Adaptive speech recognition switching methods, devices, storage media, and software products

Abstract

This disclosure provides an adaptive speech recognition switching method, apparatus, storage medium, and program product. The method includes: responding to a user's acquisition command by recording the user's speech to obtain speech samples; testing multiple preset speech recognition models based on the speech samples to obtain a comprehensive test score for each model; determining the speech recognition model with the highest comprehensive test score as the current model and deploying it to the user's voice call system, while designating the remaining speech recognition models as candidate models; monitoring the translation level of the current model in real time to obtain its comprehensive performance score; and replacing the current model with the candidate model that has the best historical score when the comprehensive performance score indicates an anomaly. This method can select and switch speech recognition models from different vendors through quantitative evaluation of comprehensive test scores and performance scores, ensuring the stability and reliability of voice calls.

Description

Technical Field

[0001] This disclosure relates to the field of natural language processing technology, and in particular to an adaptive speech recognition switching method, apparatus, storage medium, and program product. Background Technology

[0002] With the development of internet technology, intelligent call center systems are rapidly emerging as a key component of modern customer service. The technical implementation of intelligent call centers typically involves an Automatic Speech Recognition (ASR) model converting the user's voice input into text data, a Large Language Model (LLM) generating response content based on the text data, and then text-to-speech (TTS) technology converting the text back into audio for final playback to the user. In this process, the speed and quality of each step directly impact the overall call quality and user experience.

[0003] However, when connecting with ASR providers, the stability of their services and the translation effect in specific scenarios cannot be guaranteed. If the translation speed of ASR slows down or the translation quality is poor, it will directly affect the response content of the large language model, thereby affecting the quality of language communication and user experience. Summary of the Invention

[0004] In view of this, the present disclosure provides an adaptive speech recognition switching method, apparatus, storage medium, and program product, which can selectively deploy speech recognition models from different vendors into a voice call system through quantitative evaluation of comprehensive test scores and performance scores, and realize the function of abnormal switching in actual voice calls, thereby ensuring the stability and reliability of voice calls.

[0005] In a first aspect, embodiments of this disclosure provide an adaptive speech recognition switching method, employing the following technical solution:

[0006] In response to the user's collection command, the system records the user's voice and obtains voice samples.

[0007] Based on the speech samples, various preset speech recognition models are tested to obtain a comprehensive test score for each speech recognition model.

[0008] The speech recognition model with the highest comprehensive test score is determined as the current model and deployed to the user's voice call system, while the remaining speech recognition models are determined as candidate models.

[0009] Real-time monitoring of the translation level of the current model to obtain the overall performance score of the current model;

[0010] Determine whether the current model has any anomalies based on the comprehensive performance score;

[0011] If so, the current model will be replaced by the candidate model with the best historical score.

[0012] Optionally, the step of recording the user's voice and obtaining voice samples in response to the user's collection command includes:

[0013] Build a text template library;

[0014] Receive several selection requests from the user;

[0015] In response to each selection request, several target templates are selected from the text template library;

[0016] Configure a data collection button for each target template;

[0017] When the user triggers the data collection button on the front-end interface, a corresponding data collection command is generated;

[0018] In response to the acquisition command, the audio capture device is activated;

[0019] The audio capture device records the user's speech and generates speech samples.

[0020] Optionally, the step of testing multiple preset speech recognition models based on the speech samples to obtain a comprehensive test score for each speech recognition model includes:

[0021] The speech sample is divided into multiple speech segments;

[0022] The speech recognition model was tested multiple times using the multiple speech segments to obtain the single-word translation time, misspelling rate and fluency parameters of each speech recognition model in each test.

[0023] Based on the single-character translation time, obtain the time level score for each speech recognition model;

[0024] Based on the typo percentage and the fluency parameter, the test quality parameters for each speech recognition model are obtained;

[0025] Based on the single-character translation time, the percentage of misspelled words, and the fluency parameter, the learning ability parameters of each speech recognition model are obtained;

[0026] Based on the duration level score, the test quality parameter, and the learning ability parameter, a comprehensive test score is obtained for each speech recognition model.

[0027] Optionally, the step of obtaining the learning ability parameters for each speech recognition model based on the single-character translation time, the proportion of misspelled words, and the fluency parameter includes:

[0028] Based on the single-character translation time, the misspelling rate, and the fluency parameter, a time analysis graph, a misspelling analysis graph, and a semantic analysis graph are constructed for each speech recognition model.

[0029] Based on the duration analysis graph, obtain the speed improvement parameters for each speech recognition model;

[0030] Based on the misspelling analysis graph, the accuracy improvement parameters for each speech recognition model are obtained;

[0031] Based on the semantic analysis graph, semantic quality improvement parameters for each speech recognition model are obtained;

[0032] Based on the speed improvement parameter, the accuracy improvement parameter, and the semantic quality improvement parameter, the learning ability parameters of each speech recognition model are obtained.

