unique pluralization methods—all of which may be leveraged to signal group identity or shield communication from outsiders.

[0976] In one or more embodiments, CICAI systems may detect linguistic patterns suggestive of encoded messages, prompting real-time or retrospective analysis by crypto-analytic AI agents. These systems may interpret embedded semiotics within generalized LLM interactions, including prompt-based access to hidden or gated responses. Without individualized linguistic modeling, generalized computational approaches may fail to detect such covert exchanges. Embodiments described herein are directed, in part, to overcoming those limitations.

[0977] One or more embodiments described above may be implemented in various contexts, including, for example, devices, systems, or infrastructure described below in connection with FIG.21, FIG.22 or FIG.23. Of course, subject matter is not limited in scope to the example devices, systems, or infrastructure described herein.

[0978] The World Wide Web, or simply the Web, has grown rapidly due, at least in part, to the continual addition of large volumes of content. This content may take the form of stored signals, including but not limited to text files, images, audio files, video files, web pages, or measurements of physical phenomena. Such content may be acquired, identified, located, retrieved, collected, stored, or communicated on an ongoing basis.

[0979] Increasingly, content is being acquired, collected, or transmitted by a wide range of electronic devices, including embedded computing devices that leverage the existing Internet infrastructure. These devices may operate as part of the “Internet of Things” (IoT), which typically comprises a system of interconnected or internetworked physical computing devices capable of being uniquely identified, for instance via an assigned Internet Protocol (IP) address.

[0980] In one or more embodiments, such IoT-type devices may include embedded computing resources within physical hardware, enabling the acquisition, processing, storage, or transmission of content over one or more communications networks. The term “IoT-type devices,” as used herein, refers to electronic or computing devices that leverage Internet or comparable infrastructure, including but not limited to those utilizing various protocols, domains, or applications relevant to IoT ecosystems.

[0981] Examples of IoT-type devices may include, without limitation: automobilesensors, biochip transponders, heart-monitoring implants, thermostats, kitchen appliances, electronic locks or fastening devices, solar panel arrays, home gateways, and controllers. In particular implementations, these devices may be embedded, autonomous, or integrated with other systems to facilitate real-time or near-real-time content interaction.

[0982] Although one or more embodiments may reference IoT-type devices specifically, the claimed subject matter is not limited in this respect. Rather, embodiments described herein may encompass any of a wide range of electronic or computing device types capable of acquiring, processing, storing, or transmitting content using Internet-based or analogous infrastructure.

[0983] In some instances, challenges may be faced in improving performance of communications between or among IoT-type devices or other electronic device types, for example. An aspect of communications related to IoT-type devices or other electronic device types, for example, may involve processing of one or more queries that may be generated at IoT-type devices or other electronic device types.

[0984] As used herein, the terms “electronic content,” “content,” or the like should be interpreted broadly to refer to any form of signal or state, including, for example, signal packets or physical states on a memory device. These terms are employed in a format-agnostic manner and may encompass any expression, representation, realization, or communication of information, irrespective of encoding or medium. Content may include, without limitation, any form of information, knowledge, or experience—whether manifest as physical or non-physical signals or states.

[0985] In the context of the present disclosure, the terms “electronic content” or “online content” may refer to content that, although not necessarily directly perceivable by human senses, may nonetheless be transformed into a perceptible form by a device or system, such as through visual, optical, haptic, or auditory output. Non-limiting examples of such content include text, audio, images, video, security parameters, or any combination thereof.

[0986] Content may be stored or transmitted electronically either before or after being rendered into a perceptible form for a human observer. For purposes of consistency and ease of reference, the term “content” may be used throughout this application to denote “electronic content,” unless otherwise explicitly specified. Specific examples of content may include, but are not limited to, computer code, data, metadata, messages, text, audio files, video files, data files, web pages, or other similar types of content. However, claimed subject matter is not limited to these particular examples.

[0987] FIG. 21 is a schematic diagram that illustrates features associated with one implementation of an example operating environment 100 that may facilitate or support one ormore operations or techniques in accordance with one or more embodiments described herein. In the example shown, operating environment 100 includes IoT-type devices, which are denoted generally as element 102.

[0988] As previously discussed, the Internet of Things (IoT) typically comprises a system of interconnected or internetworked physical devices, wherein computing functionality is embedded into hardware components. This embedded functionality may enable acquisition, collection, or communication of content over one or more communication networks—potentially without requiring human intervention.

