CANCELLATION OF FEEDBACK BASED ON MACHINE LEARNING

Description

1 / 59 CANCELLATION OF FEEDBACK BASED ON MACHINE LEARNING CROSS-REFERENCE TO RELATED DEPOSIT REQUESTS

[0001] This application claims the benefit of U.S. Patent Application No. 18 / 611,494, entitled MACHINE LEARNING-BASED FEEDBACK CANCELLATION, filed March 20, 2024, and also the benefit of U.S. Provisional Patent Application No. 63 / 611,639, entitled MACHINE LEARNING-BASED FEEDBACK CANCELLATION, filed December 18, 2023, and claims the benefit of U.S. Provisional Patent Application No. 63 / 493,158, entitled LOW-LATENCY NOISE SUPPRESSION, filed March 30, 2023, both of which are expressly incorporated herein by reference in their entirety. TECHNICAL FIELD

[0002] Aspects of this disclosure generally relate to audio signal processing and, more particularly, to amplification systems with reduced artifacts. Some attributes can enable and provide improved audio signal processing, including improved audio quality by reducing howling sounds resulting from feedback when amplifying audio signals. INTRODUCTION

[0003] Audio playback devices are devices that can reproduce one or more audio signals, whether digital or analog signals. Audio playback can be incorporated into a wide variety of devices. By way of example, audio playback devices may include standalone audio devices, mobile phones, cellular or satellite radio phones, personal digital assistants (PDAs), display boards or tablets, gaming devices, or computing devices.

[0004] One class of audio playback devices are wearable devices (e.g., in-ear headphones, earphones, hearing aids, etc.) that can be used to improve hearing, situational awareness, and / or speech intelligibility. Generally, such devices employ processes of Petition 870250083890, dated 09 / 17 / 2025, pp. 432 / 515 2 / 59 Noise suppression is a relatively simple method to remove as much ambient noise as possible. Noise suppression operations can introduce artifacts into reproduced sounds, artifacts that reduce the signal-to-noise ratio (SNR) (of speech relative to ambient noise). This can be a problem because the desired sound, such as speech, can be obscured by the introduced artifacts. For some individuals, speech is readily intelligible only when the signal-to-noise ratio (of speech relative to ambient noise) is above a certain level, resulting in the artifacts making the speech unrecognizable. A brief summary of some examples.

[0005] The following summarizes some aspects of this disclosure to provide a basic understanding of the technology discussed. This summary is not an extensive overview of all contemplated attributes of the disclosure and is intended to identify key or critically important elements of all aspects of the disclosure, nor to delineate the scope of any or all aspects of the disclosure. Its sole purpose is to present some concepts of one or more aspects of the disclosure in summary form, as an introduction to the more detailed description that is presented later.

[0006] In some respects, audio amplification devices can use machine learning-based adaptive feedback cancellation (ML) to remove artifacts, such as howling, caused by feedback between a loudspeaker and a microphone. In some modalities, an ML model can be trained to preserve speech components and other desirable sound components (such as ambient sounds) while removing undesirable components, such as feedback components or howling. That is, an audio signal may include multiple environmental components, such as bird sounds, car noise, human speech, and feedback from a loudspeaker. Some components may be desirable to hear (e.g., bird sounds and human speech). Other components may be undesirable to hear (e.g., noise from Petition 870250083890, dated 09 / 17 / 2025, pp. 433 / 515 3 / 59 car and feedback). The ML model can be trained to recognize the sounds in an audio signal and remove the components of the audio signal that are undesirable.

[0007] ML-based feedback cancellation offers improved performance compared to conventional adaptive filters in the traditional feedback cancellation circuitry. For example, ML-based feedback cancellation can be more effective at isolating the feedback component of an audio signal and / or can preserve more of the desirable sound components by removing the feedback component. Effectiveness is increased, in some embodiments, because ML-based feedback cancellation can compensate for linear and nonlinear feedback-related components by modeling nonlinearities in an amplifier circuit.Although the ML-based model can be used in audio amplification systems in a manner similar to an adaptive filter, the ML-based model can solve other feedback cancellation problems and offer additional functionality and advantages in feedback cancellation, which are described in further detail in the detailed description of some embodiments below.

[0008] In one aspect of the disclosure, a method for signal processing involves receiving an input audio signal, wherein the input audio signal includes a desired audio component and a feedback component; and reducing the feedback component by applying a machine learning model to the input audio signal to determine an output audio signal. The machine learning model can be trained to generate feedback cancellation signals that, when combined with the input audio signal, reduce a magnitude or audible perception of feedback components. Such a machine learning model can be trained using training data that includes sample microphone signals recorded in the presence of feedback from a loudspeaker, wherein each set of training data includes a sample microphone signal and a representation of the feedback component present in the sample microphone signal and / or a representation of a signal of Petition 870250083890, dated 09 / 17 / 2025, pp. 434 / 515 4 / 59 Desired feedback cancellation for the sample microphone signal. The ML model can be trained by loading pre-computed weights or other ML model parameters at ML model initialization. Alternatively, the ML model can be trained by providing training data and configuring the ML model based on known feedback cancellation signals from the training data.

[0009] In a further aspect of the disclosure, an apparatus includes at least one processor and memory coupled to at least one processor. The at least one processor is configured to perform operations including receiving an input audio signal, wherein the input audio signal includes a desired audio component and a feedback component; and reducing the feedback component by applying a machine learning model to the input audio signal to determine an output audio signal.

[0010] In a further aspect of the disclosure, an apparatus includes means for receiving an input audio signal, wherein the input audio signal includes a desired audio component and a feedback component; and means for reducing the feedback component by applying a machine learning model to the input audio signal to determine an output audio signal.

[0011] In a further aspect of the disclosure, a non-transient, computer-readable medium stores instructions that, when executed by at least one processor, cause the processor to perform operations. The operations include receiving an input audio signal, where the input audio signal includes a desired audio component and a feedback component; and reducing the feedback component by applying a machine learning model to the input audio signal to determine an output audio signal.

[0012] The audio signal processing methods described in the present invention can be performed by a signal processing device. Audio signal processing can be applied to audio data captured by one or more microphones of the signal processing device. Devices Petition 870250083890, dated 09 / 17 / 2025, pages 435 / 515 5 / 59 Audio signal processing devices, which can play, record and / or process one or more audio recordings, can be incorporated into a wide variety of devices. By way of example, audio signal processing devices may include standalone audio devices such as entertainment devices and personal media players, handsets of wireless communication devices such as mobile phones, cellular or satellite radio phones, personal digital assistants (PDAs), tablets, gaming devices, computing devices such as webcams, video surveillance cameras or other devices with audio recording or playback capabilities.

[0013] The audio signal processing techniques described in the present invention may involve devices with microphones and a set of processing circuits (for example, application-specific integrated circuits (ASICs), digital signal processors (DSPs), graphics processing units (GPUs) or central processing units (CPUs)).

[0014] In some respects, a device may include a digital signal processor or a processor (e.g., an application processor) including specific functionality for audio processing. The methods and techniques described in the present invention may be entirely performed by the digital signal processor or the processor, or various operations may be divided between the digital signal processor and the processor and, in some respects, divided with additional processors. In some embodiments, the methods and techniques disclosed in the present invention may be adapted using input from a neural signal processor (NSP) where one or more signal processing parameters are controlled based on the output of a machine learning (ML) model executed by the NSP.

[0015] In a further aspect of the disclosure, a device configured for audio signal processing and / or audio capture is disclosed. The device includes means for recording audio. Examples of means may include a microphone. Petition 870250083890, dated 09 / 17 / 2025, pages 436 / 515 6 / 59 dynamic, a condenser microphone, a ribbon microphone, a carbon microphone, or a crystal microphone. The microphone can be interpreted as a microelectromechanical system (MEMS). These components can be controlled to capture the first and / or second sound recordings, which may correspond to the left and right channels of a recording.

[0016] For any of these types of microphones, the microphones may include analog and / or digital microphones. Analog microphones provide a sensor signal, which in some modes is conditioned or filtered. Analog microphones in a digital system include an external analog-to-digital converter (ADC) to interface with digital circuitry. Digital microphones include the ADC and other digital elements to convert the sensor signal into a digital data stream, such as a pulse-density modulated (PDM) stream or a pulse-code modulated (PCM) stream.

[0017] Aspects disclosed in the present invention describe the use of machine learning models as solutions to problems with amplification artifacts, such as howling. Although embodiments of the disclosure described in the present invention illustrate the use of a single machine learning model, embodiments may employ multiple machine learning models to perform the operations described in the present invention. For example, independent machine learning models may be configured to analyze and process respective frequency sub-bands of audio data, allowing appropriate processing per sub-band. Each of the independent machine learning models is trained and optimized to process a respective sub-band. As an example, a low-frequency sub-band of an audio segment may correspond to speech, while a high-frequency sub-band of the same audio segment corresponds to noise.A first machine learning model processes low-level sub-band audio data. Petition 870250083890, dated 09 / 17 / 2025, pp. 437 / 515 7 / 59 frequency to generate the first enhanced sub-band audio data in which speech is retained or enhanced. A second machine learning model processes high-frequency sub-band audio data to generate the second enhanced sub-band audio data in which noise is reduced. A combiner is used to combine the first enhanced sub-band audio data and the second enhanced sub-band audio data to generate enhanced audio data.As another example, the first machine learning model is trained to retain speech in low-frequency sub-band audio, and the second machine learning model is trained to reduce noise in high-frequency sub-band audio. This can benefit from having lower complexity and higher efficiency than a single machine learning model that is trained to process a larger frequency band to reduce speech in low-frequency sub-band audio and reduce noise in high-frequency sub-band audio.

[0018] One problem with a machine learning model for processing the higher frequency band is that the machine learning model is better suited to processing some sub-bands than others. For example, a masking network based on a long short-term memory network (LSTM) might be better suited to processing low-frequency sub-band audio, while a convolutional neural network might be better suited to processing high-frequency sub-band audio. Independent machine learning models can solve this problem by having different model architectures that are better suited to processing the respective sub-bands.In one example, the first machine learning model that is trained to process low-frequency sub-band audio might include an LSTM-based masking network, and the second machine learning model that is trained to process high-frequency sub-band audio might include a convolutional neural network architecture (e.g., U-Net). In some examples, procedural signal processing might be... Petition 870250083890, dated 09 / 17 / 2025, pp. 438 / 515 8 / 59 is performed for audio enhancement of a particular sub-band; audio enhancement can be ignored for another sub-band, or both.

[0019] A single complex machine learning model can have high resource usage (e.g., computation cycles, memory, etc.) which can limit the types of devices that can support the machine learning model. This problem can be solved by independent machine learning models that do not need to be co-located. For example, low-frequency sub-band audio processing can be performed on a first device that includes the first machine learning model, and high-frequency sub-band audio processing can be performed on a second device that includes the second machine learning model. At least some of the sub-band audio processing can thus be offloaded to another device.

[0020] Reconfiguring a large machine learning model can alter the processing across the entire frequency band. Independent machine learning models can solve this problem by being independently configurable. For example, an updated configuration that is better suited for low-frequency sub-band audio can be used for the first machine learning model without altering the second machine learning model. In another example, the second machine learning model can be updated to have a second configuration that is better suited for high-frequency sub-band audio. The configuration of independent machine learning models can be obtained from one or more sources, for example, other devices, based on the audio context.

[0021] In some cases, different microphones can capture audio with better audio quality in different sub-bands. For example, a first microphone is closer to a first sound source (e.g., a speech source) and a second microphone is closer to a second sound source (e.g., a music source). In this example, the first audio data Petition 870250083890, dated 09 / 17 / 2025, pp. 439 / 515 9 / 59 of the low-frequency sub-band from the first microphone are selected for processing using a first machine learning model to generate the first enhanced sub-band audio data in which speech is retained or enhanced, and the second high-frequency sub-band audio data from the second microphone are selected for processing using a second machine learning model to generate second enhanced sub-band audio data in which music is retained or enhanced.

