Information simultaneous transmission system based on cloud storage platform

Abstract

The invention discloses an information simultaneous transmission system based on a cloud storage platform, and relates to the technical field of translation systems, a cloud analysis and preprocessing module calls computing resources in the cloud storage platform, and performs noise elimination, source text generation and semantic understanding on voice input data; the distributed translation engine module dynamically loads a multi-language translation model and a domain corpus based on the cloud storage platform, and converts the source text language content into a target text language; and the translation quality evaluation module evaluates the accuracy and availability of the translation result. When the evaluation result does not reach the standard, the system starts an optimization mechanism; and the synchronous output module distributes the optimized translation result to a plurality of terminals in real time through a low-delay channel. Through the elastic resource scheduling and multi-language cooperative processing technology of the cloud storage platform, the problems of unstable translation quality, poor professional field adaptation and the like are effectively solved, cross-language cooperative scenes such as online education and the like can be applied, and the translation accuracy and efficiency are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of translation systems, and particularly to an information simultaneous translation system based on a cloud storage platform. Background Art

[0002] With the acceleration of the globalization process, the demand for cross - language communication and real - time translation is increasing day by day. Especially in international conferences, online education, multinational enterprise collaboration and other occasions, the application of simultaneous interpretation technology is particularly important. Existing simultaneous interpretation systems mainly rely on traditional translation engines and local computing resources. However, due to the limitations of their computing power, the instability of translation quality, and the lack of adaptability to professional fields, the simultaneous translation system often has problems such as high latency, low accuracy, and unstable translation quality in real - time translation tasks with high concurrency and large data volumes.

[0003] Traditional simultaneous translation systems usually rely on a single local hardware and translation model for computing, which makes it difficult for their processing capabilities to cope with the increasing data volume and complex translation tasks. In practical applications, especially when dealing with professional field translations, the translation models of existing systems often lack sufficient domain knowledge, resulting in translation results that cannot meet the requirements of high precision and high accuracy. In addition, when facing multi - language translations, existing systems often need to pre - load language models or corpora and cannot flexibly and dynamically adjust translation strategies according to different scenarios.

[0004] On the other hand, since traditional simultaneous translation systems usually adopt a centralized computing mode, the load pressure on the translation engine is often large. Especially when facing a large number of user requests, the response speed and processing capabilities of the system are particularly insufficient, easily causing delays in translation results and affecting the user experience. In addition, due to the lack of an effective translation quality evaluation mechanism, the accuracy of translation results is often difficult to guarantee, resulting in users being unable to detect translation errors or inaccuracies in a timely manner.

[0005] To solve the above problems, it is particularly important to propose a simultaneous translation system based on a cloud storage platform. Through the elastic resource scheduling and multi - language collaborative processing of the cloud computing platform, the computing bottleneck and unstable translation quality faced by traditional systems are effectively solved. At the same time, cloud computing resources are dynamically allocated according to the real - time load situation, improving the processing capabilities and response speed of the system, and ensuring the efficiency and accuracy of simultaneous translation. Summary of the Invention

[0006] The object of the present invention is to provide an information simultaneous translation system based on a cloud storage platform. Through the dynamic scheduling of cloud resources and multi - language collaborative processing technology, the defects of existing simultaneous translation systems in unstable translation quality, poor adaptation to professional fields, unreasonable allocation of computing resources, etc. are overcome, thereby improving the accuracy, efficiency of translation and the response speed of the system, and being particularly applicable to cross - language collaboration scenarios such as international conferences and online education.

[0007] To achieve the above object, the present invention provides an information simultaneous transmission system based on a cloud storage platform, including the following main modules: a voice input module, a cloud parsing and preprocessing module, a distributed translation engine module, a translation quality evaluation module, an optimization and regulation module, a synchronous output module, and a user feedback module.

[0008] Voice input module: This module is used to receive the voice input of the speaker, convert the voice signal into digital data, and upload it to the cloud storage platform as the input signal for subsequent translation.

[0009] Cloud parsing and preprocessing module: This module calls computing resources in the cloud to perform preprocessing tasks such as noise elimination, source text generation, and semantic understanding on the uploaded voice data, improving the accuracy of speech recognition and the usability of subsequent translation.