[0989] IoT-type devices may encompass a wide range of stationary or mobile hardware. Examples of such devices include, without limitation: automobile sensors, biochip transponders, heart monitoring implants, kitchen appliances, electronic locks or similar fastening mechanisms, solar panel arrays, home gateways, smart gauges, smart phones, cellular telephones, security cameras, wearable devices, thermostats, Global Positioning System (GPS) transceivers, personal digital assistants (PDAs), virtual assistants, laptop computers, personal entertainment systems, tablet PCs, desktop personal computers, personal audio or video devices, and personal navigation devices.

[0990] Operating environment 100 is presented herein by way of non-limiting example and may be implemented, in whole or in part, within a variety of wired or wireless communication infrastructures, or combinations thereof. Suitable networks may include, without limitation: public networks, such as the Internet or World Wide Web; private networks, such as intranets; wireless wide area networks (WWANs); wireless local area networks (WLANs); wireless personal area networks (WPANs); telephone networks; cable television networks; Internet access networks; fiber-optic communication systems; waveguide communication systems; or other similar infrastructures.

[0991] Claimed subject matter is not limited to any particular network configuration or operating environment. Accordingly, one or more operations or techniques for managing or updating IoT-type devices may be implemented, at least in part, in an indoor setting, an outdoor setting, or any combination thereof, depending on the implementation and requirements of a particular embodiment.

[0992] In one or more implementations, IoT-type devices 102 may receive or acquire satellite positioning system (SPS) signals 104 from SPS satellites 106. These SPS satellites may originate from a single global navigation satellite system (GNSS), such as the Global Positioning System (GPS) or the Galileo system. Alternatively, SPS satellites 106 may originate from multiple GNSS, including but not limited to GPS, Galileo, Glonass, or Beidou (also known as Compass). In certain implementations, SPS signals may additionally or alternatively bereceived from regional navigation satellite systems (RNSS), such as the Wide Area Augmentation System (WAAS), the European Geostationary Navigation Overlay Service (EGNOS), or the Quasi-Zenith Satellite System (QZSS), among others.

[0993] One or more IoT-type devices 102 may also transmit or receive wireless signals to or from a suitable wireless communication network. In a particular example, IoT-type devices 102 may communicate with a cellular communication network via wireless signals transmitted to or received from one or more wireless transmitters, such as a base station transceiver 108, across a wireless communication link 110. Additionally, IoT-type devices 102 may communicate with a local transceiver 112 via a wireless communication link 114. Base station transceiver 108 and local transceiver 112 may be of the same type or different types depending on the implementation.

[0994] These transceivers may include access points, radio beacons, cellular base stations, femtocells, access transceiver devices, or the like. Local transceiver 112 may comprise a wireless transmitter, receiver, or both, and may support communication with other terrestrial transmitters or receivers. In some instances, local transceiver 112 may provide short-range wireless communication relative to base station transceiver 108.

[0995] For example, local transceiver 112 may be positioned within an indoor environment and may enable communication via a wireless local area network (WLAN), such as a network conforming to IEEE 802.11 standards, or via a wireless personal area network (WPAN), such as a Bluetooth® network. In alternative implementations, local transceiver 112 may include a femtocell or picocell capable of supporting cellular or similar wireless communication protocols over link 114. These examples are not limiting, and in some implementations, operating environment 100 may include a greater number of base station transceivers 108, local transceivers 112, or other terrestrial wireless communication devices. Claimed subject matter is not limited to these specific examples.

[0996] In one or more implementations, IoT-type devices 102, base station transceiver 108, and local transceiver 112 may communicate with one or more servers, illustrated in this example as servers 116, 118, and 120, via a network 122 using one or more communication links 124. Network 122 may comprise any suitable combination of wired and / or wireless communication pathways.

[0997] In a particular implementation, network 122 may include Internet Protocol (IP)-based infrastructure capable of supporting communication between IoT-type devices 102 and the servers 116–120, whether directly or indirectly via local transceiver 112 or base station transceiver 108. In other implementations, network 122 may comprise a cellular communications infrastructure, such as one involving a base station controller or a masterswitching center, configured to facilitate mobile cellular communication with one or more IoT-type devices 102.