[0022] In some examples, a machine learning model is used to process sub-band audio data from multiple microphones to generate enhanced sub-band audio data. A first machine learning model is used to process the first low-frequency sub-band audio data from a first microphone and the second low-frequency sub-band audio data from a second microphone to generate enhanced low-frequency sub-band audio data. A second machine learning model is used to process the first high-frequency sub-band audio data from the first microphone and the second high-frequency sub-band audio data from the second microphone to generate enhanced high-frequency sub-band audio data.

[0023] Processing audio data from multiple microphones can improve the performance of machine learning models. For example, enhanced low-frequency sub-band audio data generated by the first machine learning model processing low-frequency sub-band audio data from multiple microphones may have improved speech and reduced noise compared to enhanced low-frequency sub-band audio data based on low-frequency sub-band audio data from a single microphone. As another example, enhanced high-frequency sub-band audio data generated by the second machine learning model processing high-frequency sub-band audio data from multiple microphones may have improved speech and reduced noise compared Petition 870250083890, dated 09 / 17 / 2025, pages 440 / 515 10 / 59 with enhanced high-frequency sub-band audio data based on high-frequency sub-band audio data from a single microphone.

[0024] Separate machine learning models that are trained to process different sub-bands may have lower complexity (e.g., fewer network nodes, network layers, etc.) and higher efficiency (e.g., faster processing time, fewer computation cycles, etc.) compared to a single machine learning model that is trained to process a larger frequency band that includes the sub-bands.

[0025] Other aspects, attributes, and implementations will become apparent to those skilled in the art by reviewing the description of specific exemplary aspects presented below, together with the attached figures. Although attributes may be discussed in relation to certain aspects and figures below, several aspects may include one or more of the advantageous attributes discussed in the present invention. In other words, although one or more aspects may be discussed as having certain advantageous attributes, one or more of these attributes may also be used in accordance with the various aspects. Similarly, although exemplary aspects may be discussed below as device, system, or method aspects, the exemplary aspects may be implemented in various devices, systems, and methods.

[0026] The method may be incorporated into a computer-readable medium in the form of computer program code comprising instructions that cause a processor to perform the steps of the method. In some embodiments, the processor may be part of a mobile device that includes a first network adapter configured to transmit data, such as images or videos (with associated or embedded sounds) in a recording or in the form of streaming data, through a first network connection among a plurality of network connections; and a processor coupled to the first network adapter and to memory. The processor may cause the image frames of Petition 870250083890, dated 09 / 17 / 2025, pp. 441 / 515 11 / 59 output described in the present invention are transmitted via a wireless communication network, such as a 5G NR communication network.

[0027] The foregoing has described, in a fairly broad manner, the attributes and technical advantages of the examples according to the disclosure so that the detailed description that follows may be better understood. Additional attributes and advantages will be described hereafter. The specific design and examples disclosed may be readily used as a basis for modifying or designing other structures to accomplish the same purposes as the present disclosure. Such equivalent constructions do not depart from the scope of the appended claims. The characteristics of the concepts disclosed in the present invention, both their organization and method of operation, together with associated advantages, will be better understood from the following description when considered in conjunction with the accompanying figures. Each of the figures is provided for the purpose of illustration and description, and not as a definition of the limits of the claims.

[0028] Although aspects and implementations are described in this application by way of illustration to a few examples, those skilled in the art will understand that additional specific implementations and use cases may arise in many different arrangements and scenarios. The innovations described in the present invention can be implemented in many different types of platforms, devices, systems, formats, sizes, and packaging arrangements. For example, aspects and / or uses may occur via embedded chip implementations and other non-module and component-based devices (e.g., end-user devices, vehicles, communication devices, computing devices, industrial equipment, retail / shopping devices, medical devices, artificial intelligence (AI) enabled devices, etc.).Although some examples may or may not be specifically targeted at use cases or applications, there may be a wide range of applicability to the innovations described. Implementations can vary in spectrum from... Petition 870250083890, dated 09 / 17 / 2025, pp. 442 / 515 12 / 59 chip-level or modular components to non-modular, non-chip-level implementations, and which additionally aggregate distributed devices or original equipment manufacturer (OEM) devices or systems that incorporate one or more aspects of the innovations described. In some practical environments, devices incorporating the described aspects and attributes may necessarily also include other components and attributes for the implementation and practice of the claimed and described aspects. It is intended that the innovations described in the present invention can be practiced in a wide variety of devices, chip-level components, systems, distributed arrangements, end-user devices, etc., of varying sizes, shapes, and constitution. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] A further understanding of the nature and advantages of this disclosure can be obtained by reference to the following drawings. In the accompanying figures, similar components or attributes may have the same reference label. Additionally, several components of the same type may be distinguished by placing, after the reference label, a dash and a second label that distinguishes them from similar components. If only the first reference label is used in the descriptive report, the description will apply to any of the similar components that have the same first reference label, regardless of the second reference label.

[0030] Figure 1 shows a block diagram of a system-on-chip (SoC) configured to perform signal processing in accordance with one or more aspects of this disclosure.

[0031] Figure 2 is a block diagram that illustrates an example data flow path for audio signal processing in a multimedia device according to one or more disclosure aspects. Petition 870250083890, dated 09 / 17 / 2025, pp. 443 / 515 13 / 59

[0032] Figure 3 is a block diagram illustrating a machine learning-based feedback cancellation audio system according to some disclosure modalities.

[0033] Figure 4 is a block diagram illustrating an audio system with machine learning-based feedback cancellation applied before forward path processing according to some disclosure modalities.

[0034] Figure 5 is a block diagram illustrating an audio system with feedback cancellation based on machine learning and another feedback cancellation system according to some disclosure modalities.

[0035] Figure 6 is a block diagram illustrating a machine learning-based feedback cancellation audio system based on additional sensors according to some disclosure modalities.

[0036] Figure 7 is a block diagram illustrating a feedback-canceling audio system based on machine learning applied through a time-domain filter according to some disclosure modalities.

[0037] Figure 8 shows a flowchart of an example method for processing audio data to perform acoustic source separation according to some disclosure modalities.

[0038] Figure 9 is an illustrative block diagram of an example machine learning (ML) model represented by an artificial neural network (ANN) according to some embodiments of the disclosure.

[0039] Figure 10 is a diagram of an operable headset for performing feedback cancellation according to some examples of the present disclosure.

[0040] Figure 11 is a diagram of a headset, such as a virtual reality, mixed reality, or augmented reality headset, operable for performing feedback cancellation, according to some examples of the present disclosure. Petition 870250083890, dated 09 / 17 / 2025, pp. 444 / 515 14 / 59

[0041] Figure 12 is a diagram of augmented reality glasses for performing feedback cancellation, according to some examples in this disclosure.

[0042] Figure 13 is a diagram of an operable wearable device for performing feedback cancellation according to some examples in this disclosure.

[0043] Figure 14 is a diagram of an operable in-ear headphone for performing feedback cancellation according to some examples in this disclosure.

[0044] Figure 15 is a diagram of an illustrative aspect of an operable system for performing machine learning-based audio sub-band processing, according to some examples in this disclosure.

[0045] Figure 16 is a diagram of an illustrative aspect of the components of a system, according to some examples in this disclosure.

[0046] Figure 17 is a diagram of an illustrative aspect of the components of a system, according to some examples from this disclosure.

[0047] Similar reference numbers and designations in the various drawings indicate similar elements. DETAILED DESCRIPTION

[0048] This disclosure provides computer-readable systems, apparatus, methods, and means that support signal processing, including techniques for adaptive feedback cancellation based on machine learning (ML) to remove artifacts, such as howls, caused by feedback between a loudspeaker and a microphone. In some embodiments, an ML model can be trained to preserve speech components and other desirable sound components (such as ambient sounds) while removing other undesirable components, such as feedback components. Several embodiments are disclosed below that provide different techniques for applying an ML model to an audio signal that can be used in an amplification system. Each of the models Petition 870250083890, dated 09 / 17 / 2025, pages 445 / 515 15 / 59 of ML can be trained to achieve adequate processing for this mode configuration to provide feedback cancellation.

[0049] Particular implementations of the subject matter described in this disclosure can be implemented to achieve one or more of the following potential advantages or benefits. In some respects, this disclosure provides techniques for improving the sound quality of output audio by reducing feedback and artifacts resulting from amplifying an audio signal from a microphone in the sound field of a loudspeaker emitting the amplified audio signal. ML-based feedback cancellation offers improved performance by being more effective at isolating the feedback component of an audio signal while preserving desirable sound components. Effectiveness is increased, in some embodiments, because ML-based feedback cancellation can compensate for both linear and nonlinear feedback-related components.

[0050] The detailed description set forth below, in connection with the accompanying drawings to which the text refers, is intended to be a description of various embodiments and is not intended to limit the scope of the disclosure. Instead, the detailed description includes specific details for the purpose of providing a complete understanding of the subject matter of this disclosure. It will be evident to those skilled in the art that these specific details are not necessary in all cases and that, in some cases, well-known structures and components are shown in block diagram form for the sake of clarity.

[0051] In the embodiment description of the present invention, numerous specific details are set forth, such as examples of specific components, circuits and processes, to provide a complete understanding of the present disclosure.The term "coupled," as used in the present invention, means directly connected to, or connected through, one or more intervening components or circuits. Furthermore, in the following description, and for explanatory purposes, specific nomenclature is established to provide a complete understanding of the present disclosure. However, it will be evident to those skilled in the art. Petition 870250083890, dated 09 / 17 / 2025, pp. 446 / 515 16 / 59 that these specific details may not be necessary for the practice of the teachings disclosed in the present invention. In other cases, well-known circuits and devices are shown in block diagram form to avoid obscuring the teachings of the present disclosure.

[0052] Some portions of the detailed descriptions that follow are presented in terms of procedures, logic blocks, processing, and other symbolic representations of bit data operations within a computer memory. In the present disclosure, a procedure, a logic block, a process, or the like, is conceived to be a self-consistent sequence of steps or instructions that lead to a desired result. Steps are those that require physical manipulations of physical quantities. In general, though not necessarily, these quantities take the form of electrical or magnetic signals capable of being stored, transferred, combined, compared, and otherwise manipulated in a computer system.

[0053] An example device for recording sounds and / or processing sound signals using one or more microphones, such as a microelectric mechanical system (MEMS) microphone, may include a configuration of one, two, three, four, or more microphones in different locations in the device. The example device may include one or more digital signal processors (DSPs), AI engines, or other suitable circuitry for processing, such as performing certain noise-canceling operations, signals captured by the microphones. The one or more digital signal processors (DSPs) may output signals representing sounds via a bus for storage in memory, for playback by an audio system, and / or for further processing by other components (such as an application processor).

[0054] The processing circuitry may perform additional processing, such as encoding, storage, transmission, or other manipulation of audio signals. In some embodiments, the example device may include an audio circuitry including an amplifier of Petition 870250083890, dated 09 / 17 / 2025, pp. 447 / 515 17 / 59 audio (e.g., a Class D amplifier) ​​to drive a transducer to reproduce the sounds represented by the audio signals. A loudspeaker may be integrated into the device and coupled to the audio amplifier to be driven by the audio amplifier to reproduce the sounds. A connection may be provided by a connector or other connector on the device to couple an external transducer (e.g., an external loudspeaker or headphones) to the audio amplifier to be driven by the audio circuitry to reproduce the sounds. In some embodiments, the connector may instead output a digital signal for conversion and amplification by an external device, such as when the connector is configured to be coupled to a digital device via a Type-C Universal Serial Bus (USB-C) connection and some or all of the audio circuitry is bypassed.