[0033] Optionally, the real-time monitoring of the translation level of the current model and the acquisition of the comprehensive performance score of the current model include:

[0034] Monitor the performance of the current model in the current voice call, and obtain the unit processing time, typo parameters, and logical parameters;

[0035] Based on the unit processing time, obtain the timeliness score of the current model;

[0036] Based on the misspelling parameters and the logical parameters, the usage quality parameters of the current model are obtained;

[0037] Based on the unit processing time, the misspelling parameters, and the logical parameters, obtain the performance improvement parameters of the current model;

[0038] Based on the timeliness score, the usage quality parameters, and the performance improvement parameters, the performance of the current model in the current voice call is evaluated to obtain a comprehensive performance score.

[0039] Optionally, determining whether the current model is abnormal based on the comprehensive performance score includes:

[0040] Determine whether the comprehensive performance score obtained by the current model is less than a first preset threshold for N consecutive times, or whether a preset proportion of the comprehensive performance score is less than a second preset threshold within a first preset time range.

[0041] If so, then the current model is determined to be abnormal;

[0042] If not, then the current model is considered to be operating normally.

[0043] Optionally, replacing the current model with the candidate model that has the best historical score includes:

[0044] Obtain the sum of the evaluation scores for each alternative model in M ​​consecutive performance analyses, wherein the evaluation scores include the comprehensive test score and / or the comprehensive performance score;

[0045] The candidate model with the highest total score is determined as the best historical score and replaces the current model.

[0046] Secondly, this disclosure also provides an adaptive speech recognition switching system, which adopts the following technical solution:

[0047] The response module is used to respond to the user's collection command, record the user's voice, and obtain voice samples;

[0048] The testing module is used to test multiple preset speech recognition models based on the speech samples and obtain a comprehensive test score for each speech recognition model.

[0049] The deployment module is used to determine the speech recognition model with the highest comprehensive test score as the current model and deploy it to the user's voice call system, and to determine the remaining speech recognition models as candidate models.

[0050] The monitoring module is used to monitor the translation level of the current model in real time and obtain the comprehensive performance score of the current model.

[0051] The judgment module is used to determine whether the current model has any abnormalities based on the comprehensive performance score; if so, the replacement module is executed.

[0052] The replacement module is used to replace the current model with the candidate model that has the best historical score.

[0053] Thirdly, this disclosure also provides a computer device, which adopts the following technical solution:

[0054] The computer device includes:

[0055] At least one processor; and,

[0056] A memory communicatively connected to the at least one processor; wherein,

[0057] The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform any of the adaptive speech recognition switching methods described above.

[0058] Fourthly, embodiments of this disclosure also provide a computer-readable storage medium storing computer instructions for causing a computer to perform any of the adaptive speech recognition switching methods described above.

[0059] Fifthly, embodiments of this disclosure also provide a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of any of the methods described above.

[0060] The adaptive speech recognition switching method provided in this disclosure involves the system recording the user's speech and acquiring speech samples after the user issues a collection command. This provides a data foundation for subsequent model testing and selection. The acquired speech samples are used to test various preset speech recognition models, and a comprehensive test score is calculated for each model. Based on the comprehensive test score, the best-performing model is selected as the current model and deployed to the user's voice call system. Simultaneously, other models are reserved as backup models for unforeseen circumstances. This method of testing and selecting models using the user's own speech samples better adapts to the different speech characteristics of users, providing more personalized and accurate speech recognition services. The system monitors the translation level of the current model in real time and calculates its comprehensive performance score, enabling timely detection of potential performance degradation or anomalies. If the comprehensive performance score of the current model does not meet expectations, indicating a possible anomaly, the system automatically switches to the backup model with the best historical score to ensure service continuity and quality. This method, by monitoring the performance of the current model in real time and automatically switching to a backup model when anomalies are detected, ensures the stability and reliability of the voice call system, reduces service interruptions caused by model failures, and increases the system's flexibility and ability to respond to emergencies. In summary, this method, through quantitative evaluation of comprehensive test scores and performance scores, makes the selection and switching process of models from different vendors more objective and efficient, reducing the need for manual intervention. Furthermore, it can dynamically adjust the model used based on the user's specific usage, achieving model adaptation and continuous optimization, thereby improving the accuracy of speech recognition and user satisfaction.

[0061] The above description is merely an overview of the technical solution disclosed herein. In order to better understand the technical means of this disclosure and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this disclosure more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0062] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0063] Figure 1A flowchart illustrating the adaptive speech recognition switching method provided in this embodiment of the disclosure;

[0064] Figure 2 A flowchart illustrating the method for obtaining comprehensive test scores provided in this embodiment of the disclosure;

[0065] Figure 3 A flowchart illustrating the method for obtaining learning ability parameters provided in this embodiment of the disclosure;

[0066] Figure 4 A schematic diagram of the adaptive speech recognition switching system provided in the embodiments of this disclosure;

[0067] Figure 5 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present disclosure. Detailed Implementation

[0068] The embodiments of this disclosure will now be described in detail with reference to the accompanying drawings.