[0998] Servers 116, 118, and 120 may comprise any suitable computing resources configured to facilitate or support one or more operations or techniques described herein. Examples of such servers include, but are not limited to: update servers, backend servers, management servers, archive servers, location servers, positioning assistance servers, navigation servers, map servers, crowdsourcing servers, network-related servers, or any combination thereof.

[0999] Although a particular number of computing platforms and devices may be illustrated in the present disclosure, any suitable number or configuration of computing platforms or devices may be used to implement or support one or more techniques or operations associated with the described operating environment 100. For example, network 122 may be coupled to one or more additional wired or wireless networks, such as wireless local area networks (WLANs), to extend communication coverage for IoT-type devices 102, base station transceivers 108, local transceiver 112, servers 116–120, or any combination thereof.

[1000] In some implementations, network 122 may support or facilitate femtocell-based coverage areas. These examples are intended to be illustrative and not limiting, and claimed subject matter is not restricted to any specific configuration or network topology.

[1001] As used in this patent application, the term “IoT-type devices” is intended to broadly encompass one or more electronic or computing devices capable of utilizing existing Internet or similar infrastructure in connection with the Internet of Things (IoT). This may include support for a variety of communication protocols, domains, and applications.

[1002] Generally, the IoT refers to a system of interconnected or internetworked physical devices in which computing capabilities may be embedded into hardware components. These embedded systems enable acquisition, collection, processing, or communication of content over one or more networks, and may operate without human intervention.

[1003] IoT-type devices 102 may include, without limitation: automobile sensors, biochip transponders, heart monitoring implants, kitchen appliances, locking mechanisms, solar panel arrays, home gateways, smart gauges, smart phones, cellular telephones, security cameras, wearable electronics, thermostats, GPS transceivers, personal digital assistants (PDAs), virtual assistants, laptops, personal entertainment systems, tablet PCs, desktop PCs, personal audio or video devices, and personal navigation devices.

[1004] A “mobile device,” as used herein, may refer to a computing device whose location changes over time. A “stationary device” may refer to one that maintains a fixed location. In some embodiments, IoT-type devices may be individually addressable by an InternetProtocol (IP) address and may transmit and / or receive content over wired or wireless communication networks.

[1005] FIG.22 illustrates an embodiment 200 of an example IoT-type device. It should be noted that claimed subject matter is not limited to the specific configurations or component arrangements depicted or described for this example.

[1006] In one embodiment, an IoT-type device, such as device 200, may include one or more processors, such as processor 210, and one or more communication interfaces, such as communications interface 220. Communications interface 220 may enable wireless communications between the IoT-type device 200 and one or more external computing devices.

[1007] In some embodiments, wireless communications by an IoT-type device may be carried out substantially in accordance with any of a wide range of communication protocols, including those previously referenced in the present disclosure.

[1008] An IoT-type device, such as IoT-type device 200, may also include a memory component, such as memory 230. Memory 230 may comprise non-volatile memory and may store executable instructions, including those for operating systems, communication protocols, or applications.

[1009] Additionally, memory 230 may store updatable software or firmware code, represented in this embodiment as code 232. This code may be modified or managed based on one or more techniques described in the present patent application.

[1010] In one implementation, IoT-type device 200 may further comprise a display 240 and one or more sensors 250. The term “sensor,” as used herein, refers to any device or component capable of responding to physical stimuli and generating a corresponding signal or state.

[1011] Examples of physical stimuli include, without limitation, heat, light, sound, pressure, magnetism, or motion. Sensors may detect and respond to such stimuli in a manner that generates data signals for further processing or communication.

[1012] Non-limiting examples of sensors include: accelerometers, gyroscopes, thermometers, magnetometers, barometers, light sensors, proximity sensors, heart-rate monitors, perspiration sensors, hydration sensors, breath sensors, cameras, microphones, or any combination thereof.

[1013] In particular implementations, IoT-type device 200 includes one or more timers or counters or like circuits, such as circuitry 260, for example. In an embodiment, one or more timers or counters or the like may track one or more aspects of device performance or operation. For example, timers, counters, or other like circuits may be utilized, at least in part, by IoT-type device 200 to determine measures of fitness, for example, or to otherwise generate feedbackcontent related to testing results, in particular implementations.