[0055] Figure 1 shows a block diagram of a system-on-a-chip (SoC) configured to perform signal processing according to one or more aspects of this disclosure. The SoC 100 may include several components coupled together via a bus 102, which may be a network-on-a-chip (NoC) or a plurality of NoCs interconnecting several components. For example, although Figure 1 illustrates several components coupled to bus 102, the various components may be coupled to different buses with additional buses connecting the different buses to provide a path for communication between the components.

[0056] An example component in the SoC 100 is a digital signal processor (DSP) 112 for signal processing. The DSP 112 can process audio signals received from microphones 130A, 130B, and 130C of the microphone array 130 (which may include one or any number of microphones, although three are shown). The DSP 112 may include custom hardware to perform a limited set of operations on specific data types. For example, a DSP may include transistors coupled together to perform operations on streaming data and use memory architectures and / or access techniques to Petition 870250083890, dated 09 / 17 / 2025, pages 448 / 515 18 / 59 fetch multiple data or instructions simultaneously. Such configurations may allow the DSP 112 to operate on real-time data, such as video data, audio data, or modem data, in a power-efficient manner.

[0057] The SoC 100 also includes a central processing unit (CPU) 104 and a memory 106 that stores instructions 108 (for example, a memory that stores processor-readable code or a non-transient, computer-readable medium that stores instructions) that can be executed by a processor of the SoC 100. The CPU 104 can be a single central processing unit (CPU) or a CPU cluster comprising two or more cores, such as the 104A core. The CPU 104 may include hardware capable of performing generic operations on many data types, such as hardware capable of executing instructions from the Advanced RISC Machines (ARM®) instruction set, such as ARMv8 and ARMv9.For example, a CPU 104 might include transistors coupled together to perform operations to support the execution of an operating system and user applications (e.g., a camera application, a multimedia application, a gaming application, a productivity application, a messaging application, a video calling application, an audio recording application, a video recording application). The CPU 104 might execute instructions 108 retrieved from memory 106. In some embodiments, the CPU 104 running an operating system might coordinate the execution of instructions by multiple components within the SoC 100. For example, the CPU 104 might retrieve instructions 108 from memory 106 and execute the instructions in the DSP 112.

[0058] The SoC 100 may additionally include a neural signal processor (NSP) 124 to run machine learning (ML) models related to multimedia applications. The NSP 124 may include hardware configured to perform and accelerate convolution operations involved in running machine learning algorithms. For example, the NSP 124 may improve performance when running predictive models, such as artificial neural networks (ANNs) (including Petition 870250083890, dated 09 / 17 / 2025, pp. 449 / 515 19 / 59 multilayer feedforward neural networks (MLFFNN), recurrent neural networks (RNN), and / or radial basis functions (RBF). The ANN executed by NSP 124 can access predefined training weights stored in memory 106 to perform operations on user data.

[0059] The SoC 100 can be coupled to a display 114 to interact with a user. The SoC 100 can also include a graphics processing unit (GPU) 126 to render images on the display 114. The images can be coordinated with the processed sound emitted by the audio processing circuitry, such as when the SoC 100 is running a user application for multimedia playback or game applications. In some embodiments, the CPU 104 can perform rendering to the display 114 without a GPU 126. In some embodiments, the GPU 126 can be configured to execute instructions to perform operations unrelated to image rendering, such as processing large volumes of datasets in parallel.

[0060] Processing algorithms, techniques, and methods described in the present invention can be executed by at least one SoC 100 processor, which may include execution by all steps on one of the processors (e.g., DSP 112, CPU 104, NSP 124, GPU 126) or may include execution of steps through a combination of one or more of the processors (e.g., DSP 112, CPU 104, NSP 124, GPU 126). In some embodiments, at least one of the processors executes instructions to perform various operations described in the present invention, including machine learning-based feedback cancellation.For example, the execution of instructions by CPU 104 as part of a multimedia application (e.g., a voice recorder, a sound recording, or a video recorder) may instruct DSP 112 to start or stop capturing audio from one or more microphones 130A to 130C and a digital audio signal processed in NSP 124 to reduce the presence of feedback (e.g., howling sounds) in an audio signal. The operations of CPU 104 may... Petition 870250083890, dated 09 / 17 / 2025, pages 450 / 515 20 / 59 being based on user input. For example, a voice recorder application running on processor 104 may receive a user command to initiate a voice recording whereby audio comprising one or more channels is captured and processed for playback and / or storage. Audio processing to determine output or corrected signals, as per the techniques described in the present invention, may be applied to one or more audio segments in the recording sequence.

[0061] Input / output components can be coupled to the SoC 100 via an input / output (I / O) hub 116. An example of a hub 116 is an interconnect to a peripheral component interconnect express (PCIe) bus. Example components coupled to the hub 116 may be components used to interact with a user, such as a touchscreen interface and / or physical buttons. Some components coupled to the hub 116 may also include network interfaces for communication with other devices, including a wide area network (WAN) adapter (e.g., WAN adapter 152), a local area network (LAN) adapter (e.g., LAN adapter 153), and / or a personal area network (PAN) adapter (e.g., PAN adapter 154). A WAN 152 adapter can be a 4G LTE or 5G NR wireless network adapter.A LAN adapter 153 can be an IEEE 802.11 WiFi wireless network adapter. A PAN adapter 154 can be a Bluetooth wireless network adapter. Each of the WAN adapters 152, LAN adapter 153, and / or PAN adapter 154 can be coupled to an antenna that can be shared by each of the adapters 152, 153, and 154, or coupled to multiple antennas configured for primary and diversity reception and / or configured to receive specific frequency bands. In some embodiments, the WAN adapter 152, LAN adapter 153, and / or PAN adapter 154 can share a set of circuits, such as portions of a radio frequency front end (RFFE). Petition 870250083890, dated 09 / 17 / 2025, pages 451 / 515 21 / 59 In some embodiments, the data transmitted through the I / O hub 116 may include processed audio signals in accordance with aspects of this disclosure. For example, the processed audio signals may be output through the WAN adapter 152 or LAN adapter 153 to a network media playback device or may be output through the PAN adapter 154 to a user's personal audio device (e.g., speaker, headset or in-ear headphones).

[0062] The audio circuitry set 156 can be integrated into the SoC 100 as a dedicated circuitry set for coupling the SoC 100 to an external loudspeaker 120, which can be a transducer, such as a loudspeaker (internal or external to a device incorporating the SoC 100) or headphones. The audio circuitry set 156 can include encoder / decoder (CODEC) functionality for processing digital audio signals. The audio circuitry 156 may additionally include one or more amplifiers (e.g., a class D amplifier) ​​to drive a transducer coupled to the SoC 100 to emit sounds generated during the execution of applications by the SoC 100. The functionality related to the audio signals described in the present invention may be performed by a combination of the audio circuitry 156 and / or other processors (e.g., CPU 104, DSP 112, GPU 126, NSP 124) of the SoC 100.

[0063] The SoC 100 can couple with external devices outside the SoC package. 100. For example, the SoC 100 can be coupled to a power source 118, such as a battery or an adapter to couple the SoC 100 to a power source. The signal processing described in the present invention can be adapted and achieve power efficiency to support the operation of the SoC 100 from a limited-capacity power source 118, such as a battery. For example, operations can be performed on a portion of the SoC 100 configured to perform the operation at a lower power consumption. As another example, the operations themselves are performed in a way that reduces the number of computations to perform the operation, so that the algorithm is optimized for Petition 870250083890, dated 09 / 17 / 2025, pp. 452 / 515 22 / 59 extend the operating time of a device while powered by a limited-capacity power supply 118. In some embodiments, the operations described in the present invention can be configured based on a type of power supply 118 that provides power to the SoC 100. For example, a first set of operations can be performed to perform a function when the power supply 118 is a wall adapter. As another example, a second set of operations can be performed to perform a function when the power supply 118 is a battery.

[0064] The SoC 100 may also include, or be coupled to, additional attributes or components not shown in Figure 1. Although the components are shown integrated as a single SoC 100, which may include all components built on a single semiconductor wafer with a common semiconductor substrate, other arrangements of the illustrated blocks with different numbers of wafers, substrates and / or packages may be arranged to effect the same functionality described in this disclosure.

[0065] Memory 106 may include a non-temporary or non-transient computer-readable medium that stores computer-executable instructions such as instructions 108 to perform all or part of one or more operations described in this disclosure. Instructions 108 may include a multimedia application (or other suitable application, such as a message-passing application) to be executed by the SoC 100 that records, processes, or outputs audio signals. Instructions 108 may also include other applications or programs executed by the SoC 100, such as an operating system and applications, in addition to multimedia processing.

[0066] In addition to instructions 108, memory 106 can also store audio data. The SoC 100 can be coupled to external memory and configured to access the memory in order to record output audio files for later playback or long-term storage. For example, the SoC 100 can be coupled to a flash storage device comprising NAND memory. Petition 870250083890, dated 09 / 17 / 2025, pp. 453 / 515 23 / 59 for storing video files (e.g., MP4 container-formatted files) including audio tracks and / or storing audio recordings (e.g., MPEG-1 Layer 3 files, also called MP3 files). Portions of the video or audio files can be transferred to memory 106 for processing by SoC 100, with the resulting signals after processing encoded as video or audio files in memory 106 for transfer to long-term storage.

[0067] Although the SoC 100 is mentioned in the examples in the present invention to carry out aspects of the present disclosure, some device components may not be shown in Figure 1 to avoid obscuring aspects of the present disclosure. Furthermore, other components, numbers of components, or combinations of components may be included in a suitable device to carry out aspects of the present disclosure. Thus, the present disclosure is not limited to a particular device or component configuration.

[0068] The SoC in Figure 1 can be operated to obtain improved audio recordings and / or an enhanced user experience through higher quality audio playback by applying feedback cancellation to an audio signal based on a machine learning (ML) model. An example method for performing multimedia operations is shown in Figure 2 and described below.

[0069] Figure 2 is a block diagram illustrating an example data flow path for audio signal processing in a multimedia device according to one or more aspects of the disclosure. The System 200 SoC 100 may execute the multimedia control 210, as part of an operating system or driver, to control the capture of sounds from microphones or other audio sources and / or to control the configuration of the audio processing circuitry 156. The audio configuration applied by the multimedia control 210 to output devices (e.g., speakers) or input devices (e.g., microphones) may include parameters that specify, for example, Petition 870250083890, dated 09 / 17 / 2025, pp. 454 / 515 24 / 59 for example, a bit depth, a sampling rate, a data rate, a magnitude, or other parameters.

[0070] The multimedia control 210 can be managed by or provide services to a multimedia application 204. The multimedia application 204 can also run on the SoC 100 including one or more processors of the SoC 100. The multimedia application 204 provides user-accessible settings so that a user can specify individual playback settings or select a profile with corresponding playback settings. The multimedia application 204 can be, for example, a video recording application, a screen sharing application, a virtual conferencing application, an audio playback application, a messaging application, a video communications application, or another application that processes audio data. The multimedia application 204 can include feedback cancellation 206 to improve the quality of the audio presented to the user during the execution of the multimedia application 204.Feedback cancellation 206 can perform one or more or a combination of the techniques described in the present invention and, in some embodiments, feedback cancellation 206 can be performed by multiple processing units within the SoC 100 (such as with portions performed by DSP 112 and portions performed by NSP 124).

[0071] An example implementation for feedback cancellation 206 is shown in Figure 3. Figure 3 is a block diagram illustrating a machine learning-based feedback-cancelled audio system according to some disclosure modalities. Microphone 130 can receive audio signals 370 from a desired audio signal source. Microphone 130 can also receive feedback 324 from the output audio signals 372 of speaker 120 (or another transducer in the reception field of microphone 130). Microphone 130 generates audio signals 304 that include a feedback component corresponding to feedback 324 and a desired audio component corresponding to audio signals 370. The audio signals 304 can be processed in a direct path. Petition 870250083890, dated 09 / 17 / 2025, pages 455 / 515 25 / 59 322, as by amplifying the audio signals 304. The output of the direct path 322 can be processed with one or more machine learning models 326 to determine the output audio signals 314. The output audio signals 314 can be amplified in the amplifier circuit 336 to reproduce the output audio signals 314 through the loudspeaker 120 as sounds 372. The amplifier circuit 336 can be, for example, a class D amplifier.