[0069] It should be understood that the following specific examples illustrate the implementation of this disclosure, and those skilled in the art can easily understand other advantages and effects of this disclosure from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. This disclosure can also be implemented or applied through other different specific implementation methods, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this disclosure. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0070] It should be noted that various aspects of embodiments within the scope of the appended claims are described below. It will be apparent that the aspects described herein can be embodied in a wide variety of forms, and any particular structure and / or function described herein is merely illustrative. Based on this disclosure, those skilled in the art will understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects set forth herein can be used to implement the device and / or practice the method. Additionally, this device and / or method can be implemented using structures and / or functionalities other than one or more of the aspects set forth herein.

[0071] It should also be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of this disclosure. The drawings only show the components related to this disclosure and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0072] Furthermore, specific details are provided in the following description to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that the described aspects can be practiced without these specific details.

[0073] Reference Figure 1 This disclosure provides an adaptive speech recognition switching method, including the following steps:

[0074] S1: In response to the user's collection command, record the user's voice and obtain voice samples;

[0075] S2: Based on speech samples, test multiple preset speech recognition models and obtain the comprehensive test score for each speech recognition model;

[0076] S3: The speech recognition model with the highest comprehensive test score is determined as the current model and deployed to the user's voice call system, while the remaining speech recognition models are determined as candidate models;

[0077] S4: Monitor the translation level of the current model in real time and obtain the overall performance score of the current model;

[0078] S5: Determine whether the current model has any anomalies based on the overall performance score; if yes, proceed to S6; otherwise, return to S4.

[0079] S6: Replace the current model with the candidate model that has the best historical score;

[0080] The adaptive speech recognition switching method disclosed herein involves the system recording the user's speech and acquiring speech samples after the user issues a collection command. This provides a data foundation for subsequent model testing and selection. The acquired speech samples are used to test various preset speech recognition models, and a comprehensive test score is calculated for each model. Based on the comprehensive test score, the best-performing model is selected as the current model and deployed to the user's voice call system. Simultaneously, other models are reserved as backup models for unforeseen circumstances. This method of testing and selecting models using the user's own speech samples better adapts to the different speech characteristics of users, providing more personalized and accurate speech recognition services.

[0081] The system monitors the translation performance of the current model in real time and calculates its overall performance score, enabling it to promptly detect potential performance degradation or anomalies. If the overall performance score of the current model does not meet expectations, indicating a possible anomaly, the system automatically switches to the backup model with the best historical score to ensure service continuity and quality. This method, by monitoring the performance of the current model in real time and automatically switching to the backup model when anomalies are detected, ensures the stability and reliability of the voice call system, reduces service interruptions caused by model failures, and increases the system's flexibility and ability to respond to emergencies.

[0082] In summary, this method, through quantitative evaluation of comprehensive test scores and performance scores, makes the selection and switching process of models from different vendors more objective and efficient, reducing the need for manual intervention. Furthermore, it can dynamically adjust the model used based on the user's specific usage, achieving model adaptation and continuous optimization, thereby improving the accuracy of speech recognition and user satisfaction.

[0083] In S1, a text template library is built, integrating various preset text templates. These templates are tagged according to different application scenarios, such as news reports and literary works, to cover diverse voice call needs. Each preset text template is displayed to the user on the front-end interface, and the user selects one or more target text templates based on their needs. Users can initiate selection requests by clicking on preset text templates on the front-end interface or using voice commands. The back-end receives these selection requests and filters the corresponding target template from the text template library based on the user's selection.

[0084] Each target template is designed and configured with a dedicated acquisition button, which is intuitively displayed on the front-end interface for easy user access and operation. When a user triggers any acquisition button on the front-end interface via mouse click or touchscreen operation, the system immediately recognizes the action and generates an acquisition command corresponding to that template. This command contains all necessary information, such as template identifier and user ID, to ensure correct association and processing of subsequent voice data.

[0085] In response to the capture command, the system automatically activates the audio capture device, such as a built-in or external microphone. During startup, it checks the device's connection status and permission settings to ensure the audio capture device is functioning correctly. Additionally, the system automatically adjusts the microphone sensitivity based on the current ambient noise level to optimize recording quality. Once the audio capture device is ready, the user reads the text from the target template, and the system automatically saves the recorded user speech as a voice sample. When the user selects multiple target templates, the user needs to click the corresponding capture button sequentially and read the text from the target template. The system will record each reading and save it as multiple independent voice samples.

[0086] By identifying target templates that meet user needs and recording corresponding voice samples, speech recognition models from different vendors (i.e., different types of speech recognition models (ASR)) can be tested in subsequent stages. Furthermore, during the testing process, the voice samples can also be used to train the speech recognition model, enabling it to learn and adapt to the user's individual language characteristics and habits, thereby providing translation services that better meet the user's personalized needs.

