Speech recognition method and device, equipment and storage medium

Abstract

The invention relates to the technical field of voice recognition, and provides a voice recognition method and device, equipment and a storage medium, and the method comprises the steps: obtaining to-be-recognized voice information; based on the trained voice recognition model, performing voice recognition on the to-be-recognized voice information to obtain a preset recognition result corresponding to the to-be-recognized voice information; determining the voice type of the to-be-recognized voice information according to a preset recognition result; when it is determined that the to-be-recognized voice information meets a preset re-recognition condition according to the voice type, performing voice recognition on the to-be-recognized voice information based on a trained voice recognition large model to obtain a target recognition result corresponding to the to-be-recognized voice information; wherein the training data of the large voice recognition model is different from the training data of the voice recognition model, so that the voice recognition convenience and the voice recognition accuracy of voice information related to dialects, accent, noise and other complex voice situations are improved.

Description

Technical Field

[0001] The present application relates to the field of speech recognition technology, and in particular to a speech recognition method, apparatus, device and storage medium. Background Art

[0002] With the development of artificial intelligence, speech recognition models can be widely used to perform speech recognition on voice information. Speech recognition models include, for example, end-to-end speech recognition (Automatic Speech Recognition, ASR) models. In the related art, the end-to-end ASR model can generally only perform speech recognition on speech information related to Mandarin. However, in the case where the voice information includes complex speech such as dialects, accents, noise, etc., if the voice information is voice recognized by an end-to-end ASR model, it is easy to cause poor accuracy in speech recognition of the voice information. For example, for customer service business in the financial field, customer service needs to answer calls from different customers, and customers may use dialects or have accents when speaking. In the case where the end-to-end ASR model has poor recognition accuracy for call-related voice information, it is easy for customer service to be unable to understand the customer's problem, thereby adversely affecting the quality of customer service provided by customer service to customers. Summary of the invention

[0003] The main purpose of this application is to provide a speech recognition method, apparatus, device and storage medium, aiming to solve the technical problem of poor speech recognition accuracy of speech information due to the fact that the speech information includes complex speech such as dialects, accents, noise, etc.

[0004] In a first aspect, the present application provides a speech recognition method, comprising:

[0005] Obtaining voice information to be recognized;

[0006] Based on the trained speech recognition model, speech recognition is performed on the speech information to be recognized to obtain a preset recognition result corresponding to the speech information to be recognized;

[0007] Determining the voice type of the voice information to be recognized according to the preset recognition result;

[0008] When it is determined according to the voice type that the voice information to be recognized meets the preset re-recognition conditions, voice recognition is performed on the voice information to be recognized based on the trained voice recognition large model to obtain a target recognition result corresponding to the voice information to be recognized; wherein the training data of the voice recognition large model is different from the training data of the voice recognition model.

[0009] In a second aspect, the present application further provides a speech recognition device, the speech recognition device comprising:

[0010] An acquisition module, used to acquire voice information to be recognized;

[0011] A first recognition module, configured to perform speech recognition on the speech information to be recognized based on a trained speech recognition model, and obtain a preset recognition result corresponding to the speech information to be recognized;

[0012] A determination module, used to determine the voice type of the voice information to be recognized according to the recognition result;

[0013] The second recognition module is used to perform speech recognition on the speech information to be recognized based on a trained speech recognition large model when it is determined that the speech information to be recognized meets the preset re-recognition conditions according to the speech type, so as to obtain a target recognition result corresponding to the speech information to be recognized; wherein the training data of the speech recognition large model is different from the training data of the speech recognition model.

[0014] In a third aspect, the present application further provides a computer device, the computer device comprising a memory and a processor;

[0015] The memory is used to store computer programs;

[0016] The processor is used to execute the computer program and implement the steps of the above-mentioned speech recognition method when executing the computer program.

[0017] In a fourth aspect, the present application further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the speech recognition method as described above are implemented.