[0010] Distributed translation engine module: This module dynamically loads multi-language translation models and domain-specific corpora based on the cloud storage platform, and performs translation processing on the generated source text content. This module can support multi-language switching and dynamically adjust translation strategies according to specific translation tasks to ensure the efficiency and professionalism of translation quality.

[0011] Translation quality evaluation module: The translation result will be evaluated in real time through a translation quality evaluation model. This model combines various data such as semantic consistency analysis, context coherence detection, and user feedback to comprehensively evaluate the accuracy and usability of the translation. When the evaluation result shows that the translation quality is lower than the set standard, an optimization mechanism is started for adjustment.

[0012] Optimization and regulation module: When the translation quality evaluation fails to meet the standard, the system will automatically start an optimization mechanism, including but not limited to: switching the translation engine, calling the backup corpus, or prompting the user to optimize the input content. If the requirements still cannot be met after multiple rounds of optimization, the speaker will be reminded to re-enter the voice.

[0013] Synchronous output module: This module synchronizes the optimized translation result to multiple terminals in real time through a low-latency channel. The output forms include the synchronous presentation of voice, text, and subtitles, ensuring that the translation content among all terminals is synchronized and consistent.

[0014] User feedback module: This module receives the user's feedback on the translation quality. The user provides feedback by means of scoring, error correction, etc. This data will be uploaded to the cloud platform in real time as an important basis for updating the quality analysis model.

[0015] Furthermore, during the preprocessing of the input data, the cloud parsing and preprocessing module eliminates noise, accents, and non-standard sentences in the voice input, improving the accuracy of speech recognition and the reliability of translation.

[0016] Furthermore, the calculation formula for denoising the speech signal is as follows:

[0017]

[0018] where S raw (f) is the frequency-domain representation of the original speech signal, θ(f) = μ * E[|S bg (f)|] is the dynamic threshold, μ is the adjustment coefficient adaptively adjusted by the cloud platform according to historical noise data, and S bg (f) is the background noise spectrum.

[0019] Furthermore, the determination condition for the segmentation point of semantic segmentation in semantic understanding is as follows:

[0020]

[0021] where e i is the semantic embedding vector of the i-th sentence, k is the size of the sliding window, is the similarity threshold dynamically adjusted by the cloud, and t is the current time step in the sliding window.

[0022] Furthermore, the working process of the distributed translation engine module includes: selecting a translation engine instance deployed on the public cloud or hybrid cloud from the cloud storage platform according to the current network load; performing chunk-based parallel translation on long texts and temporarily storing intermediate results in the in-memory database of the cloud platform; and recording the response time and resource occupancy rate of the engine in real time during translation for optimizing the dynamic scheduling strategy.

[0023] Furthermore, the translation quality evaluation module conducts analysis based on a quality analysis model, which comprehensively considers the accuracy of the translation result, the grammatical structure, the context consistency, and the effectiveness of user feedback. If the evaluation fails to meet the standard, an optimization mechanism is triggered to improve the translation quality.

[0024] Furthermore, the calculation formula of the quality analysis model of the translation quality evaluation module is as follows:

[0025]

[0026] where BLEU is the machine translation score based on n-gram, cos(V src , V trans ) is the cosine similarity between the source text and the target text word vectors, Delay is the total delay from cloud processing to terminal output, PacketLoss is the network packet loss rate, and α, β, γ, and δ are dynamic weight coefficients adaptively adjusted by the cloud platform according to historical data.

[0027] Furthermore, the calculation formulas for the weights α, β, γ, and δ are as follows:

[0028] α t+1 = α t + η * (BLEU user - BLEU model )

[0029] β t+1 = β t + η * (cos(V src , V trans ) user - cos(V src , V trans ) model )

[0030]

[0031] δ t+1 = δ t + η * (PacketLoss user - PacketLoss model )

[0032] where η is the step size for controlling the weight update of the learning rate, usually taking values from 0.01 to 0.1, BLEU is the n-gram based machine translation score, cos(V src , V trans ) is the cosine similarity between the source text and the translated text word vectors, Delay is the total delay from cloud processing to terminal output, and PacketLoss is the network packet loss rate; BLEU user , cos(V src , V trans ) user , Delay user , PacketLoss user are the actual values of user feedback, obtained through user ratings or corrected data; BLEU model , cos(V src , V trans ) model , Delay model , PacketLoss model are the predicted values of the model, generated by the translation quality evaluation module.