[0087] In S2, a unified interface is designed to integrate speech recognition models from different vendors, ensuring standardized input parameters and storage, including but not limited to parameter types, parameter meanings, return value types, return value meanings, and audio codecs. Since commonly used softswitches (such as Freeswitch) in voice calls typically employ PCMU encoding, which some ASR vendors may not support, the system also needs to perform audio transcoding adaptation to ensure compatibility with various ASR products. To track and optimize ASR product performance, vendor tagging is performed for each test and actual voice call to clearly record and identify the ASR vendor used in the current test and call. This helps monitor the performance of each vendor and allows for corresponding training and optimization of the ASR technology based on actual usage.

[0088] Reference Figure 2 The flowchart illustrating the method for obtaining the comprehensive test score shows that "based on speech samples, multiple preset speech recognition models are tested to obtain the comprehensive test score for each speech recognition model," which includes the following steps:

[0089] S21: Divide the speech sample into multiple speech segments;

[0090] S22: Test each speech recognition model multiple times using multiple speech segments, and obtain the single-word translation time, misspelling rate and fluency parameters of each speech recognition model in each test;

[0091] S23: Based on the translation time of a single character, obtain the time-level score of each speech recognition model;

[0092] S24: Based on the misspelling rate and fluency parameters, obtain the test quality parameters for each speech recognition model;

[0093] S25: Based on single-character translation time, misspelling rate, and fluency parameters, obtain the learning ability parameters of each speech recognition model;

[0094] S26: Based on the duration level score, test quality parameters, and learning ability parameters, obtain the comprehensive test score for each speech recognition model.

[0095] In S21, before conducting actual testing, the collected speech samples are preprocessed to divide each sample into multiple speech segments. Specifically, the natural pauses when the user speaks are used as reference points, and other acoustic features such as speech intensity and intonation changes are combined to segment the speech samples, thereby obtaining continuous and complete speech segments.

[0096] In S22, each speech recognition model is tested for each speech segment separately. The total translation time and number of translated characters for each speech segment are obtained. The translation time per character of a speech segment is equal to the ratio of the total translation time to the number of translated characters. The speech segment is matched with the source target template (the target template to which the speech segment belongs). The corresponding text content is determined through comparative analysis. The translation accuracy and fluency of the speech recognition model are then tested based on the text content.

[0097] Regarding translation accuracy, if the translation is completely correct, the error rate is set to 0; if errors exist, the error rate is equal to the ratio of the number of errors to the total number of words in the text. Regarding translation fluency, if the speech recognition model translates the speech segment fluently, the fluency parameter is set to a first preset score (e.g., 1 or 10); if the translation is incoherent, the fluency parameter is set to a second preset score (e.g., 0). Alternatively, N-gram models, GPT models, or subjective evaluation methods can be used to detect translation fluency, and the specific value of the fluency parameter can be determined based on the translation fluency, but the minimum value is the second preset score.

[0098] In S23, multiple numerical ranges are preset, each with different priorities. The translation time for a single character is matched against these ranges, and the duration score is equal to the priority of the successfully matched range. For example, when the translation time for a single character is less than 100 milliseconds, the duration score is 10; when the translation time is greater than or equal to 100 milliseconds and less than 200 milliseconds, the score is 9; when the translation time is greater than or equal to 200 milliseconds and less than 300 milliseconds, the score is 7; when the translation time is greater than or equal to 300 milliseconds and less than 400 milliseconds, the score is 5; and when the translation time is greater than or equal to 400 milliseconds, the score is 2.

[0099] In S24, if the misspelling percentage is 0 and the fluency parameter is not equal to the second preset score, then the test quality parameter is determined to be equal to the third preset score (e.g., 10); if the misspelling percentage is 0 but the fluency parameter is equal to the second preset score, or if the misspelling percentage is not equal to 0 but the fluency parameter is not equal to the second preset score, then the test quality parameter is determined to be equal to the fourth preset score (e.g., 9); if the misspelling percentage is not equal to 0 and the fluency parameter is equal to the second preset score, then the test quality parameter is determined to be equal to the fifth preset score (e.g., 8).

[0100] In S25, refer to Figure 3 The flowchart illustrating the method for obtaining learning ability parameters shows that "based on single-character translation time, misspelling rate, and fluency parameters, the learning ability parameters for each speech recognition model are obtained" includes the following steps:

[0101] S251: Based on the single-character translation time, misspelling rate, and fluency parameters, construct the time analysis graph, misspelling analysis graph, and semantic analysis graph for each speech recognition model;

[0102] S252: Based on the duration analysis graph, obtain the speed improvement parameters for each speech recognition model;

[0103] S253: Based on the misspelling analysis graph, obtain the accuracy improvement parameters for each speech recognition model;

[0104] S254: Based on the semantic analysis graph, obtain the semantic quality improvement parameters for each speech recognition model;

[0105] S255: Based on speed improvement parameters, accuracy improvement parameters, and semantic quality improvement parameters, obtain the learning ability parameters for each speech recognition model.