[0018] The present application provides a speech recognition method, device, equipment and storage medium, the method comprising: obtaining speech information to be recognized; based on a trained speech recognition model, performing speech recognition on the speech information to be recognized, and obtaining a preset recognition result corresponding to the speech information to be recognized; determining the speech type of the speech information to be recognized according to the preset recognition result; when it is determined that the speech information to be recognized meets the preset re-recognition condition according to the speech type, performing speech recognition on the speech information to be recognized based on a trained speech recognition large model, and obtaining a target recognition result corresponding to the speech information to be recognized; wherein the training data of the speech recognition large model is different from the training data of the speech recognition model, so as to improve the speech recognition convenience and speech recognition accuracy of speech information involving complex speech situations such as dialects, accents, and noise. For example, it can be used for customer service business in the financial field. For customer service business, when the customer service receives a call from a customer, the computer device can preferentially perform speech recognition on the speech information to be recognized corresponding to the telephone communication process between the customer service and the customer through the trained speech recognition model, so as to reduce the computing resources required for speech recognition. Accordingly, based on the preset recognition results output by the speech recognition model, it is determined that the speech type of the speech information to be recognized meets the re-recognition conditions. For example, if it involves dialects, accents, noise, etc., the computer equipment can further perform speech recognition on the speech information to be recognized through the trained speech recognition large model to obtain the corresponding target recognition results, so as to more accurately recognize the problems described by the customer through voice, so that the customer service can better provide customer service to the customer. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0020] Figure 1 It is a flowchart of a speech recognition method provided by an embodiment of the present application;

[0021] Figure 2 is a flow chart of a speech recognition method involved in an embodiment of the present application;

[0022] Figure 3 It is a schematic diagram of the training process of a large speech recognition model involved in an embodiment of the present application;

[0023] Figure 4 is a schematic block diagram of a speech recognition device provided in an embodiment of the present application;

[0024] Figure 5 It is a schematic block diagram of the structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0025] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0026] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0027] The embodiments of the present application provide a speech recognition method, apparatus, device and storage medium. The speech recognition method can be applied to a computer device, which can be a mobile phone, a tablet computer, a laptop computer, a desktop computer and the like. It can also be applied to a server, which can be a separate server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0028] In conjunction with the accompanying drawings, some embodiments of the present application are described in detail below. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0029] See also Figure 1 , Figure 1 It is a flow chart of a speech recognition method provided in an embodiment of the present application. It should be noted that the speech recognition method provided in an embodiment of the present application can be used in a computer device, and of course can also be used in a server. For example, the server can obtain speech information to be recognized from a computer device, process the speech information according to the speech recognition method, and obtain a target recognition result corresponding to the speech information. Exemplarily, the server can, for example, send the target recognition result to a computer device; of course, it is not limited to this and is not limited here.

[0030] In specific implementation, computer devices include but are not limited to: any one of mobile phones, tablet computers, laptop computers, and desktop computers; the server can be a single server or a server cluster, or a cloud server that provides cloud computing services.

[0031] like Figure 1As shown, the speech recognition method includes steps S101 to S104.

[0032] Step S101: Acquire voice information to be recognized.

[0033] When it is necessary to perform speech recognition on the speech information to be recognized, the computer device can acquire the speech information to be recognized.

[0034] In some embodiments, the voice information to be recognized can be determined based on at least one of video voice and recording. For example, for customer service business, when answering a customer call, the customer service can record the telephone communication process between the customer service and the customer to form the corresponding voice information to be recognized. Correspondingly, the computer device can subsequently perform voice recognition on the voice information to be recognized, for example, so that the customer service can clarify the customer's consultation questions. For example, when a user is in class, he can record the class to form voice information to be recognized. Correspondingly, the computer device can subsequently perform voice recognition on the voice information to be recognized, for example, to form class notes. Of course, the voice information to be recognized is not limited to this, and is not limited here.

[0035] When the voice information to be recognized is obtained, the voice information to be recognized can be used to subsequently determine a target recognition result corresponding to the voice information.

[0036] Step S102: Based on the trained speech recognition model, speech recognition is performed on the speech information to be recognized to obtain a preset recognition result corresponding to the speech information to be recognized.

[0037] For example, the speech recognition model may include an end-to-end ASR model. The end-to-end ASR model may adopt an end-to-end network structure. For example, the end-to-end network architecture may include a network architecture that integrates a Conformer neural network, Connectionist Temporal Classification (CTC), and an attention mechanism, which is not limited here.

[0038] In some implementations, the training data of the speech recognition model may include speech information of a speech type carrying a common language. The computer device may perform speech recognition on the speech information to be recognized by using the trained speech recognition model.

[0039] When the computer device obtains the voice information to be recognized, it can input the voice information to be recognized into the voice recognition model, and the voice recognition model can perform voice recognition on the voice information to be recognized to obtain a preset recognition result corresponding to the voice information to be recognized. The preset recognition result may include text corresponding to the voice information to be recognized. When the training data of the voice recognition model includes voice information of a voice type carrying a commonly used language, the computer device can determine that the trained voice recognition model has a high voice recognition accuracy for voice information of a voice type carrying a commonly used language. Accordingly, the computer device can also determine that the voice recognition model has a low voice recognition accuracy for voice information of a voice type carrying an uncommon language.