[0072] The machine learning model(s) 326 can be trained to preserve the desired input, such as the audio signal 370, while removing artifacts resulting from feedback 324. An example artifact is howling, which results from amplification in the forward path 322 increasing the magnitude of the feedback 324 until the feedback 324 is loud enough in magnitude to create an audible howling noise from the loudspeaker 120. The feedback 324 can have linear and non-linear components, and the machine learning model(s) 326 can be configured to cancel the linear and / or non-linear components. The feedback 324 can be represented by a transfer function between the loudspeaker 120 and the microphone 130. State-of-the-art techniques for feedback cancellation have involved adaptive filters configured to estimate this transfer function.However, adaptive filters were restricted to canceling linear components of the transfer function for feedback 324.

[0073] The machine learning model(s) 326 can cancel nonlinear components of the transfer function and, in some embodiments, the linear components. The machine learning model(s) 326 can, in some embodiments, also be combined with other feedback cancellation processing to further improve the audio quality of the sounds 372 emitted by the loudspeaker 120. Example models for the machine learning model(s) 326 include masking, complex masking, SGN and its variants, including the SGN example embodiments illustrated and described with reference to Figure 15, Figure 16 and Figure 17 and Unet and its variant ShrimpNet. Petition 870250083890, dated 09 / 17 / 2025, pp. 456 / 515 26 / 59

[0074] In some modes, the machine learning model(s) 326 can switch between transparency and noise cancellation modes, in which the output signals are generated by processing audio signals with different coefficients. The operating mode (whether transparency, noise cancellation, or other) can be selected by the user or determined based on one or more criteria.

[0075] In some embodiments, feedback cancellation is performed by the machine learning model(s) at the output of the forward path 322, as shown in Figure 3. In other embodiments, feedback cancellation can be applied to the audio signals 304 before processing in the forward path 322. An example of an audio system for pre-amplification feedback cancellation with machine learning models is shown in Figure 4. Figure 4 is a block diagram illustrating an audio system with machine learning-based feedback cancellation applied before forward path processing according to some embodiments of the disclosure.

[0076] In Figure 4, the machine learning model(s) 426, which may be any of the machine learning models described in relation to Figure 3, may modify the audio signals 304 before processing in the forward path 322. The machine learning models 426 may determine a feedback cancellation signal to be combined with the audio signal 304 received from microphone 130, with the combined signal processed in the forward path 322. The machine learning model(s) 426 may output the feedback cancellation signal based on the audio signals 314 from the output of the forward path 322 being played by the loudspeaker 120.

[0077] In some embodiments, the machine learning model(s) 326 may be combined with one or more adaptive filters. The combination may be configured so that the adaptive filter, which may be executed by the DSP 112, removes linear feedback components 324 and the machine learning model(s), which may be executed by the NSP 124, remove(s) Petition 870250083890, dated 09 / 17 / 2025, pp. 457 / 515 27 / 59 nonlinear feedback components 324. An example of combining machine learning models with other feedback cancellation is shown in Figure 5. Figure 5 is a block diagram illustrating an audio system with feedback cancellation based on machine learning and other feedback cancellation according to some disclosure modalities.

[0078] In Figure 5, the machine learning model(s) 526, which can be configured similarly to the machine learning model(s) 326 in Figure 3, can modify the audio signals after processing in the forward path 322. The feedback canceller 528 can determine a feedback cancellation signal to match the audio signals 304 before processing in the forward path 322. The feedback canceller 528 can be an adaptive filter that reduces the linear components of the feedback 324 before the forward path 322. The machine learning model(s) 526 can be configured to reduce non-linear components of feedback 324 before output to the loudspeaker 120. The machine learning model(s) 526 can improve the feedback cancellation performed by the audio system in Figure 5.Although certain blocks are described as providing linear component cancellation and / or nonlinear component cancellation, each of the 526 and 528 blocks can perform linear component cancellation, nonlinear component cancellation, or a combination of linear and nonlinear component cancellation.

[0079] The machine learning model(s) 526 can be configured to receive feedback (as input parameters related to the feedback cancellation signal) from the feedback canceller 528. For example, an adaptive filter of the feedback canceller 528 can provide information to the machine learning model(s) 526 regarding the configuration of the adaptive filter that the machine learning model(s) 526 use(s) as input to estimate nonlinear components in the output of the forward path 322, and this estimate is used to modify the output of the forward path 322 to reduce artifacts. Petition 870250083890, dated 09 / 17 / 2025, pages 458 / 515 28 / 59 as howls. In some embodiments, the feedback canceller 528 provides the machine learning model(s) 526 with an indicator of whether the howl was detected in the audio signal 304, which can be used to activate the machine learning model(s) 526. In some embodiments, the feedback canceller 528 provides the machine learning model(s) 526 with one or more of the adaptive filter's FIR coefficients F(q), forward path coefficients D(q) 322, open-loop transfer function 1+F(q)D(q), an estimate of the gain margin (e.g., a distance of the open-loop transfer function from the Nyquist point (-1,0)) and / or proximity to an adaptive filter instability point.

[0080] In some modes, the machine learning model(s) 626, which can be configured similarly to the machine learning model(s) 526 in Figure 5, can receive information from one or more additional sensors 626. Figure 6 is a block diagram illustrating a machine learning-based feedback cancellation audio system based on additional sensors according to some disclosure embodiments. The data from the additional sensor(s) 626 can be fed into the machine learning model(s) 626 to improve the feedback cancellation provided by the machine learning model(s) 626.The additional sensor(s) 626 may be located outside the feedback path 324 or may be separate (e.g., a different type of sensor than an audio sensor) from the feedback 324, which may improve the ability of the machine learning model(s) 626 to discriminate between howling resulting from feedback 324 processed in the direct path 322 and howling in the audio signal 370 (e.g., music, bird sounds, sirens, etc.). An example of an additional sensor is a camera, which receives optical signals that are not part of the feedback 324.

[0081] Figure 7 is a block diagram illustrating an audio system with feedback cancellation based on machine learning applied through a time-domain filter according to some modalities of Petition 870250083890, dated 09 / 17 / 2025, pp. 459 / 515 29 / 59 disclosure. This configuration for a machine learning model can reduce latency by performing analysis on the ML model in the frequency domain, with the output coefficients of the ML model for a time-domain filter. In some embodiments, the machine learning model(s) 726 is / are trained to generate output data that includes target audio components from the frequency-domain audio data of the forward path 322 and omits or suppresses non-target audio components from the frequency-domain audio data. Such models may be referred to as inline models. In some embodiments, the machine learning model(s) 726 is / are trained to generate output data that includes the non-target audio components from the frequency-domain audio data of the forward path 322 and omits or suppresses target audio components from the frequency-domain audio data. Such models are referred to as masking models.The output data from the masking models can be used directly as the noise suppression output and combined with the output of the forward path 322. The output data from the inline models can be further processed to generate the noise suppression output 128.

[0082] In some embodiments, the output coefficients of the ML model may be mask coefficients. These mask coefficients may be inserted into a time-domain filter and applied to the output of the forward path. In Figure 7, the machine learning model(s) 726, which may be configured similarly to the machine learning model(s) 326, receive(s) the audio signal from the forward path 322 to determine coefficients for a time-domain filter 724. The time-domain filter 724 also receives the output audio signal from the forward path 322 to produce output audio signals 314 for playback by the loudspeaker 120.

[0083] System 200 in Figure 2 can be configured to perform the operations described with reference to the modal aspects of Figure 3, Figure 4, Figure 5, Figure 6 and / or Figure 7 to determine an audio signal. Figure Petition 870250083890, dated 09 / 17 / 2025, pages 460 / 515 Figure 30 / 59 shows a flowchart of an example method for processing audio data to perform acoustic source separation according to some embodiments of this disclosure. The operations in Figure 3 can result in an audio signal with improved sound representation, resulting in an enhanced user experience. Each of the operations described with reference to Figure 3 can be performed by one or a combination of the SoC 100 processors.

[0084] In block 802, incoming audio signals are received. Incoming audio signals can be received, for example, from microphones. Alternatively, audio data can be received from a wireless microphone, wherein the audio data is received through one or more of the WAN adapter 152, LAN adapter 153, and / or PAN adapter 154. Alternatively, audio data can be received from a memory location or a network storage location, such as when the audio signal was previously captured and is now retrieved from memory 106 and / or a remote location through one or more of the WAN adapter 152, LAN adapter 153, and / or PAN adapter 154. In some embodiments, the reception (e.g., capture or retrieval) of audio signals can be initiated by the multimedia application 204 running on the SoC 100.The audio data, comprising the audio signals, can be retrieved in block 802 and subsequently processed by SoC 100 according to the operations described in one or more of the following blocks.

[0085] In block 804, the input audio signals are processed to reduce the feedback component by applying a machine learning model to the input audio signal received in block 802. The application of the machine learning model to the input audio signals can be performed according to aspects of Figure 3, Figure 4, Figure 5, Figure 6 and / or Figure 7.

[0086] The operations described with reference to blocks 802 and 804 of Figure 8 can be performed on a digital signal processor (DSP), such as the DSP 112 of the SoC 100 illustrated in Figure 1. However, the operations can, Petition 870250083890, dated 09 / 17 / 2025, pp. 461 / 515 31 / 59 Alternatively, they may be performed by one or more of the processors in Figure 1, including one or more of the CPU 104, DSP 112, GPU 126, or NSP 124. For example, CPU 104 may control the recording of input audio signals from the microphone array 130 to memory 106 as part of block operations 802. DSP 112 and / or NSP 124 may then perform block operations 804 on the audio signals stored in memory 106, after which the output signals determined by DSP 112 and / or NSP 124 may be stored in memory 106, output to audio circuitry 156 for playback by the speaker 120, and / or transmitted to another device via one or more of the WAN 152, LAN 153, and / or PAN 154. In another example, the processor performing the operations of 802 and / or 804 blocks can be a set of dedicated logic circuits for performing certain operations.

[0087] Certain aspects and techniques, as described in the present invention, can be implemented, at least in part, using an artificial intelligence (AI) program, for example, a program that includes a machine learning (ML) model. For example, the machine learning model(s) 326 can implement an AI program as described with reference to Figure 9. The ML model can be trained offline to receive an input including clean speech, unstable feedback (e.g., howls or other artifacts), and ambient sounds (e.g., howl-like sounds). The ML model can be trained to emit clean speech and ambient sounds without the unstable feedback (e.g., feedback artifacts such as howling).

[0088] An example ML model might define computing capabilities to make determinations from input data, with the determinations made based on patterns identified in the input data. Computing capabilities can be defined in terms of weights and propensities. Weights might indicate relationships between certain input data and certain determinations. Propensities might indicate a starting point for the determinations. An example ML model operating on input data might start at a defined determination. Petition 870250083890, dated 09 / 17 / 2025, pp. 462 / 515 32 / 59 by the propensities and then alter its determination based on a combination of the input data and weights. The determinations of an ML model can be one or more of decisions, predictions, inferences, or values. The decisions, predictions, or inferences can be represented as output values ​​of an ML model. In some embodiments of this disclosure, an ML model can be configured to provide computing capabilities for feedback cancellation in audio signal processing. Such an ML model can be configured with weights and / or propensities to perform feedback cancellation by recognizing desirable aspects of audio (e.g., speech and ambient sounds) and reducing or eliminating undesirable aspects of audio (e.g., feedback artifacts).Thus, during the operation of a device, the ML model can receive input data (e.g., microphone signals) and make determinations (e.g., output audio signals with reduced feedback artifacts) based on weights and / or biases. ML models that can be configured in this way according to the modalities of this disclosure include supervised ML models, unsupervised ML models, and ML models for classification and / or regression.