[0106] Record the data collection times for single-word translation time, misspelling percentage, and fluency parameters (assuming these three parameters are collected simultaneously). Construct a time-based analysis graph based on the collection time and single-word translation time; a misspelling analysis graph based on the collection time and misspelling percentage; and a semantic analysis graph based on the collection time and fluency parameters. All three graphs are two-dimensional coordinate graphs, where the horizontal axis represents the collection time and the vertical axis represents the collected parameter values.

[0107] Mathematical methods were used to process each analysis graph, calculating the corresponding average slope. The average slope of the duration analysis graph represents the speed improvement parameter, the average slope of the misspelling analysis graph represents the accuracy improvement parameter, and the average slope of the semantic analysis graph represents the semantic quality improvement parameter. Based on the obtained speed improvement parameter, accuracy improvement parameter, and semantic quality improvement parameter, the learning ability parameter of each speech recognition model was calculated. The formula for calculating the learning ability parameter is as follows:

[0108] ;(Formula 1)

[0109] In Formula 1, This serves as an identifier for the speech recognition model. For the first The learning ability parameters of a speech recognition model; , , These are preset weights; For the first Speed ​​improvement parameters for each speech recognition model; For the first Parameters for improving the accuracy of a speech recognition model; For the first Semantic quality improvement parameters for a speech recognition model.

[0110] In S26, the formula for calculating the comprehensive test score is as follows:

[0111] ;(Formula 2)

[0112] In Formula 2, For identifying speech segments, ; This represents the total number of speech segments; For the first The comprehensive test score of each speech recognition model; For the first The speech recognition model is for the first... The duration rating of each audio segment; For the first The speech recognition model is for the first... Test quality parameters for each audio segment; , , The preset weights.

[0113] In S3, the speech recognition model with the highest overall test score is selected and identified as the optimal model. This model is then deployed to the user's voice call system to provide the best speech recognition service. Simultaneously, other participating speech recognition models are designated as backup models. This allows for rapid switching to the best-performing backup model should the current model encounter performance or network issues, ensuring the user's voice call system always operates at its best. Furthermore, the existence of backup models provides more options and flexibility for future performance optimization and model upgrades.

[0114] In S4, the total translation time of the current model for the current voice call is obtained by recording the start and end times of the translation. When the user completes the current voice call using the current model, the translated text data is output, and the number of translated characters is detected based on the text data. The unit processing time of the current model for the current voice call is equal to the ratio of the total translation time to the number of translated characters; the size of the unit processing time reflects the translation speed. Based on the unit processing time, the timeliness score of the current model is obtained, and its principle is the same as that of S23, which will not be repeated here.

[0115] The current model transmits the translated text data to a Large Language Model (LLM) as input. The LLM not only generates a response based on the input and converts the text back to audio using text-to-speech (TTS) technology, but also performs sentence analysis on the text data, such as detecting typos and fluency using prompts. Based on the typo and fluency detection results from the LLM, typo parameters and logical parameters are obtained.

[0116] If the large language model reports no typos, the typo parameter is set to 0. If the large language model reports typos, the typo parameter is set to the ratio of the number of typos reported by the large language model to the total number of characters in the text. If the large language model reports a fluent translation, the logical parameter is set to the first preset score. If the large language model reports an incoherent translation, the logical parameter is set to the second preset score. The large language model can also detect the degree of fluency and determine the specific value of the logical parameter based on the degree of fluency, but the minimum value is the second preset score.

[0117] Based on the misspelling parameter and the logical parameter, the current model's usability parameters are obtained. The principle is the same as that of S24, so it will not be repeated here.

[0118] Record the collection time for unit processing time, typo parameters, and logical parameters (assuming these three parameters are collected simultaneously). After each acquisition of unit processing time and its collection time, update the duration analysis graph based on the unit processing time and collection time, recalculate the average slope of the duration analysis graph, and update the value of the speed improvement parameter. Similarly, after each acquisition of typo parameters and its collection time, update the typo analysis graph based on the typo parameters and its collection time, recalculate the average slope of the typo analysis graph, and update the value of the accuracy improvement parameter. After each acquisition of logical parameters and its collection time, update the semantic analysis graph based on the logical parameters and its collection time, recalculate the average slope of the semantic analysis graph, and update the value of the semantic quality improvement parameter.

[0119] After the current model has translated the user's current voice call, the performance improvement parameters of the current model are calculated based on the updated speed improvement parameters, accuracy improvement parameters, and semantic quality improvement parameters. The formula for calculating the performance improvement parameters is as follows:

[0120] ;(Formula 3)

[0121] In formula 3, For the first Each speech recognition model has performance improvement parameters for current voice calls; For the first The speed improvement parameters after updating the speech recognition model; For the first Parameters for improving the accuracy of the updated speech recognition model; For the first The semantic quality improvement parameters after updating the speech recognition model.