[0040] For example, the voice information to be recognized may include multiple voices. The voice information to be recognized may be represented as s i =[s1,s2,…,s n ], i = 1, 2, ..., n, where i is used to indicate that the speech information to be recognized includes the speech ordinal numbers of the respective multiple speech messages. i In the case of , the speech recognition model can treat the recognized speech information s i Perform speech recognition to obtain the speech information to be recognized s i The preset recognition results of the multiple voices are included. The preset recognition results may include the text corresponding to the multiple voices. The preset recognition results can be expressed as w j =[w1,w2,…,w m ], j = 1, 2, ..., m, where j is used to indicate the sentence ordinal number corresponding to the text included in the preset recognition result. Since the user may have corresponding pauses or corresponding associated words when speaking, the speech recognition model can combine the pause information, associated words, etc. between adjacent voices to divide the text included in the preset recognition result into corresponding sentences when identifying the preset recognition results corresponding to the multiple voices included in the voice information to be recognized. Based on this, the voice ordinal number i of the voice included in the voice information to be recognized and the sentence ordinal number corresponding to the text included in the preset recognition result can be one-to-one corresponding.

[0041] In this way, when the trained speech recognition model is used to perform speech recognition on the speech information to be recognized and a preset recognition result corresponding to the speech information is obtained, the computer device can subsequently determine the speech type of the speech information based on the preset recognition result. For example, the computer device can subsequently evaluate the speech recognition accuracy of the speech recognition model on the speech information to be recognized based on whether the speech type of the speech information to be recognized indicated by the preset recognition result is consistent with the speech type corresponding to the training data of the speech recognition model, so as to determine whether the speech information needs to be re-recognized.

[0042] Exemplarily, the computer device may pre-train the language recognition model to obtain a trained speech recognition model. The trained speech recognition model may be used to perform speech recognition on the speech information to be recognized.

[0043] In some embodiments, a third training sample set is obtained, the third training sample set including multiple third voice information and second recognition result labels corresponding to each third voice information; based on the trained voice recognition model, voice recognition is performed on the third voice information to obtain a second recognition result corresponding to the third voice information; based on the second recognition result and the second recognition result label, model optimization parameters of the voice recognition model are determined; and based on the model optimization parameters, model parameters of the voice recognition model are adjusted.

[0044] For example, when training the speech recognition model, a third training sample set may be obtained. The third training sample set may include a plurality of third speech information. Each third speech information in the third training sample set may be provided with a corresponding second recognition result label.

[0045] The third voice information may include voice information of a commonly used language. For example, the commonly used language of Chinese people is Mandarin, so when training the voice recognition model, the third voice information may include Mandarin voice information. The commonly used language of British people is English, so when training the voice recognition model, the third voice information may include English voice information. And so on.

[0046] In an exemplary embodiment, when the third voice information is obtained, the second recognition result label corresponding to the third voice information can be determined by a preset annotation method, which is not limited here.

[0047] For example, when training a speech recognition model, it is necessary to learn how to determine the second recognition result corresponding to the third speech information.

[0048] Take the network architecture of the speech recognition model including an encoder, an attention mechanism layer, a decoder and an activation function layer as an example. The speech recognition model can perform speech recognition on the third voice information through the encoder, the attention mechanism layer, the decoder and the activation function layer to obtain the second recognition result corresponding to the third voice information. For example, the speech recognition model can capture the key information of the third voice information through the encoder to obtain the encoding vector corresponding to the third voice information. The attention mechanism layer can assign attention weights to the decoder in combination with the key information output by the encoder and the state of the decoder. Accordingly, the decoder can convert the encoding vector into a text sequence in combination with the encoding vector and the attention weight. The activation function layer can normalize the text sequence output by the decoder to determine the second recognition result corresponding to the third voice information. By analogy, the speech recognition model can determine the second recognition results corresponding to each third voice information in the third training sample set.

[0049] For example, when the second recognition result corresponding to the third voice information is determined by the speech recognition model, the second recognition result is equivalent to the predicted result of the speech recognition model for the preset recognition result corresponding to the third voice information. The second recognition result label corresponding to the third voice information is equivalent to the actual result of the speech recognition model for the preset recognition result corresponding to the third voice information. Based on the second recognition result and the second recognition result label, the loss value between the actual result and the predicted result of the speech recognition model when determining the preset recognition result corresponding to the third voice information can be determined.