[0033] Furthermore, the optimization and control module includes the following optimization mechanisms: switching to a backup translation engine: when the current translation quality does not meet the requirements, the system automatically selects and calls an alternative translation engine on the cloud to ensure translation quality; calling a domain-specific corpus: according to different translation scenarios (including law, medicine, technology, and education), automatically switching to the corpus in the cloud storage platform corresponding to the translation scenario to improve the professionalism and accuracy of the translation; localized re-translation: when the network delay is high or packet loss is severe, triggering the edge node to perform localized translation, reducing cloud dependence and improving translation efficiency and quality.

[0034] Furthermore, the triggering rules of the tuning mechanism are as follows: the optimization strategy is triggered in order of priority (the priority is switching engines, calling corpora, and localization re-translation in descending order): if Q deviates from the threshold by less than or equal to 20%: prioritize calling the domain-specific corpus and retain the current translation engine; if Q deviates from the threshold by more than 20%: force switching to the backup translation engine and synchronously update the corpus. model Or if the packet loss rate PacketLoss>15%, cloud processing is skipped and localized re-translation at the edge node is directly triggered.

[0035] Furthermore, the synchronous output module supports simultaneous output on multiple terminal devices and ensures that the translation results remain synchronized between all terminal devices, supporting output modes including but not limited to audio playback, text display and subtitle presentation.

[0036] Furthermore, the synchronous output module supports multi-terminal collaboration, and the specific implementation method is: generate a global timing identifier through a cloud storage platform to ensure the playback synchronization of translated content on each terminal; when it is detected that the network delay of a terminal exceeds a threshold, the compressed text summary is sent to it first, and the complete data is resent after the network is restored.

[0037] Furthermore, the user feedback module includes a real-time scoring function and an error correction feedback function. Users score the translation results during the translation process and correct translation errors in the system through error correction feedback. The feedback data is updated in real time to the quality analysis model through the cloud storage platform.

[0038] Compared with the prior art, the advantages of the present invention are:

[0039] (1) Flexible resource scheduling of cloud storage platform: The present invention realizes dynamic scheduling of computing resources through cloud storage platform, and can automatically allocate resources according to real-time load. Compared with the local computing mode of traditional simultaneous interpretation system, cloud processing provides higher computing power, which can effectively cope with high concurrency and large-scale data processing, ensuring stable translation quality and faster response speed.

[0040] (2) Multi - language collaborative processing and enhanced domain adaptability: The distributed translation engine of the present invention can dynamically load multi - language translation models and domain - specific corpora, flexibly switch between different languages and domains (such as law, medicine, technology, etc.), and solve the problem of poor adaptability of traditional systems in the face of professional - field translation.

[0041] (3) Real - time translation quality assessment and optimization mechanism: The present invention monitors the translation quality in real - time through a quality assessment module, and starts an automatic optimization mechanism when the translation quality does not meet the standard through an optimization and regulation module, such as switching the translation engine, calling a backup corpus, or prompting the user to adjust the input content. Compared with the traditional simultaneous interpretation system lacking an effective quality assessment and correction mechanism, the present invention can ensure that the translation result is more accurate and meets the actual needs, avoiding the impact of translation errors on the meeting process. Brief Description of the Drawings

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0043] Figure 1 It is the system module diagram of the present invention;

[0044] Figure 2 It is the structure diagram of the cloud parsing and pre - processing module of the present invention;

[0045] Figure 3 It is the structure diagram of the optimization and regulation module of the present invention. Detailed Embodiments

[0046] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all of them. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0047] Embodiment 1: Please refer to Figure 1 As shown, the information simultaneous interpretation system based on a cloud storage platform in this embodiment includes a voice input module, a cloud parsing and pre - processing module, a distributed translation engine module, a translation quality assessment module, an optimization and regulation module, a synchronous output module, and a user feedback module;

[0048] Speech Input Module: This module is used to convert the speech input of the speaker into digital speech data, usually implemented through a microphone or other speech input devices. The speech data is sent to the Cloud Parsing and Preprocessing Module.