[0122] The performance of the current model in the current voice call is evaluated by using timeliness scores, quality parameters, and performance improvement parameters to obtain a comprehensive performance score. The formula for calculating the comprehensive performance score is as follows:

[0123] ;(Formula 4)

[0124] In formula 4, For the first The overall performance score of each speech recognition model; For manual weighting; For the first Each speech recognition model scores the timeliness of the current voice call; For the first Each speech recognition model is designed for the quality parameters of the current voice call. , , The preset weights.

[0125] Existing large language models can correct and identify individual typos, meaning they can ignore some inaccurate inputs. Furthermore, translation quality assessment relies heavily on the feedback from these large language models. However, inherent limitations of these models can lead to errors in the assessment results. Therefore, designing... Greater than This allows the calculation of the overall performance score to focus more on translation speed.

[0126] Through the above methods, as users continue to use the voice call system, the corresponding analysis graph can be updated after each translation by the current model, acquiring new performance improvement parameters, and collecting new timeliness scores and usage quality parameters. This results in a comprehensive performance score for the current model in each actual voice call. Furthermore, during actual voice calls, the current model can further learn from users' individual language characteristics and habits through continuous use, achieving adjustments and optimizations.

[0127] This method also introduces an adjustable human weight for each vendor's speech recognition model, with an initial value set to 1. In practical applications, the value of the human weight can be adjusted according to specific circumstances to achieve human intervention in a particular speech recognition model. For example, in cases of abnormal performance degradation, user complaints, or privacy risks associated with the model, the human weight can be modified to 0, triggering the system to determine that the current model is abnormal, issue an alarm, immediately stop using the current model, and switch to other alternative models.

[0128] In S5, besides manual anomaly detection, real-time monitoring can automatically determine if the current model is abnormal. Specifically, it checks if the current model's overall performance score is lower than a first preset threshold for N consecutive times, or if a preset percentage of the overall performance score is lower than a second preset threshold within a first preset time range. If so, the current model is considered abnormal; otherwise, it is considered to be running normally, requiring no model switching, and the translation level of the current model continues to be monitored. For example, if N is 2, the first preset time range is 1 minute, the preset percentage is 30%, and both the first and second preset thresholds are 7, then the current model is considered abnormal if the overall performance score obtained by the current model is below 7 for two consecutive times, or if more than 30% of the overall performance scores obtained within 1 minute are below 7.

[0129] In S6, when the current model malfunctions, an alternative model is selected to replace it. Specifically, the sum of the evaluation scores for each alternative model across M consecutive performance analyses is obtained, where M is 3. Performance analyses include testing and / or real-world application evaluations, and evaluation scores include a comprehensive test score and / or a comprehensive performance score. This means that if the last three performance analyses of the current model were conducted during testing, the comprehensive test scores from the last three tests are obtained and summed. If the last performance analysis of the current model was conducted during actual voice call translation for a user, and the two-to-last performance analyses were conducted during testing, the comprehensive performance score obtained from the first analysis and the comprehensive test scores from the last two tests are obtained and summed. If the last three performance analyses of the current model were conducted during actual voice call translation for a user, the comprehensive performance scores from the last three analyses of the current model are summed.

[0130] The candidate model with the highest total score is identified as the best historically rated model and replaces the current model, becoming the new current model. This model switching method ensures the smoothness and accuracy of user voice calls. Furthermore, once the abnormal model recovers, the recovered model is used as a backup model for future use if needed.

[0131] Speech recognition technology typically utilizes mature products from established industry vendors. These vendors, leveraging their long-term expertise in their respective fields, can provide ASR (Automatic Speech Recognition) products that perform well within their areas of strength. In the technical chain of intelligent call centers or other voice call scenarios, the translation quality and speed of ASR are crucial to the entire system. Slow ASR speed will affect the speed of dialogue response, leading to a negative user experience of being unresponsive. Poor translation quality, such as semantic errors or excessive typos, will directly impact the response content of the large language model, thus affecting the quality of the voice call and the user experience. This method, in the context of multiple ASR vendors coexisting, achieves vendor redundancy and fault tolerance. It allows for rapid switching to other vendors' ASR products when one vendor's product encounters problems or performs poorly, maintaining a high level of service and user experience.

[0132] Reference Figure 4 This disclosure provides an adaptive speech recognition switching system, including:

[0133] Response module 101 is used to respond to the user's acquisition command, record the user's voice, and acquire voice samples;

[0134] The test module 102 is used to test various preset speech recognition models based on speech samples and obtain the comprehensive test score of each speech recognition model.

[0135] The deployment module 103 is used to determine the speech recognition model with the highest comprehensive test score as the current model and deploy it to the user's voice call system, and to determine the remaining speech recognition models as candidate models.

[0136] The monitoring module 104 is used to monitor the translation level of the current model in real time and obtain the comprehensive performance score of the current model.

[0137] The judgment module 105 is used to determine whether there is an anomaly in the current model based on the comprehensive performance score; if so, the replacement module 106 is executed.

[0138] Replacement module 106 is used to replace the current model with the candidate model that has the best historical score.