[0089] The description in the present invention illustrates, by way of example, how one or more tasks / problems in feedback cancellation in audio signal processing can benefit from the application of one or more ML models using an ANN. In some embodiments, other type(s) of ML models may be used instead of an ANN. Therefore, unless expressly stated, the subject matter relating to an ML model is not necessarily intended to be limited to an ANN solution. Furthermore, it should be understood that, except where specifically indicated otherwise, terms such as AI / ML model, ML model, trained ML model, ANN, model, algorithm or the like are intended to be interchangeable.

[0090] Figure 9 is an illustrative block diagram of an example machine learning (ML) model represented by an artificial neural network (ANN) 900. The ANN 900 can receive input data 906 which may include one or Petition 870250083890, dated 09 / 17 / 2025, pp. 463 / 515 33 / 59 plus bits of 902 data, preprocessed data output from preprocessor 904 (optional), or some combination thereof. Here, the 902 data may include training data, verification data, application-related data, or similar, for example, depending on the deployment stage of the ANN 900. Preprocessor 904 may be included within the ANN 900 in some other implementations. Preprocessor 904 may, for example, process all or part of the 902 data, which may result in some of the 902 data being altered, replaced, deleted, etc. In some implementations, preprocessor 904 may add additional data to the 902 data. In some implementations, preprocessor 904 may be an ML model, such as an ANN.

[0091] The ANN 900 includes at least one first layer 908 of artificial neurons 910 to process input data 906 and provide data resulting from the first layer via edges 912 to at least a portion of at least a second layer 914. The second layer 914 processes data received via edges 912 and provides output data from the second layer via edges 916 to at least a portion of at least a third layer 918. The third layer 918 processes data received via edges 916 and provides output data from the third layer via edges 920 to at least a portion of a final layer 922 including one or more neurons to provide output data 924. All or part of the output data 924 may be further processed in some way by the (optional) postprocessor 926. Thus, in certain examples, the ANN 900 may provide output data 928 that are based on data from Output 924, postprocessor postprocessor output 926, or some combination thereof.The 926 postprocessor can be included within the ANN 900 in some other implementations. The 926 postprocessor can, for example, process all or part of the 924 output data, which may result in 928 output data being different, at least in part, from 924 output data, for example, as a result of data being changed, replaced, deleted, etc. In some implementations, the 926 postprocessor can be configured to add additional data to the output. Petition 870250083890, dated 09 / 17 / 2025, pp. 464 / 515 34 / 59 output data 924. In this example, the second layer 914 and the third layer 918 represent intermediate or hidden layers that may be arranged in a hierarchical or similar structure. Although not explicitly shown, there may be one or more additional intermediate layers between the second layer 914 and the third layer 918. In some implementations, the post-processor 926 may be an ML model, such as an ANN.

[0092] The ANN 900 or other ML models can be implemented in various types of processing circuits, along with the memory and instructions applicable thereto. For example, general-purpose hardware circuits, such as one or more central processing units (CPUs) and one or more graphics processing units (GPUs), can be employed to implement a model. One or more tensor processing units (TPUs), neural processing units (NPUs), or other special-purpose processors and / or field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), or similar can also or alternatively be employed. In some implementations, the ML model can be implemented by an NPU and / or a TPU embedded in a system-on-a-chip (SoC) along with other components, such as one or more CPUs and / or GPUs.A SoC includes several components manufactured on a shared semiconductor substrate. The NPU and / or TPU can be controlled by one or more CPUs by configuring the ML model implemented by the NPU and / or TPU with weights and / or biases, providing certain training data to the ML model to configure the ML model, and / or providing input data to the ML model to obtain determinations. The one or more CPUs can also receive the determinations and be configured to perform certain actions based on the determinations made by the ML model.

[0093] Aspects of signal processing described in Figure 3, Figure 4, Figures 5, 6, and / or 7 can be incorporated into example devices. Petition 870250083890, dated 09 / 17 / 2025, pages 465 / 515 35 / 59 as the example device in Figure 10, Figure 11, Figure 12, Figure 13 and Figure 14. Examples of such wearable devices include, without limitation, a headset device as further described with reference to Figure 10; a virtual reality, mixed reality or augmented reality headset as described with reference to Figure 11; augmented reality glasses as described with reference to Figure 12; a hearing aid as described with reference to Figure 13; or in-ear headphones as described with reference to Figure 14.

[0094] Figure 10 depicts an implementation 1000 of a headset device 1002, which may include the SoC 100 configured according to one of Figures 2, 3, 4, 5, 6, 7, and / or 8. The headset device 1002 includes microphone(s) 130 and speaker(s) 120. In the example illustrated in Figure 10, microphone 130A is positioned primarily to detect the speech of a person using the headset device 1002, and microphone 130B is positioned to detect ambient sound, such as another person's speech or other sounds. The components of the SoC 100, including the feedback canceller 206, are integrated into the headset device 1002 and depicted using dashed lines to indicate internal components that are generally not visible to a user of the headset device 1002.

[0095] In a particular example of operation, microphone 130B can detect sound in an environment around headset device 1002 and generate audio data representing the sound. The audio data can be provided to SoC 100, which can process the audio data in the time domain to generate a noise-suppressed output signal and can process the audio data in the frequency domain to generate (or update) time-domain filter coefficients that are subsequently applied to the received audio data. In this example, the time-domain filter coefficients determined through frequency-domain processing provide high-quality noise suppression, and since the time-domain filter coefficients are applied to the audio data Petition 870250083890, dated 09 / 17 / 2025, pages 466 / 515 36 / 59 received in the time domain, little or no latency is added using these time-domain filter coefficients to process the audio data. In this way, the 1002 headset device can provide high-quality, low-latency noise suppression.

[0096] Figure 11 depicts an implementation 1100, which includes a portable electronic device corresponding to a virtual reality, mixed reality, or augmented reality headset device 1102. The headset device 1102 includes the microphone(s) 130 and the speaker(s) 120. Additionally, the components of the SoC 100, including the feedback cancellation 206, are integrated into the headset device 1102. In a particular operating example, the microphone(s) 130 can detect sound in an environment around the headset device 1102 and generate audio data representing the sound. Audio data can be provided to the SoC 100, which can process the audio data in the time domain to generate a noise-suppressed output signal and can process the audio data in the frequency domain to generate (or update) time-domain filter coefficients that are subsequently applied to the received audio data.In this example, the time-domain filter coefficients determined through frequency-domain processing provide high-quality noise suppression, and since the time-domain filter coefficients are applied to receive time-domain audio data, little or no latency is added by using these time-domain filter coefficients to process the audio data. In this way, the 1102 headset device can provide high-quality, low-latency noise suppression.

[0097] Figure 12 depicts an implementation 1200, which corresponds to augmented reality or mixed reality glasses 1202. The glasses 1202 include a holographic projection unit 1204 configured to project visual data onto a lens surface 1206 or to reflect visual data off a lens surface 1206 and onto the user's retina. The glasses 1202 also include microphone(s) 130, speaker(s) 120 and SoC 100. Petition 870250083890, dated 09 / 17 / 2025, pp. 467 / 515 37 / 59

[0098] In a particular example of operation, microphone(s) 130 can detect sound in an environment around glasses 1202 and generate audio data representing the sound. The audio data can be provided to feedback cancellation 206, which can process the audio data in the time domain to generate a noise-suppressed output signal and can process the audio data in the frequency domain to generate (or update) time-domain filter coefficients that are applied to the subsequently received audio data. In this example, the time-domain filter coefficients determined through frequency-domain processing provide high-quality noise suppression, and since the time-domain filter coefficients are applied to receive time-domain audio data, little or no latency is added by using these time-domain filter coefficients to process the audio data.In this way, the 1202 glasses can provide high-quality, low-latency noise suppression.

[0099] In some implementations, the holographic projection unit 1204 may display information related to the sound detected by the microphone(s) 130. For example, the holographic projection unit 1204 may display a notification indicating that speech has been detected. In another specific example, the holographic projection unit 1204 may display a notification indicating a detected audio event. For example, the notification may be superimposed on the user's field of view at a particular position that coincides with the location of the sound source associated with the audio event.

[0100] Figure 13 is a diagram of an operable wearable device for performing latency noise suppression, according to some examples in this disclosure. In the example illustrated in Figure 13, the wearable device is a hearing aid device 1302. The hearing aid 1302 includes microphone(s) 130, speaker(s) 120, and SoC 100, which includes feedback cancellation 206. In the example illustrated in Figure 13, the hearing aid 1302 is shown, which is configured to be worn behind one of the user's ears, a Petition 870250083890, dated 09 / 17 / 2025, pages 468 / 515 38 / 59 portion 1308 configured to extend over the ear and a portion 1306 to be used in or near a user's ear canal. In other examples, the hearing aid 1302 has a different configuration or form factor. To illustrate, the hearing aid 1302 may be an in-the-ear device that does not include portion 1304 configured to be used behind an ear and portion 1308 configured to extend over the ear.

[0101] In a particular example of operation of the hearing aid device 1302, the microphone(s) 130 can detect sound in an environment around the hearing aid device 1302 and generate audio data representing the sound. The audio data can be provided to the audio components 940, which can process the audio data in the time domain to generate a noise-suppressed output signal and can process the audio data in the frequency domain to generate (or update) time-domain filter coefficients that are subsequently applied to the received audio data.In this example, the time-domain filter coefficients determined through frequency-domain processing provide high-quality noise suppression, and since the time-domain filter coefficients are applied to receive time-domain audio data, little or no latency is added by using these time-domain filter coefficients to process the audio data. In this way, the 1302 hearing aid can provide high-quality, low-latency noise suppression.

[0102] Figure 14 depicts an implementation 1400 that includes a portable electronic device corresponding to one or more in-ear headphones 1406 (for example, a first in-ear headphone 1402, a second in-ear headphone 1404, or both). Although in-ear headphones 1406 are described, it should be understood that the present technology can be applied to other in-ear or over-the-ear audio devices.

[0103] In the example illustrated in Figure 14, the first in-ear headphone 1402 includes a first microphone 1410A, as a high signal-to-noise ratio microphone. Petition 870250083890, dated 09 / 17 / 2025, pp. 469 / 515 39 / 59 positioned to capture the voice of a user of the first in-ear headphone 1402, one or more other microphones configured to detect ambient sounds and spatially distributed to support beamforming, illustrated as microphone(s) 1412A, an internal microphone 1414A adjacent to the user's ear canal (e.g., to assist with active noise cancellation) and a self-speech microphone 1416A, such as a bone conduction microphone configured to convert sound vibrations from the ear bone or skull of a user into an audio signal.

[0104] The second in-ear headset 1404 can be configured in a substantially similar manner to the first in-ear headset 1402. For example, the second in-ear headset may include a microphone 1410B positioned to capture the voice of a user of the second headset 1404, one or more other microphones 1412B configured to detect ambient sounds and spatially distributed to support beamforming, an internal microphone 1414B and an auto-speech microphone 1416B.

[0105] In some implementations, the in-ear headphones 1402, 1404 are configured to automatically switch between various modes of operation, such as a pass-through mode in which ambient sound is processed by the audio components 940 for output via speaker(s) 120, and a playback mode in which non-ambient sound (e.g., streaming audio corresponding to a telephone conversation, media playback, video game, etc.) is played through the speaker(s) 120. In other implementations, the in-ear headphones 1402, 1404 may support fewer modes or may support one or more other modes in place of, or in addition to, the modes described. In an illustrative example of operation in pass-through mode, one or more of the microphones 130 (for example, microphone(s) 1412a, 1412b) can detect sound in an environment around the in-ear headphones 1402, 1404 and generate audio data representing the sound.