[0049] Cloud Parsing and Preprocessing Module: As shown in Figure 2 , this module calls computing resources in the cloud to perform preprocessing tasks such as noise cancellation, text generation, and semantic understanding on the uploaded speech data, improving the accuracy of speech recognition and the usability of subsequent translation.

[0050] Deploy high-performance servers on the cloud storage platform, with the functions of mass storage and high-speed networking. Pre-deploy quality assessment models and various translation engines, so as to dynamically load and schedule according to needs, and achieve real-time and efficient translation processing.

[0051] The collected audio signals often contain noise or other interferences, and preprocessing steps are required to improve the accuracy of speech recognition. The preprocessing process includes operations such as noise removal and audio enhancement to ensure the clarity and quality of the signal, thereby improving the accuracy of subsequent recognition and translation. The calculation formula for noise cancellation is:

[0052]

[0053] where S raw (f) is the frequency-domain representation of the original speech signal, θ(f) = μ * E[|S bg (f)|] is the dynamic threshold, μ is the adjustment coefficient adaptively adjusted by the cloud platform according to historical noise data, and S bg (f) is the background noise spectrum.

[0054] In speech recognition, it is first necessary to convert the audio signal into a mathematical feature representation. This process uses technologies such as Mel Frequency Cepstral Coefficients (MFCC) to convert the audio signal into a series of feature vectors, and then uses a trained acoustic model to map these feature vectors to text units at the phoneme or sub-word level. Commonly used acoustic models include Hidden Markov Model (HMM), Convolutional Neural Network (CNN), and Recurrent Neural Network (RNN);

[0055] In the decoding stage, the system uses the probability distribution generated by the acoustic model and, with the help of the language model, finds the most likely text sequence. This process is often implemented using technologies such as the Viterbi algorithm. However, the decoded text sequence may contain errors or unnatural parts, so it needs to be further optimized through post-processing steps such as spelling correction and grammar correction. Finally, the decoded text sequence is output as the text representation of the speaker's speech content.

[0056] The output text sequence also needs to be semantically understood. This part analyzes and understands the semantics of the text to obtain the speaker's intention and the content expressed, then splits the text into words or phrases, and determines the part of speech and basic attributes of each word. The judgment condition for the segmentation point of semantic segmentation is:

[0057]

[0058] where, e i is the semantic embedding vector of the i-th sentence, k is the size of the sliding window, is the similarity threshold dynamically adjusted by the cloud, and t is the current time step in the sliding window.

[0059] To further understand the subject-predicate-object relationship and modification relationship of the text, it is also necessary to analyze the sentence structure, determine the grammatical relationship and hierarchical structure between words. Mark each word in the sentence as different semantic roles, such as subject, action, object, etc., to help understand the logical structure of the sentence. For identifying specific entities in the text, such as person names, place names, organization names, etc., to obtain specific information, it is also necessary to analyze the dependency relationship between words to determine the relationship between each word and other words in the sentence.

[0060] In addition to the above steps, it is also necessary to convert the sentence into a semantic representation, model the relationship between words and sentence meanings, and then infer the speaker's intention and purpose based on the semantic representation. In some cases, sentiment analysis is also required to judge the sentiment color expressed by the sentence, so as to better understand the speaker's attitude.

[0061] Distributed translation engine module: This module dynamically loads multilingual translation models and domain-specific corpora based on a cloud storage platform, and performs translation processing on the source language content. This module can support multilingual switching, and dynamically adjusts translation strategies according to specific translation tasks to ensure the efficiency and professionalism of translation quality. The translation data is sent to the translation quality assessment module;

[0062] Before machine translation, it is necessary to perform necessary preprocessing on the source language text. This includes operations such as word segmentation, punctuation removal, and conversion to lowercase, etc., to provide more standardized input data for the translation model, thereby improving translation quality.