[0139] The various variations and specific examples of the adaptive speech recognition switching method provided above are also applicable to the adaptive speech recognition switching system provided in this disclosure. Through the foregoing detailed description of the adaptive speech recognition switching method, those skilled in the art can clearly understand the implementation method of the adaptive speech recognition switching system. For the sake of brevity, they will not be described in detail here.

[0140] A computer device according to embodiments of the present disclosure includes a memory and a processor. The memory is used to store non-transitory computer-readable instructions. Specifically, the memory may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may, for example, include random access memory (RAM) and / or cache memory. The non-volatile memory may, for example, include read-only memory (ROM), hard disk, flash memory, etc.

[0141] The processor may be a central processing unit (CPU) or other processing unit with data processing capabilities and / or instruction execution capabilities, and may control other components in the computer device to perform desired functions. In one embodiment of this disclosure, the processor is used to execute computer-readable instructions stored in the memory, causing the computer device to perform all or part of the steps of the adaptive speech recognition switching method of the foregoing embodiments of this disclosure.

[0142] Those skilled in the art will understand that, in order to solve the technical problem of how to achieve a good user experience, this embodiment may also include well-known structures such as communication buses and interfaces, and these well-known structures should also be included within the protection scope of this disclosure.

[0143] like Figure 5This is a schematic diagram of a computer device provided for an embodiment of the present disclosure. It illustrates a structural schematic diagram suitable for implementing the computer device in the embodiments of the present disclosure. Figure 5 The computer device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments disclosed herein.

[0144] like Figure 5 As shown, a computer device may include a processor (such as a central processing unit, graphics processing unit, etc.), which can perform various appropriate actions and processes based on programs stored in read-only memory (ROM) or programs loaded from storage devices into random access memory (RAM). The RAM also stores various programs and data required for the operation of the computer device. The processor, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0145] Typically, the following devices can be connected to the I / O interface: input devices, such as sensors or visual information acquisition devices; output devices, such as displays; storage devices, such as magnetic tapes or hard drives; and communication devices. Communication devices allow the computer device to communicate wirelessly or wiredly with other devices (such as edge computing devices) to exchange data. Although Figure 5 A computer apparatus with various devices is shown, but it should be understood that it is not required to implement or have all of the devices shown. More or fewer devices may be implemented or included alternatively.

[0146] In particular, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from a storage device, or installed from a ROM. When the computer program is executed by a processor, all or part of the steps of the adaptive speech recognition switching method of embodiments of this disclosure are performed.

[0147] For a detailed description of this embodiment, please refer to the corresponding descriptions in the foregoing embodiments, which will not be repeated here.

[0148] A computer-readable storage medium according to embodiments of the present disclosure stores non-transitory computer-readable instructions. When these non-transitory computer-readable instructions are executed by a processor, all or part of the steps of the adaptive speech recognition switching methods described in the foregoing embodiments of the present disclosure are performed.

[0149] The aforementioned computer-readable storage media include, but are not limited to: optical storage media (e.g., CD-ROM and DVD), magneto-optical storage media (e.g., MO), magnetic storage media (e.g., magnetic tape or portable hard drive), media with built-in rewritable non-volatile memory (e.g., memory card), and media with built-in ROM (e.g., ROM cartridge).

[0150] For a detailed description of this embodiment, please refer to the corresponding descriptions in the foregoing embodiments, which will not be repeated here.

[0151] The basic principles of this disclosure have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this disclosure are merely examples and not limitations, and should not be considered as essential features of each embodiment of this disclosure. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the scope of this disclosure to the necessity of employing the aforementioned specific details for implementation.

[0152] In this disclosure, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The block diagrams of devices, apparatuses, devices, and systems involved in this disclosure are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, devices, and systems can be connected, arranged, and configured in any manner. Words such as "comprising," "including," "having," etc., are open-ended terms meaning "including but not limited to," and are used interchangeably with them. The terms "or" and "and" as used herein refer to the terms "and / or," and are used interchangeably with them unless the context clearly indicates otherwise. The term "such as" as used herein refers to the phrase "such as but not limited to," and is used interchangeably with it.

[0153] Additionally, as used herein, the "or" used in a list of items beginning with "at least one" indicates a separate list, such that a list of, for example, "at least one of A, B, or C" means A or B or C, or AB or AC or BC, or ABC (i.e., A and B and C). Furthermore, the word "exemplary" does not imply that the described example is preferred or better than other examples.

[0154] It should also be noted that in the systems and methods of this disclosure, the components or steps can be decomposed and / or recombined. These decompositions and / or recombinations should be considered as equivalent solutions to this disclosure.

[0155] Various changes, substitutions, and modifications can be made to the technology described herein without departing from the teachings defined by the appended claims. Furthermore, the scope of the claims of this disclosure is not limited to the specific aspects of the processes, machines, manufactures, events, means, methods, and actions described above. Currently existing or later-developed processes, machines, manufactures, events, means, methods, or actions that perform substantially the same function or achieve substantially the same result as the corresponding aspects described herein can be utilized. Therefore, the appended claims include such processes, machines, manufactures, events, means, methods, or actions within their scope.