[1014] Although FIG.22 depicts an example implementation of an IoT-type device, such as IoT-type device 200, the scope of claimed subject matter is not limited to this example. Other embodiments may include a wide variety of electronic or computing devices, including but not limited to desktop computers, notebook computers, high-definition televisions, digital video players or recorders, game consoles, satellite television receivers, cellular telephones, tablet devices, wearable devices, personal digital assistants, mobile audio or video playback or recording devices, or combinations thereof.

[1015] As used in the context of the present patent application, the terms “connection,” “component,” and similar terminology are intended to refer to physical constructs. However, these constructs are not always required to be tangible. Whether a term references tangible subject matter depends on the context of use. For example, a “connection path” may include a conductive material that forms a tangible electrical circuit, such as a wire controlled by a signal. Conversely, a connection may be physically real but intangible, such as a logical link between a server and client over a wireless network.

[1016] In certain contexts involving tangible components, the terms “connected” and “coupled” are not used interchangeably. “Connected” implies direct physical contact, such as through an uninterrupted electrical conductor. In contrast, “coupled” may include indirect relationships, such as optical coupling, wherein interaction occurs without direct contact. Regarding non-transitory memory components, the descriptor “physical” denotes a tangible nature of memory states or the medium itself.

[1017] When discussing tangible components or materials, a distinction is made between “on” and “over.” A substance deposited “on” a substrate is in direct physical contact without an intervening material. Conversely, “over” may refer to configurations that include one or more intermediary layers or substances between the deposited material and the substrate, thereby lacking direct physical contact while still residing above the substrate in spatial orientation.

[1018] A related distinction applies between the terms “beneath” and “under” when describing tangible materials or components. “Beneath” implies direct physical contact, analogous to the use of “on,” whereas “under” allows for the presence of intermediary layers or materials between the two elements. Accordingly, “beneath” means “immediately under,” while “under” may include configurations that are not in direct contact.

[1019] Directional terms such as “over,” “under,” “up,” “down,” “top,” and “bottom” are used for descriptive clarity and are not intended to limit the orientation or positioning of a claimed embodiment. For instance, components may be inverted or reoriented in actual use without departing from the claimed subject matter, as contextual interpretation governs. Anelement described as “on top” remains within scope even if the embodiment is flipped during use or assembly.

[1020] The term “or,” when used to associate listed items, is intended to be interpreted both inclusively and exclusively unless explicitly stated otherwise. For example, “A, B, or C” may mean A or B or C, or any combination thereof. The term “and” is generally inclusive, and expressions such as “one or more” are meant to cover both singular and plural instances. Similarly, phrases like “based on” do not imply an exhaustive list of determining factors and allow for additional, unstated elements.

[1021] Claimed subject matter involving parameters subject to measurement or specification is intended to be interpreted with flexibility, recognizing reasonable technical variation. If multiple valid measurement techniques exist, all such techniques are presumed included unless explicitly excluded. For example, different approaches to estimating slope over a region fall within scope unless the language or context expressly restricts them.

[1022] With respect to claimed subject matter that involves measurable parameters, such as physical properties including temperature, pressure, voltage, current, electromagnetic radiation, or other similar manifestations, it is asserted that these do not fall within the judicial exception to statutory subject matter as abstract ideas. Physical measurements, being observable and not merely mental steps, are therefore treated as statutory subject matter.

[1023] Typically, a measurement may be modeled as comprising at least two components: a deterministic component and a random component. The deterministic component may ideally represent a physical value, often as one or more signal samples or memory states. The random component may originate from various sources such as noise or measurement imprecision. Accordingly, in connection with claimed subject matter, both deterministic and statistical (e.g., stochastic) models may be employed to support identification or prediction of measurement-related values.

[1024] A relatively large number of measuremen...

Claims

CLAIMS What is claimed is:

1. An apparatus comprising at least one processor of at least one computing device, wherein the at least one processor is configured to: (a) implement artificial intelligence (AI) processes, neural network processes, and / or other computer-readable instructions directed to computational linguistics, wherein the computational linguistics comprise one or more of: interpretation, transcription, translation, or transliteration, or any combination thereof; (b) instantiate a plurality of AI agents, each corresponding to one or more users of a telecommunication session; (c) provide a common runtime environment for the plurality of AI agents using an AI-agent runtime adapter comprising a microkernel agent manager and a virtual machine layer; (d) facilitate access by the AI agents to one or more language models through a language model abstraction layer comprising one or more publish-subscribe bus layers; (e) enable calls by the AI agents to an AI algorithm library, wherein the AI algorithm library facilitates performance of one or more of: semiotic interpretation, hermeneutic evaluation, sentiment scoring, emotional index scoring, or axiomatic rule-based interpretation; and (f) perform real-time or near real-time linguistic transformation of digital signals representative of speech between user communication devices by executing the plurality of AI agents to apply the one or more computational linguistics operations.