[0063] Feature extraction (for SMT): In statistical machine translation (SMT), it is necessary to convert the source language text into a feature vector representation. This process usually involves the application of a vocabulary, a phrase table, and a language model to capture the statistical characteristics of the text and prepare for subsequent translation.

[0064] For neural machine translation (NMT), the source language text is first encoded into a continuous vector representation. This step is achieved through an encoder in a recurrent neural network (RNN) or Transformer architecture, aiming to convert the input text into a form that can be understood by the machine for subsequent processing.

[0065] Decoding is the process of converting the feature vectors or encoded representations of the source language into target language text. In SMT, this process relies on a phrase translation table and a language model; while in NMT, the target language text is generated through a decoder. Regardless of the method, decoding is a crucial step in achieving the conversion from one language to another.

[0066] During the decoding process, the system gradually generates the text of the target language, which can be words, phrases, or sub-word units. The generated content initially forms the basis of the translation. When it comes to voice output, the synthesized voice data needs to be sent to a speaker or other audio playback devices through a suitable audio transmission method so that the listener can clearly hear the translation result. Finally, the translation module outputs the generated target language text as the translation result.

[0067] Translation quality assessment module: The translation result will be evaluated in real time through a quality assessment model. This model comprehensively evaluates the accuracy and usability of the translation. When the evaluation result shows that the translation quality is below the set standard, an optimization mechanism is initiated for adjustment.

[0068] The calculation formula of the quality analysis model of the translation quality assessment module is:

[0069]

[0070] where BLEU is the n-gram-based machine translation score, cos(V src ,V trans ) is the cosine similarity between the source text and the translated text word vectors, Delay is the total delay from cloud processing to terminal output, PacketLoss is the network packet loss rate, and α, β, γ, and δ are dynamic weight coefficients all greater than 0, which are adaptively adjusted by the cloud platform according to historical data.

[0071] The calculation formulas for the weights α, β, γ, and δ are:

[0072] α t+1 =α t +η*(BLEU user -BLEU model )

[0073] β t+1 =β t +η*(cos(V src ,V trans )user -cos(V src ,V trans ) model )

[0074]

[0075] δ t+1 =δ t +η*(PacketLoss user -PacketLoss model )

[0076] where η is the step size for controlling the weight update of the learning rate, usually taking values from 0.01 to 0.1, BLEU is the n-gram based machine translation score, cos(V src ,V trans ) is the cosine similarity between the source text and the target text word vectors, Delay is the total delay from cloud processing to terminal output, PacketLoss is the network packet loss rate; BLEU user , cos(V src ,V trans ) user , Delay user , PacketLoss user are the actual values of user feedback, obtained through user ratings or corrected data; BLEU model , cos(V src ,V trans ) model , Delay model , PacketLoss model are the predicted values of the model, generated by the translation quality evaluation module.

[0077] Optimization and regulation module: As shown in Figure 3 , when the translation quality evaluation fails to meet the standard, the system will automatically activate the optimization mechanism, including but not limited to: switching the translation engine, calling the backup corpus, or prompting the user to optimize the input content. If the requirements still cannot be met after multiple rounds of optimization, the speaker will be reminded to re-enter the speech.

[0078] The specific rules for triggering the optimization strategy are: trigger in order of priority (from highest to lowest: switching the engine, calling the corpus, local retranslation). After obtaining the quality coefficient Q, if the deviation of Q from the threshold is less than or equal to 20%: preferentially call the domain-specific corpus and retain the current translation engine. If the deviation of Q from the threshold is greater than 20%: forcefully switch to the backup translation engine and synchronously update the corpus. If the network delay Delay > Delay model or the packet loss rate PacketLoss > 15%, skip cloud processing and directly trigger local retranslation at the edge node.

[0079] Synchronous output module: This module synchronizes the optimized translation results to multiple terminals in real time through a low-latency channel. The output forms include the synchronous presentation of voice, text, and subtitles, ensuring that the translation content among all terminals is synchronized and consistent.