[0106] Audio data can be provided for feedback cancellation. 206, which can process audio data using a machine learning model. Petition 870250083890, dated 09 / 17 / 2025, pages 470 / 515 40 / 59 in the time domain to generate a noise-suppressed output signal and can process audio data in the frequency domain to generate (or update) time-domain filter coefficients that are subsequently applied to received audio data. In this example, the time-domain filter coefficients determined through frequency-domain processing provide high-quality noise suppression, and since the time-domain filter coefficients are applied to the received audio data in the time domain, little or no latency is added by using these time-domain filter coefficients to process the audio data. In this way, the 1402, 1404 in-ear headphones can provide high-quality, low-latency noise suppression.

[0107] With reference to Figure 15, a diagram of an illustrative aspect of an operable 660 system for performing machine learning-based audio sub-band processing is shown. In one particular aspect, the system of Figures 1 to 14 may include one or more components of the 660 system. The audio frequency divider 142 generates sub-band audio data for audio data. For example, the audio frequency divider 142 processes audio data 617B to generate sub-band audio data 618AA associated with a first frequency sub-band and sub-band audio data 618AB associated with a second frequency sub-band. The sub-band audio data 618AA represents the first audio frequency sub-band captured by a microphone. The sub-band audio data 618AB represents the second audio frequency sub-band captured by a microphone.

[0108] Additionally, the audio frequency divider 142 processes audio data 627B to generate sub-band audio data 628AA associated with the first frequency sub-band and sub-band audio data 628AB associated with a second frequency sub-band. The sub-band audio data 628AA represents the first audio frequency sub-band captured by one of the Petition 870250083890, dated 09 / 17 / 2025, pp. 471 / 515 41 / 59 microphones. The 628AB sub-band audio data represents the second audio frequency subband captured by one of the microphones.

[0109] One or more audio sub-band enhancers 144 process audio sub-band data sets associated with a corresponding frequency sub-band to generate enhanced audio data from the frequency sub-band. For example, audio sub-band enhancer 144A processes audio sub-band data from the first frequency sub-band to generate enhanced audio data 135A. To illustrate, audio sub-band enhancer 144A processes audio sub-band data 618AA and audio sub-band data 628AA to generate enhanced audio data 135A. As another example, audio sub-band enhancer 144B processes audio sub-band data 618AB and audio sub-band data 628AB to generate enhanced audio data 135B.

[0110] With reference to Figure 16, a diagram 700 is shown of an illustrative aspect of the components of a system. In a particular aspect, diagram 700 includes an example of an illustrative optional embodiment of the enhanced sub-band audio generator 140 and an example of an illustrative optional embodiment of the combiner 148.

[0111] Audio sub-band enhancers 144 include a plurality of machine learning models (e.g., LSTMs) associated with the respective sub-bands. For example, an LSTM 704A coupled to an LSTM 706A and an LSTM 708a corresponds to an audio sub-band enhancer 144a associated with a first frequency sub-band. As another example, an LSTM 704B coupled to an LSTM 706B and an LSTM 708B corresponds to an audio sub-band enhancer 144B associated with a second frequency sub-band. Yet another example, an LSTM 704C coupled to an LSTM 706C and an LSTM 708C corresponds to an audio sub-band enhancer 144C associated with a third frequency sub-band. In a further example, an LSTM 704D coupled to an LSTM 706D and an LSTM 708D corresponds to an enhancer of Petition 870250083890, dated 09 / 17 / 2025, pp. 472 / 515 42 / 59 audio sub-band 144D associated with a fourth frequency sub-band. It should be understood that the enhanced sub-band audio generator 140 including audio sub-band enhancers 144 corresponding to four frequency sub-bands is provided as an illustrative example; in other examples, the enhanced sub-band audio generator 140 may include audio sub-band enhancers 144 associated with fewer than four frequency sub-bands or more than four frequency sub-bands.

[0112] The 148 combiner includes a 748A concatenation layer coupled to a fully connected 750A layer. The 148 combiner also includes a 748B concatenation layer coupled to a fully connected 750B layer. The 144A audio sub-band enhancer processes audio data (e.g., 118A sub-band audio data, 128A sub-band audio data, one or more additional sets of sub-band audio data, or a combination thereof) representing the first frequency sub-band to generate an output that is provided to each of the 706A and 708A LSTMs of the 144A audio sub-band enhancer.

[0113] Audio frequency divider 142 processes audio data 117 to generate sub-band audio data 118A, sub-band audio data 118B, sub-band audio data 118C, and sub-band audio data 118D corresponding to the first frequency sub-band, the second frequency sub-band, the third frequency sub-band, and the fourth frequency sub-band, respectively. Audio frequency divider 142 processes audio data 127 to generate sub-band audio data 128A, sub-band audio data 128B, sub-band audio data 128C, and sub-band audio data 128D corresponding to the first frequency sub-band, the second frequency sub-band, the third frequency sub-band, and the fourth frequency sub-band, respectively.

[0114] The 142 audio frequency divider provides sub-band audio data from a frequency sub-band to a corresponding LSTM 704. For example, Petition 870250083890, dated 09 / 17 / 2025, pp. 473 / 515 43 / 59 The audio frequency divider 142 provides the sub-band audio data 118A, the sub-band audio data 128A, or both, to the LSTM 704A. As another example, the audio frequency divider 142 provides the sub-band audio data 118B, the sub-band audio data 128B, or both, to the LSTM 704B. An output from an LSTM 704 is provided to the corresponding LSTM 706, the corresponding LSTM 708, or both. For example, an output from the LSTM 704A is provided to the LSTM 706A, the LSTM 708A, or both.

[0115] The outputs of the 706 LSTMs are provided to the concatenation layer. The outputs of the 748A and 708 LSTMs are provided to the 748B concatenation layer. For example, an output from the 706A LSTM is provided to the 748A concatenation layer, and an output from the 708A LSTM is provided to the 748B concatenation layer. In a particular aspect, the output of the 706A LSTM, the output of the 708A LSTM, or both, correspond to the enhanced sub-band audio data 136A.

[0116] The 748A concatenation layer concatenates outputs from the 706A LSTM, from LSTM 706B, LSTM 706C, LSTM 706D, one or more additional LSTMs, or a combination thereof, to generate the first concatenated audio data representing a frequency band. In one example, the frequency band includes the first frequency sub-band, the second frequency sub-band, the third frequency sub-band, the fourth frequency sub-band, one or more additional frequency sub-bands, or a combination thereof. The first concatenated audio data is processed by the fully connected layer 750A. The combiner 148 applies a sigmoid function 752 to an output of the fully connected layer 750A to generate mask values ​​764. For example, an output of the fully connected layer 750A includes a first count of values ​​(e.g., 257 integer values). Applying the sigmoid function 752 to the output of the fully connected layer 750A generates the first count of mask values ​​764 (e.g., 257 mask values).In a particular optional mode, a mask value is either a 0 or a 1. Petition 870250083890, dated 09 / 17 / 2025, pp. 474 / 515 44 / 59

[0117] Combinator 148 applies a delay of 740 to audio data 127 to generate delayed audio data 762. Combinator 148 includes a multiplier 754 that applies mask values ​​764 to delayed audio data 762 to generate masked audio data 766. For example, delayed audio data 762 includes the first count of values ​​(e.g., values ​​257), and applying mask values ​​764 to delayed audio data 762 includes applying a first mask value to a first value of delayed audio data 762 to generate a first value of masked audio data 766. In a particular optional embodiment, if the first mask value is 0, the first value of masked audio data 766 is 0. Alternatively, if the first mask value is 1, the first value of masked audio data 766 is the same as the first value of the data. delayed audio 762.The mask values ​​764 thus enable the selected values ​​from the delayed audio data 762 to be included in the masked audio data 766.

[0118] The 748B concatenation layer concatenates outputs from the 708A LSTM, from LSTM 708B, LSTM 708C, LSTM 708D, one or more additional LSTMs, or a combination thereof, to generate the second concatenated audio data representing the frequency band. The second concatenated audio data is processed by the fully connected layer 750B to generate audio data 768. The combiner 148 generates the enhanced audio data 135 based on a combination of the masked audio data 766 and the audio data 768.

[0119] In a particular optional embodiment, the model architectures of the 144 audio subband enhancers are based on the 346 subband enhancer data. In one example, each of the LSTMs of the 144A audio subband enhancer includes two hidden layers, each of the LSTMs of the 144B audio subband enhancer includes two hidden layers, each of the LSTMs of the 144C audio subband enhancer includes four hidden layers, and each of the LSTMs of the 144D audio subband enhancer includes four Petition 870250083890, dated 09 / 17 / 2025, pages 475 / 515 45 / 59 hidden layers. In one particular aspect, the 144 audio sub-band enhancers and the 148 combiner correspond to an SGN. In one particular aspect, the 144 audio sub-band enhancers include multiple LSTMs to generate enhanced audio data from the respective sub-bands, which are smaller as a group than a single LSTM that is configured to generate enhanced audio data for the entire frequency band.

[0120] It should be understood that the application of the 740 delay to audio data 127 to generate the delayed audio data 762 is provided as an illustrative example. In another example, the delay 740 can be applied to audio data 117, audio data 127, or a combination thereof, to generate delayed audio data 762.

[0121] With reference to Figure 17, an 800 diagram is shown of an illustrative aspect of the operations of a system component. In one particular aspect, the 800 diagram includes an example of an illustrative optional implementation of the enhanced sub-band audio generator 140 and an example of an illustrative optional implementation of the combiner 148.

[0122] In a particular optional embodiment, one or more of the 144 audio sub-band enhancers are configured to perform procedural signal processing. For example, the 144 audio sub-band enhancers include an audio sub-band enhancer 144E configured to use procedural signal processing to process fifth frequency sub-band audio data to generate 136E enhanced fifth frequency sub-band audio data.

[0123] In an illustrative example, audio frequency divider 142 processes audio data 117 to generate sub-band audio data 118E from a fifth frequency sub-band, in addition to generating sub-band audio data 118A, sub-band audio data 118B, sub-band audio data 118C, and sub-band audio data 118D. Audio frequency divider 142 processes audio data 127 to generate sub-band audio data 128E from the fifth sub-band. Petition 870250083890, dated 09 / 17 / 2025, pp. 476 / 515 46 / 59 frequency band, in addition to generating audio data for sub-bands 118A, 118B, 118C, and 118D. The audio frequency divider 142, which generates audio data associated with five frequency sub-bands, is provided as an illustrative example; in other examples, the audio frequency divider 142 may generate audio data associated with fewer than five or more than five frequency sub-bands.

[0124] The 144E audio sub-band enhancer applies procedural signal processing to the 118E sub-band audio data, the 128E sub-band audio data, or a combination thereof, to generate the enhanced 136E sub-band audio data. In an optional embodiment, the 144E audio sub-band enhancer applies procedural signal processing based on the voice activity information 810 from one or more of the 144A to 144D audio sub-band enhancers. In a particular aspect, the fifth frequency sub-band (e.g., 8 to 16 kHz) corresponds to a higher frequency range, and sub-band SGN processing (e.g., using generative networks such as LSTMs) is skipped for the higher frequency range because speech in the higher frequency range appears similar to noise to generative networks. In some optional modes, the 144E audio sub-band enhancer includes a machine learning model other than a generative network.

[0125] Combiner 148 generates audio data 864 based on a combination of masked audio data 766 and audio data 768. Audio data 864 is from a particular frequency sub-band (e.g., the first frequency sub-band, the second frequency sub-band, the third frequency sub-band, and the fourth frequency sub-band). Combiner 148 includes a concatenation layer 812 that concatenates audio data 864 and enhanced sub-band audio data 136E to generate enhanced audio data 135. Enhanced audio data 135 is from a frequency band (e.g., the particular frequency sub-band and the fifth frequency sub-band). Petition 870250083890, dated 09 / 17 / 2025, pp. 477 / 515 47 / 59

[0126] In one or more aspects, techniques for supporting signal processing may include additional aspects, such as any single aspect or any combination of aspects described below or in connection with one or more other processes or devices described elsewhere in the present invention. In a first aspect, the supporting signal processing may include an apparatus configured for feedback reduction. The apparatus is further configured to perform feedback reduction using a trained machine learning (ML) model that determines an output audio signal by matching an output from the ML model with a microphone signal to reduce the feedback components in the microphone signal.