[0156] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use this disclosure. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of this disclosure. Therefore, this disclosure is not intended to be limited to the aspects shown herein, but rather to be carried out within the widest scope consistent with the principles and novel features disclosed herein.

[0157] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this disclosure to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations therein.

Claims

1. A method of adaptive speech recognition switching, the method comprising: include: In response to the user's collection command, the system records the user's voice and obtains voice samples. Based on the speech samples, various preset speech recognition models are tested to obtain a comprehensive test score for each speech recognition model. The step of testing multiple preset speech recognition models based on the speech samples and obtaining a comprehensive test score for each speech recognition model includes: The speech sample is divided into multiple speech segments; The speech recognition model was tested multiple times using the multiple speech segments to obtain the single-word translation time, misspelling rate and fluency parameters of each speech recognition model in each test. Based on the single-character translation time, obtain the time level score for each speech recognition model; Based on the typo percentage and the fluency parameter, the test quality parameters for each speech recognition model are obtained; Based on the single-character translation time, the percentage of misspelled words, and the fluency parameter, the learning ability parameters of each speech recognition model are obtained; Based on the duration level score, the test quality parameter, and the learning ability parameter, a comprehensive test score is obtained for each speech recognition model; The step of obtaining the learning ability parameters for each speech recognition model based on the single-character translation time, the percentage of misspelled words, and the fluency parameter includes: Based on the single-character translation time, the misspelling rate, and the fluency parameter, a time analysis graph, a misspelling analysis graph, and a semantic analysis graph are constructed for each speech recognition model. Based on the duration analysis graph, obtain the speed improvement parameters for each speech recognition model; Based on the misspelling analysis graph, the accuracy improvement parameters for each speech recognition model are obtained; Based on the semantic analysis graph, semantic quality improvement parameters for each speech recognition model are obtained; Based on the speed improvement parameter, the accuracy improvement parameter, and the semantic quality improvement parameter, the learning ability parameters of each speech recognition model are obtained; In the duration analysis graph, the misspelling analysis graph, and the semantic analysis graph, the horizontal axis represents the collection time, and the vertical axis represents the collected parameter values. The speech recognition model with the highest comprehensive test score is determined as the current model and deployed to the user's voice call system, while the remaining speech recognition models are determined as candidate models. Real-time monitoring of the translation level of the current model to obtain the overall performance score of the current model; Determine whether the current model has any anomalies based on the comprehensive performance score; If so, the current model will be replaced by the candidate model with the best historical score.

2. The adaptive speech recognition switching method according to claim 1, characterized in that, The step of responding to the user's collection command by recording the user's voice and obtaining voice samples includes: Build a text template library; Receive several selection requests from the user; In response to each selection request, several target templates are selected from the text template library; Configure a data collection button for each target template; When the user triggers the data collection button on the front-end interface, a corresponding data collection command is generated; In response to the acquisition command, the audio capture device is activated; The audio capture device records the user's speech and generates speech samples.

3. The adaptive speech recognition switching method according to claim 1, characterized in that, The real-time monitoring of the current model's translation level and the acquisition of the current model's overall performance score include: Monitor the performance of the current model in the current voice call, and obtain the unit processing time, typo parameters, and logical parameters; Based on the unit processing time, obtain the timeliness score of the current model; Based on the misspelling parameters and the logical parameters, the usage quality parameters of the current model are obtained; Based on the unit processing time, the misspelling parameters, and the logical parameters, obtain the performance improvement parameters of the current model; Based on the timeliness score, the usage quality parameters, and the performance improvement parameters, the performance of the current model in the current voice call is evaluated to obtain a comprehensive performance score.

4. The adaptive speech recognition switching method according to claim 1, characterized in that, The step of determining whether the current model has an anomaly based on the comprehensive performance score includes: Determine whether the comprehensive performance score obtained by the current model is less than a first preset threshold for N consecutive times, or whether a preset proportion of the comprehensive performance score is less than a second preset threshold within a first preset time range. If so, then the current model is determined to be abnormal; If not, then the current model is considered to be operating normally.

5. The adaptive speech recognition switching method according to claim 1, characterized in that, The step of replacing the current model with the candidate model that has the best historical score includes: Obtain the sum of the evaluation scores for each alternative model in M ​​consecutive performance analyses, wherein the evaluation scores include the comprehensive test score and / or the comprehensive performance score; The candidate model with the highest total score is determined as the best historical score and replaces the current model.

6. A computer device, characterized in that, The computer device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the adaptive speech recognition switching method according to any one of claims 1-5.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to perform the adaptive speech recognition switching method according to any one of claims 1-5.

8. A computer program product comprising computer instructions, characterized in that, When executed by a processor, the computer instructions implement the steps of the method according to any one of claims 1-5.

Images