[0080] Convert the translated target language text into voice, which is achieved through text-to-speech synthesis technology. Analyze the target language text, extract information such as content, tone, and emotion to help determine the appropriate pronunciation, intonation, and speech rate. Then, select a suitable speech synthesis engine to generate natural and fluent speech according to different technologies and models, where the models are driven by statistics or neural networks.

[0081] The system also provides a user-friendly operation interface to help users operate and control the system functions, through methods such as graphical interfaces and voice interactions. The visualization interface and interaction functions are realized, and users interact with the system by clicking, filling in text, etc. At the same time, users can also interact with the system through voice commands, and the system recognizes the intention and executes the corresponding operations.

[0082] The system allows users to control various functions of the system, such as starting voice recognition and adjusting the volume. In a multilingual environment, it supports users to select the source language and the target language.

[0083] In the present invention, the cloud parsing and preprocessing module performs noise elimination, text generation, and semantic segmentation on the input data to ensure the accuracy of voice recognition and the reliability of translation; the distributed translation engine module dynamically loads multilingual translation models and domain corpora based on the cloud storage platform to convert the source text language content into the target language, supporting multilingual switching and professional field adaptation; the translation quality evaluation module combines semantic consistency analysis, context coherence detection, and user feedback to evaluate the accuracy and usability of the translation results in real time; when the evaluation result does not meet the standard, the optimization and regulation module starts the following optimization mechanisms: switching to the backup translation engine, calling the domain-specific corpus, or performing localized retranslation through edge nodes to improve the translation quality; the synchronous output module distributes the optimized translation results to multiple terminals in real time through a low-latency channel, supporting the synchronous presentation of voice, text, and subtitles; the user feedback module receives the user's score and correction instructions for the translation quality, and stores the feedback data in the cloud platform to update the quality analysis model, forming a closed-loop optimization mechanism. When this translation system is used for video conference or online education translation, it can evaluate the translation quality in real time and make corresponding processing, effectively ensuring the translation accuracy and thus ensuring the stable progress of the conference.

[0084] The above formulas are all in dimensionless form and only use numerical values for calculation. These formulas are obtained based on a large amount of data and through software simulation, aiming to be as close to the actual situation as possible. The preset parameters in the formulas can be adjusted by those skilled in the art according to specific requirements.

[0085] In the description of this specification, the descriptions referring to terms such as "one embodiment", "example", "specific example", etc. mean that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in a suitable manner in any one or more embodiments or examples.

[0086] The preferred embodiments of the present invention disclosed above are only used to help illustrate the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the present invention to only the specific implementation manners. Obviously, many modifications and variations can be made according to the content of this specification. These embodiments are selected and specifically described in this specification in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can well understand and utilize the present invention. The present invention is only limited by the claims and their full scope and equivalents.

Claims

1. An information simultaneous interpretation system based on a cloud storage platform, characterized in that: Includes the following modules: Voice input module: used to receive voice input and upload input data to the cloud storage platform; Cloud parsing and preprocessing module: calls computing resources in the cloud storage platform to eliminate noise, generate source text, and understand semantics of speech input data; Distributed translation engine module: dynamically loads multilingual translation models and domain corpora based on the cloud storage platform, and converts the processed source text language into the target text language; Translation quality assessment module: evaluates the semantic consistency, contextual coherence and network transmission stability of translation results through historical translation data stored in the cloud and dynamically updated quality analysis models; Optimization and control module: When the translation quality is lower than the threshold, at least one of the following optimization mechanisms is triggered: switching to a backup translation engine in the cloud, calling a domain-specific corpus in the cloud storage platform, or performing localized re-translation through edge nodes; Synchronous output module: distributes the optimized translation results to the terminal device through a low-latency channel, and supports the synchronous presentation of voice, text and subtitles; User feedback module: Receives user ratings and correction instructions on translation quality, and stores feedback data in the cloud platform to update the quality analysis model.

2. According to claim 1, the information simultaneous interpretation system based on the cloud storage platform is characterized in that: The cloud-based parsing and preprocessing module removes noise, stress and irregular sentences in the voice input during the preprocessing of the input voice data.