[0127] Additionally, the apparatus may perform or operate in accordance with one or more aspects, as described below. In some implementations, the apparatus includes a wireless device, such as a UE. In some implementations, the apparatus includes a remote server, such as a cloud-based computing solution, which receives image data for processing to determine output image frames. In some implementations, the apparatus may include at least one processor, and a memory coupled to the processor. The processor may be configured to perform operations described in the present invention with respect to the apparatus. In some other implementations, the apparatus may include a non-transient computer-readable medium having program code recorded thereon, and the program code may be executable by a computer to cause the computer to perform operations described in the present invention with reference to the apparatus.In some implementations, the device may include one or more means configured to perform operations described in the present invention. In some implementations, a wireless communication method may include one or more operations described in the present invention with reference to the device.

[0128] In a second aspect, in combination with the first aspect, the device is additionally configured to receive an input audio signal, Petition 870250083890, dated 09 / 17 / 2025, pp. 478 / 515 48 / 59 wherein the input audio signal includes a desired audio component and a feedback component; and determine an output audio signal by applying a machine learning model to the input audio signal, in which the machine learning model is configured to reduce the feedback component.

[0129] In a third aspect, in combination with one or more of the first or second aspect, the machine learning model is configured to preserve the desired component and remove the feedback component.

[0130] In a fourth aspect, in combination with one or more of the first to third aspects, the apparatus additionally includes an amplification circuit coupled to one or more processors and configured to drive a transducer from the output audio signal.

[0131] In a fifth aspect, in combination with one or more of the first aspect through the fourth aspect, one or more processors are configured to reduce the feedback component by having the combination of the input audio signal with a cancellation signal generated by the machine learning model determine the output audio signal, and the machine learning model is configured to generate the cancellation signal to cancel nonlinearities created by the amplification circuit that amplifies the output audio signal.

[0132] In a sixth aspect, in combination with the one or more from the first to the fifth aspect, the one or more processors are additionally configured to determine a feedback cancellation signal to reduce linear components of the feedback component of the input audio signal; and combine the feedback cancellation signal with the input audio signal before determining the output audio signal by applying the machine learning model.

[0133] In a seventh aspect, in combination with one or more of the first to sixth aspects, the apparatus further comprises an amplification circuit coupled to one or more processors and configured to amplify Petition 870250083890, dated 09 / 17 / 2025, pp. 479 / 515 49 / 59 the output audio signal to drive an output audio signal transducer, where the machine learning model is configured to reduce the feedback component by reducing nonlinearities in the amplification circuit.

[0134] In an eighth aspect, in combination with one or more of the first to seventh aspects, the apparatus further comprises an additional amplification circuit coupled to one or more processors and configured to amplify the input audio signal after combining the feedback cancellation signal with the input audio signal and before reducing the feedback component by applying the machine learning model, and wherein the machine learning model is configured to reduce the feedback component by reducing nonlinearities of the additional amplification circuit.

[0135] In a ninth aspect, in combination with one or more of the first to eighth aspects, the machine learning model is configured to reduce the feedback component based on parameters related to the feedback cancellation signal.

[0136] In a tenth aspect, in combination with one or more of the first to ninth aspects, one or more processors comprise a digital signal processor configured to determine the feedback cancellation signal and to output parameters related to the feedback cancellation signal; and a neural signal processor configured to execute the machine learning model based on parameters related to the feedback cancellation signal.

[0137] In an eleventh aspect, in combination with the one or more from the first to the tenth aspect, the machine learning model is configured to reduce the feedback component based on the input parameters corresponding to the input from a sensor that is not correlated with the feedback component.

[0138] In a twelfth aspect, in combination with one or more of the first aspect through the eleventh aspect, the learning model of Petition 870250083890, dated 09 / 17 / 2025, pp. 480 / 515 The 50 / 59 machine is configured to reduce one or more artifacts resulting from the amplification circuit without reducing other howls in the input audio signal.

[0139] In a thirteenth aspect, in combination with one or more of the first to twelfth aspects, the one or more processors are configured to reduce the feedback component by applying the machine learning model comprising applying a time-domain filter to the input audio signal after amplifying the input audio signal, the time-domain filter being configured based on the machine learning model.

[0140] In a fourteenth aspect, in combination with one or more of the first to thirteenth aspects, the apparatus further comprises a first microphone coupled to the one or more processors, wherein the input audio signal is received from the first microphone; and a transducer coupled to the one or more processors, wherein the transducer is configured to reproduce the output audio signal.

[0141] In a fifteenth aspect, in combination with one or more of the first to fourteenth aspects, a method comprises receiving an input audio signal, wherein the input audio signal includes a desired audio component and a feedback component; and reducing the feedback component by applying a machine learning model to the input audio signal to determine an output audio signal.

[0142] In a sixteenth aspect, in combination with one or more of the first aspect through the fifteenth aspect, the machine learning model is configured to preserve the desired component and remove the feedback component.

[0143] In a seventeenth aspect, in combination with one or more of the first to sixteenth aspects, the method additionally includes amplifying the output audio signal to output to a transducer, wherein the reduction of the feedback component comprises combining the input audio signal with a cancellation signal generated by the machine learning model to Petition 870250083890, dated 09 / 17 / 2025, pp. 481 / 515 51 / 59 determine the output audio signal before amplifying the output audio signal and where the machine learning model is configured to generate the cancellation signal to cancel nonlinearities created by amplifying the output audio signal.

[0144] In an eighteenth aspect, in combination with one or more of the first to seventeenth aspects, the method additionally includes determining a feedback cancellation signal to reduce linear components of the feedback component of the input audio signal; combining the feedback cancellation signal with the input audio signal before reducing the feedback component by applying the machine learning model; and amplifying the output audio signal to drive a transducer of the output audio signal, wherein the machine learning model is configured to reduce the feedback component by reducing amplification nonlinearities of the output audio signal.

[0145] In a nineteenth aspect, in combination with one or more of the first to eighteenth aspects, the method further comprises: amplifying the input audio signal after combining the feedback cancellation signal with the input audio signal and before reducing the feedback component by applying the machine learning model, wherein the machine learning model is configured to reduce the feedback component by reducing amplification nonlinearities of the input audio signal.

[0146] In a twelfth aspect, in combination with one or more of the first aspect through the nineteenth aspect, the machine learning model is configured to reduce the feedback component based on input parameters related to the feedback cancellation signal.

[0147] In a twenty-first aspect, in combination with one or more of the first to twelfth aspects, the machine learning model is configured to reduce the feedback component based on Petition 870250083890, dated 09 / 17 / 2025, pages 482 / 515 52 / 59 input parameters corresponding to the input of a sensor not correlated with the feedback component.

[0148] In a twenty-second aspect, in combination with the one or more from the first aspect to the twenty-first aspect, amplification results in one or more artifacts resulting from the feedback component in the input audio signal, the one or more artifacts comprising howling, and the machine learning model is configured to reduce the one or more artifacts resulting from amplification without reducing other howling in the input audio signal.

[0149] In a twenty-third aspect, in combination with the one or more from the first to the twenty-second aspect, the method additionally includes amplifying the input audio signal, wherein the amplification results in one or more artifacts resulting from the feedback component in the input audio signal and wherein the machine learning model is configured to reduce the one or more artifacts.

[0150] In the figures, a single block can be described as performing one or more functions. The one or more functions performed by this block can be performed in a single component or in multiple components, and / or can be performed using hardware, software, or a combination of hardware and software. To clearly illustrate this interchangeability between hardware and software, several illustrative components, blocks, modules, circuits, and stages are described below, generally in terms of their functionality. The possibility of such functionality being implemented as hardware or software depends on the particular application and the design constraints imposed on the overall system. Those skilled in the art may implement the described functionality in various ways for each particular application, but such implementation decisions should not be interpreted as causing a deviation from the scope of this disclosure.Furthermore, example devices may include components different from those shown, including well-known components such as a processor, memory, and the like. Petition 870250083890, dated 09 / 17 / 2025, pp. 483 / 515 53 / 59

[0151] Unless specifically stated otherwise, as will become evident from the discussions that follow, it should be understood that, throughout this application, discussions using terms such as access, receive, send, use, select, determine, normalize, multiply, average, monitor, compare, apply, update, measure, derive, solve, generate, or similar terms refer to actions and processes of a computer system or similar electronic computing device that manipulates and transforms data represented as physical (electronic) quantities in the computer system's registers and memories into other data, similarly represented as physical quantities in the computer system's registers, memories, or other information storage, transmission, or display devices. The use of different terms referring to actions or processes of a computer system does not necessarily indicate different operations.For example, determining data can refer to generating data. As another example, determining data can refer to retrieving data.

[0152] The terms device and apparatus are not limited to one or a specific number of physical objects (such as a smartphone, a camera controller, a processing system, and so forth). As used in the present invention, a device can be any electronic device with one or more parts that can implement at least some portions of the disclosure. Although the description and examples in the present invention use the term device to describe various aspects of the disclosure, the term device is not limited to a specific configuration, type, or number of objects. As used in the present invention, an apparatus may include a device, or a portion of a device, to perform the operations described.

[0153] Certain components in a device or apparatus described as means for accessing, means for receiving, means for sending, means for using, means for selecting, means for determining, means for normalizing, means for multiplying, or other terms with similar names referring to one or more Petition 870250083890, dated 09 / 17 / 2025, pp. 484 / 515 54 / 59 operations on data, such as image data, may refer to a set of processing circuits (e.g., application-specific integrated circuits (ASICs), digital signal processors (DSPs), graphics processing units (GPUs), central processing units (CPUs)), computer vision processors (CVPs), or neural signal processors (NSPs)) configured to perform the aforementioned function through hardware, software, or a combination of hardware configured by software.

[0154] Those skilled in the art will understand that information and signals can be represented using any of a variety of different technologies and techniques. For example, data, instructions, commands, information, signals, bits, symbols, and chips that may be mentioned throughout the description above can be represented by voltages, currents, electromagnetic waves, magnetic particles or fields, optical particles or fields, or any combination thereof.

[0155] The components, functional blocks and modules described in the present invention in relation to the Figures mentioned above include processors, electronic devices, hardware devices, electronic components, logic circuits, memories, software code, firmware code, among other examples, or any combination thereof. Software should be interpreted broadly to mean instructions, instruction sets, code, code segments, program code, programs, subprograms, software modules, applications, software applications, software packages, routines, subroutines, objects, executable portions, execution threads, procedures and / or functions, among other examples, whether called software, firmware, middleware, microcode, hardware description language or otherwise.Furthermore, the attributes discussed in the present invention can be implemented by means of a set of specialized processor circuits, via executable instructions, or combinations thereof. Petition 870250083890, dated 09 / 17 / 2025, pages 485 / 515 55 / 59

[0156] Those skilled in the art know that one or more blocks (or operations) described with reference to Figure 3 can be combined with one or more blocks (or operations) described with reference to another figure. For example, one or more blocks (or operations) in Figure 3 can be combined with one or more blocks (or operations) in Figure 1 or Figure 2. As another example, one or more blocks associated with Figures 4 to 9 can be combined with one or more blocks (or operations) associated with Figures 1 to 3. As a further example, one or more blocks associated with Figures 15 to 17 can be combined with one or more blocks (or operations) associated with Figures 3 to 9. Yet another example, one or more blocks associated with Figures 10 to 14 can be combined with one or more blocks (or operations) associated with Figures 1 to 2 and / or Figures 3 to 9.