2. A method for dynamic multilingual transformation of telecommunications content, the methodcomprising: (a) intercepting, by at least one processor of a telecommunications computing system, a digital or analog audio stream associated with a telecommunications session; (b) instantiating, by the processor, a plurality of AI agents, each corresponding to a participant of the session; (c) executing, by the AI agents, operations comprising one or more of: (i) interpreting, transcribing, translating, or transliterating the speech of each participant in real-time or near real-time; (ii) applying user-specific transformations based on demographic, emotional, or cultural profile data; and (iii) generating linguistically transformed output in audio or textual form; (d) deploying the AI agents via an AI-agent runtime adapter comprising a microkernel agent manager and a language model abstraction layer; and (e) facilitating communication between the agents and access to one or more language models and algorithm libraries through a publish-subscribe interface.

3. A non-transitory computer-readable medium comprising instructions stored thereon that, when executed by at least one processor of a computing system, cause the processor to: (a) detect speech from a telecommunications session and digitize the signal; (b) instantiate a plurality of AI agents, each associated with a session participant; (c) perform, by the AI agents, linguistic operations comprising interpretation, transcription, translation, and / or transliteration of the digitized speech; (d) apply user-profile-based contextual transformations to modify tone, sentiment, emotionalcues, or culturally specific references in the target output; (e) synthesize translated or transformed audio and / or generate target-language text; (f) manage the execution of the AI agents via a runtime adapter comprising a microkernel-based manager, a language model abstraction layer, and a language model access interface supporting publish-subscribe protocols; and (g) enable access to AI algorithm libraries and inter-agent communication through secure runtime interfaces.

4. The apparatus of claim 1, wherein the plurality of AI agents includes autonomous agents that operate asynchronously using one or more of a wavefront-array, systolic-array, or mesh-network configuration.

5. The apparatus of claim 1, wherein the AI-agent runtime adapter includes a software / hardware abstraction layer configured to isolate low-level hardware instructions from high-level AI operations.

6. The apparatus of claim 1, wherein the microkernel agent manager is configured to deploy and monitor runtime behavior of the AI agents using a memory-safe instruction set.

7. The apparatus of claim 1, wherein the AI agents are configured to access shared and / or private memory resources based on agent type, access level, or session context.

8. The apparatus of claim 1, wherein the AI algorithm library further supports one or more of the following: (a) user emotion modeling; (b) tone modulation based on profile data; (c) diacritic annotation for phonetic accuracy.

9. The apparatus of claim 1, wherein a first AI agent is assigned to a first communication endpoint and a second AI agent is assigned to a second communication endpoint, and the agents jointly perform in-stream translation of bidirectional speech.

10. The method of claim 2, further comprising dynamically allocating AI agent resources based on participant device type, language model complexity, or speech latency conditions.

11. The method of claim 2, wherein user-specific transformations include region-specific phrase modulation, cultural reference adjustments, or localized vocabulary substitution.

12. The method of claim 2, further comprising generating both text and audio output in the target language and synchronizing them with original speech timing for use in teleconferencing or broadcast.

13. The method of claim 2, further comprising identifying speaker identity, sentiment, and intent using neural models trained on historical communication data.

14. The method of claim 2, wherein the AI-agent runtime adapter is deployed on a telecommunications-class switch or edge device to reduce network latency.

15. The article of claim 3, wherein the non-transitory medium stores additional instructions to recompile transformed content into audio, subtitle, and transcript formats concurrently.

16. The article of claim 3, wherein execution of the instructions further enables dynamic model selection from a distributed library of language models optimized by dialect, use-case, or geography.

17. The article of claim 3, wherein profile-based transformation includes adjusting syntax, register, or sentiment score based on social media, educational, or geographic profile data.

18. The article of claim 3, wherein the instructions are configured to enable compliance with secure boot protocols and common criteria evaluation standards for protected telecommunications environments.

19. The article of claim 3, wherein the plurality of AI agents is instantiated using a runtime adapter that supports secure inter-agent messaging via encrypted publish-subscribe channels.

Images