3. According to claim 2, the information simultaneous interpretation system based on the cloud storage platform is characterized in that: The calculation formula for the noise elimination is: Among them, S raw (f) is the frequency domain representation of the original speech signal, θ(f) = μ*Ε[|S bg (f)|] is the dynamic threshold, μ is the adjustment coefficient of the cloud platform based on historical noise data, S bg (f) is the background noise spectrum.

4. The information simultaneous interpretation system based on a cloud storage platform according to claim 3, characterized in that: The segmentation point determination condition of the semantic segmentation in the semantic understanding is: Among them, e i is the semantic embedding vector of the i-th sentence, k is the sliding window size, is the similarity threshold dynamically adjusted by the cloud, and t is the current time step in the sliding window.

5. The system according to claim 1, characterized in that The workflow of the distributed translation engine module includes: selecting a translation engine instance deployed in a public cloud or hybrid cloud from a cloud storage platform according to the current network load; adopting block-by-block parallel translation for long texts, and temporarily storing intermediate results through the cloud platform's memory database; and recording the engine's response time and resource occupancy rate in real time during the translation process.

6. The information simultaneous interpretation system based on a cloud storage platform according to claim 1, characterized in that: The translation quality assessment module performs analysis based on a quality analysis model, which comprehensively assesses the accuracy, grammatical structure, contextual consistency and effectiveness of user feedback of the translation results. If the assessment does not meet the standards, an optimization mechanism is triggered to improve the translation quality.

7. The system according to claim 6, characterized in that The calculation formula of the quality analysis model of the translation quality assessment module is: Among them, BLEU is the machine translation score based on n-gram, cos(V src ,V trans ) is the cosine similarity between the original and translated word vectors, Delay is the total delay from cloud processing to terminal output, PacketLoss is the network packet loss rate, α, β, γ and δ are dynamic weight coefficients, which are adaptively adjusted by the cloud platform based on historical data.

8. The information simultaneous interpretation system based on the cloud storage platform according to claim 7, characterized in that: The calculation formula of the weights α, β, γ and δ is: alpha t+1 =α t +η*(BLUE user -BLUE model ) b t+1 =b t +η*cos(V src ,V trans ) user -cos(V src ,V trans ) model δ t+1 =δ t +η*(PacketLoss user -Packet Loss model ) Where η is the step size of the learning rate to control the weight update, which ranges from 0.01 to 0.1, BLEU is the machine translation score based on n-gram, cos(V src ,V trans ) is the cosine similarity between the original and translated word vectors, Delay is the total delay from cloud processing to terminal output, PacketLoss is the network packet loss rate; BLEU user , cos(V src ,V trans ) user , Delay user ,PacketLoss user Feedback actual value to users, obtained through user ratings or correction data; BLEU model , cos(V src ,V trans ) model , Delay model ,PacketLoss model is the model prediction value, generated by the translation quality assessment module.

9. The information simultaneous interpretation system based on a cloud storage platform according to claim 1, characterized in that: The optimization control module includes the following optimization mechanisms: Switching to an alternative translation engine: When the current translation quality does not meet the requirements, the system automatically selects and calls an alternative translation engine on the cloud to ensure the translation quality; Calling domain-specific corpora: According to different translation scenarios, automatically switch to the corpus corresponding to the translation scenario in the cloud storage platform to improve the professionalism and accuracy of the translation; Localized re-translation: When the network delay is high or the packet loss is serious, trigger the edge node for localized translation, reduce cloud dependence and improve translation efficiency and quality.

10. The information simultaneous interpretation system based on a cloud storage platform according to claim 1, characterized in that: The synchronous output module supports simultaneous output on multiple terminal devices and ensures that the translation results are synchronized between all terminal devices, supporting output modes including but not limited to audio playback, text display and subtitle presentation; The user feedback module includes real-time scoring and error correction feedback functions. Users score the translation results during the translation process and correct translation errors in the system through error correction feedback. The feedback data is updated in real time to the quality analysis model through the cloud storage platform.

Images