[0157] Those skilled in the art will further recognize that the various illustrative logic blocks, modules, circuits, and algorithm steps described in connection with the present disclosure in the present invention can be implemented in the form of electronic hardware, computer software, or combinations of both. To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, modules, circuits, and steps have been described above in terms of their functionality. The possibility of such functionality being implemented as hardware or software depends on the particular application and the design constraints imposed on the overall system. Those skilled in the art may implement the described functionality in various ways for each particular application, but such implementation decisions should not be interpreted as causing a deviation from the scope of the present disclosure.Those skilled in the art will also readily recognize that the order or combination of components, methods, or interactions described in the present invention are merely examples, and that the components, methods, or interactions of the various aspects of the present disclosure may be... Petition 870250083890, dated 09 / 17 / 2025, pp. 486 / 515 56 / 59 combined or carried out in ways other than those illustrated and described in the present invention.

[0158] Different illustrative logics, logic blocks, modules, circuits, and algorithm processes described in connection with the implementations disclosed in the present invention may be implemented in the form of electronic hardware, computer software, or combinations of both. The interchangeability of hardware and software has been described in general terms of functionality and illustrated in the various components, blocks, modules, circuits, and illustrative processes described above. Whether such functionality will be implemented in hardware or software depends on the particular application and the design limitations imposed on the system as a whole.

[0159] In one or more aspects, the operations described may be implemented in hardware, a set of digital electronic circuits, computer software, firmware, including the structures disclosed in this descriptive report and their structural equivalents, or in any combination thereof. Implementations of the subject matter described in this descriptive report may also be implemented as one or more computer programs, which are one or more modules of computer program instructions, encoded in computer storage media for execution by, or to control the operation of, a data processing device.

[0160] The operations of a method or algorithm disclosed in the present invention can be implemented in a processor-executable software module, which may reside on a computer-readable medium and be commercially available as a computer program product or software. Computer-readable media include both computer storage media and communication media, including any media that may be enabled to transfer a computer program from one place to another. Storage media may be any available media that can be accessed by a computer. By way of example, and not limitation, such computer-readable media Petition 870250083890, dated 09 / 17 / 2025, pages 487 / 515 57 / 59 Computers may include random-access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other means that can be used to store the desired program code in the form of instructions or data structures, and that can be accessed by a computer. Furthermore, any connection may be appropriately termed a computer-readable medium. Disks (disk and disc), as used in the present invention, include compact disc (CD compact disc), laser disc, optical disc, digital versatile disc (DVD), floppy disk, and Blu-ray disc, wherein disks typically reproduce data magnetically, and discs reproduce data optically by means of lasers.Combinations of the above items should also be included within the scope of computer-readable media.

[0161] Various modifications to the implementations described in this disclosure may become readily apparent to those skilled in the art, and the generic principles defined in the present invention may be applied to some other implementations without departing from the spirit or scope of this disclosure. Thus, the claims are not intended to be limited to the implementations shown in the present invention, but rather to be given the broadest scope consistent with the present disclosure, the principles and the innovative attributes disclosed in the present invention.

[0162] Additionally, a person skilled in the art will readily recognize that opposite terms such as superior and inferior, or front and back, or top and bottom, or forward and backward, or left and right are sometimes used for ease of description of figures and indicate relative positions corresponding to the orientation of the figure on a properly oriented page and Petition 870250083890, dated 09 / 17 / 2025, pp. 488 / 515 58 / 59 may not reflect the proper orientation of any device as implemented.

[0163] Certain attributes described in this descriptive report, in the context of separate implementations, may also be implemented in combination in a single implementation. Conversely, several attributes described in the context of a single implementation may also be implemented separately in multiple implementations, or in any suitable subcombination. Furthermore, although attributes may be described above as acting in certain combinations and even initially claimed in this way, one or more attributes of a claimed combination may, in some cases, be removed from the combination, and the claimed combination may be directed to a subcombination or a variation of a subcombination.

[0164] Similarly, although operations are depicted in the drawings in a particular order, this should not be understood as requiring that such operations be performed in the particular order depicted or in sequential order, or that all illustrated operations be performed, to achieve desirable results. Additionally, the drawings may schematically depict one or more example processes in the form of a flow diagram. However, other operations not depicted may be incorporated into the example processes that are schematically illustrated. For example, one or more additional operations may be performed before, after, simultaneously with, or between any of the illustrated operations. In certain circumstances, parallel or multitasking processing may be advantageous.Furthermore, the separation of various system components in the implementations described above should not be understood as a requirement for such separation in all implementations, and it should be understood that the program and system components described may, in general, be integrated into a single software product or provided as packages in multiple software products. Additionally, some other implementations are within the scope of the following claims. In some cases, the... Petition 870250083890, dated 09 / 17 / 2025, pages 489 / 515 59 / 59 actions mentioned in the claims can be performed in a different order and still achieve the desired results.

[0165] As used in the present invention, including in the claims, the term "or," when used in a list of two or more items, means that any one of the items mentioned may be used alone, or any combination of two or more of the items mentioned may be used. For example, if a composition is described as containing components A, B, or C, the composition may contain only A; only B; only C; A and B in combination; A and C in combination; B and C in combination; or A, B, and C in combination. Furthermore, as used in the present invention, including in the claims, "or," as used in a list of items preceded by "at least one of," indicates a disjunctive list such that, for example, a list of at least one of A, B, or C means A or B or C or AB or AC or BC or ABC (that is, A and B and C), or any of these in any combination thereof.

[0166] The term substantially is defined as largely, but not necessarily entirely, what is specified (and includes what is specified; for example, substantially 90 degrees includes 90 degrees and substantially parallel includes parallel), as understood by a person skilled in the art. In any disclosed implementations, the term substantially may be replaced by within [a percentage] of what is specified, where the percentage includes 0.1, 1.5, or 10 percent.

[0167] The preceding description of the disclosure is provided to enable any person skilled in the art to make or use the disclosure. Various modifications to the disclosure will be readily apparent to those skilled in the art, and the generic principles defined in the present invention can be applied to other variations without departing from the spirit or scope of the disclosure. Thus, the disclosure is not intended to be limited to the examples and designs described in the present invention, but rather to be given the broadest scope consistent with the innovative principles and attributes disclosed in the present invention. Petition 870250083890, dated 09 / 17 / 2025, pp. 490 / 515

Claims

1 / 5 CLAIMS 1. Apparatus characterized by comprising: a memory that stores processor-readable instructions; and one or more processors coupled to the memory, the one or more processors configured to: receive an input audio signal, wherein the input audio signal includes a desired audio component and a feedback component; and determine an output audio signal by applying a machine learning model to the input audio signal, wherein the machine learning model is configured to reduce the feedback component.

2. Device according to claim 1, characterized in that the machine learning model is configured to preserve a desired component.

3. Apparatus, according to claim 1, characterized by further comprising an amplification circuit coupled to one or more processors and configured to drive a transducer from the output audio signal, wherein: the one or more processors are configured to reduce the feedback component by having the combination of the input audio signal with a cancellation signal generated by the machine learning model determine the output audio signal, and the machine learning model is configured to generate the cancellation signal to cancel nonlinearities created by the amplification circuit that amplifies the output audio signal.

4. Apparatus, according to claim 1, characterized in that one or more processors are further configured to: determine a feedback cancellation signal to reduce linear components of the feedback component of the input audio signal; and combine the feedback cancellation signal with the input audio signal before determining the output audio signal by applying the machine learning model, and wherein the apparatus further comprises: an amplification circuit coupled to one or more processors and configured to amplify the output audio signal to drive a transducer of the output audio signal, wherein the machine learning model is configured to reduce the feedback component by reducing nonlinearities of the amplification circuit.

5. Apparatus, according to claim 4, characterized by further comprising: an additional amplification circuit coupled to one or more processors and configured to amplify the input audio signal after combining the feedback cancellation signal with the input audio signal and before reducing the feedback component by applying the machine learning model, and wherein the machine learning model is configured to reduce the feedback component by reducing nonlinearities of the additional amplification circuit.

6. Device, according to claim 4, characterized in that the machine learning model is configured to reduce the feedback component based on parameters related to the feedback cancellation signal.

7. Apparatus, according to claim 6, characterized in that one or more processors comprise: a digital signal processor configured to determine the feedback cancellation signal and to output parameters related to the feedback cancellation signal; and Petition 870250083890, dated 09 / 17 / 2025, p. 512 / 515 3 / 5 a neural signal processor configured to execute the machine learning model based on parameters related to the feedback cancellation signal.

8. Device according to claim 4, characterized in that the machine learning model is configured to reduce the feedback component based on input parameters corresponding to the input of a sensor uncorrelated with the feedback component.

9. Device according to claim 8, characterized in that the machine learning model is configured to reduce one or more artifacts resulting from the amplification circuit without reducing other howls in the input audio signal.

10. Device, according to claim 9, characterized in that one or more processors are configured to reduce the feedback component by applying a machine learning model comprising applying a time-domain filter to the input audio signal after amplifying the input audio signal, the time-domain filter being configured based on the machine learning model.

11. Apparatus, according to claim 1, characterized by further comprising: a first microphone coupled to one or more processors, wherein the input audio signal is received from the first microphone; and a transducer coupled to one or more processors, wherein the transducer is configured to reproduce the output audio signal.

12. A method characterized by understanding: receiving an input audio signal, wherein the input audio signal includes a desired audio component and a feedback component; and reducing the feedback component by applying a machine learning model to the input audio signal to determine an output audio signal. Petition 870250083890, dated 09 / 17 / 2025, pp. 513 / 515 4 / 5 13. Method, according to claim 12, characterized in that the machine learning model is configured to preserve a desired component.

14. A method according to claim 12, characterized by further comprising: amplifying the output audio signal to a transducer, wherein the reduction of the feedback component comprises combining the input audio signal with a cancellation signal generated by the machine learning model to determine the output audio signal before amplifying the output audio signal, and wherein the machine learning model is configured to generate the cancellation signal to cancel nonlinearities created by the amplification of the output audio signal.

15. A method according to claim 12, characterized by further comprising: determining a feedback cancellation signal to reduce linear components of the feedback component of the input audio signal; combining the feedback cancellation signal with the input audio signal before reducing the feedback component by applying the machine learning model; and amplifying the output audio signal to drive a transducer of the output audio signal, wherein the machine learning model is configured to reduce the feedback component by reducing amplification nonlinearities of the output audio signal.

16. Method according to claim 15, characterized by further comprising: amplifying the input audio signal after combining the feedback cancellation signal with the input audio signal and before reducing the feedback component by applying the machine learning model, Petition 870250083890, dated 09 / 17 / 2025, pp. 514 / 515 5 / 5, wherein the machine learning model is configured to reduce the feedback component by reducing amplification nonlinearities of the input audio signal.

17. A method according to claim 15, characterized in that the machine learning model is configured to reduce the feedback component based on input parameters related to the feedback cancellation signal.

18. A method according to claim 15, characterized in that the machine learning model is configured to reduce the feedback component based on input parameters corresponding to the input of a sensor that is uncorrelated with the feedback component.

19. A method according to claim 18, characterized in that: the amplification results in one or more artifacts resulting from the feedback component in the input audio signal, the one or more artifacts comprising howls, and the machine learning model is configured to reduce the one or more artifacts resulting from the amplification without reducing other howls in the input audio signal.

20. A method according to claim 12, characterized by further comprising: amplifying the input audio signal, wherein the amplification results in one or more artifacts resulting from the feedback component in the input audio signal, and wherein the machine learning model is configured to reduce the one or more artifacts.

21. Product, process, system, kit, means or use characterized by comprising one or more elements described in the descriptive report, claims, drawings, sequence listing or summary of this application, when applicable. Petition 870250083890, dated 17 / 09 / 2025, pp. 515 / 515

Images