[0050] The speech recognition model can determine the model optimization parameters of the speech recognition model according to the loss value. The model optimization parameters can be used to subsequently adjust the model parameters of the speech recognition model so that the speech recognition model can more accurately move the prediction result of the speech recognition model for the preset recognition result corresponding to the third speech information closer to the actual result, thereby achieving the purpose of training.

[0051] For example, the speech recognition model can adjust the model parameters of the speech recognition model according to the model optimization parameters, such as adjusting the model parameters of at least one of the encoder, attention mechanism layer, decoder and activation function layer, which is not limited here.

[0052] Since the speech recognition model is trained based on the third speech information, the third speech information may include speech information whose speech type is a commonly used language. When the speech recognition model is used to perform speech recognition on the speech information to be recognized later, if the speech information to be recognized includes speech information whose speech type is a commonly used language, the speech recognition model generally has a high speech recognition accuracy for the speech information to be recognized. Based on this, when the computer device subsequently performs speech recognition on the speech information to be recognized, if it is determined that the speech type of the speech information to be recognized includes a commonly used language based on the preset recognition result corresponding to the speech information to be recognized, the computer device may determine that the speech information to be recognized does not meet the preset re-recognition condition. Accordingly, when it is determined that the speech information to be recognized does not meet the preset re-recognition condition, the computer device may determine the preset recognition result corresponding to the speech information to be recognized as the target recognition result corresponding to the speech information.

[0053] In this way, the computer device trains the speech recognition model through the third training sample set, which is helpful to improve the speech recognition accuracy and convenience of the speech recognition model for speech information carrying common languages.

[0054] Step S103: Determine the voice type of the voice information according to the preset recognition result.

[0055] For example, different voice information to be recognized may correspond to different users, and different users may have different languages, accents, speech speeds, the noisiness of the environment, the number of languages ​​stated, and the like. Take the voice information to be recognized as including voice information 1 to voice information 3. When user A tells voice information 1, he uses a dialect and speaks at a fast speed. When user B tells voice information 2, he uses Mandarin with an accent, and the environment is noisy. When user C tells voice information 3, he uses Mandarin, with a moderate speech speed, and the environment is quiet. Based on this, when the computer device performs voice recognition on different voice information to be recognized through a voice recognition model, the preset recognition results corresponding to different voice information to be recognized are easily affected by different users' languages, accents, speech speeds, the noisiness of the environment, the number of languages ​​stated, and the like. Based on this, the computer device can determine the voice type of the voice information to be recognized based on the preset recognition results.

[0056] Exemplarily, the voice type may include at least one of carrying common language, carrying dialect, carrying accent, carrying noise, speaking speed greater than or equal to a preset speaking speed threshold, and the number of languages ​​greater than or equal to a preset number threshold.

[0057] The voice type of the voice information to be recognized can be used by the computer device to subsequently evaluate the recognition accuracy of the voice recognition model on the voice information to be recognized, so as to determine whether the voice information needs to be re-recognized.

[0058] For example, when the speech type corresponding to the training data of the speech recognition model is a commonly used language, if the speech type of the speech information to be recognized includes at least one of carrying a dialect, carrying an accent, carrying noise, a speech speed greater than or equal to a preset speech speed threshold, and a number of languages ​​greater than or equal to a preset number threshold, then it can be determined that the speech type of the speech information to be recognized does not match the speech type corresponding to the training data of the speech recognition model. Based on this, the computer device can determine that the speech recognition model has poor recognition accuracy for the speech information to be recognized, and the speech information to be recognized needs to be re-recognized.

[0059] In some embodiments, based on a confidence assessment model, a confidence assessment is performed on a preset recognition result to obtain a corresponding confidence score; when the confidence score is less than or equal to a preset confidence threshold, the speech type of the speech information to be recognized is determined to include at least one of speech with a dialect, speech with an accent, speech with noise, speech with a speaking speed greater than or equal to a preset speaking speed threshold, and speech with a number of languages ​​greater than or equal to a preset number threshold.

[0060] For example, when a preset recognition result is obtained, the computer device can input the preset recognition result into a confidence assessment model, and the confidence assessment model can perform a confidence assessment on the preset recognition result to obtain a confidence score corresponding to the preset recognition result. The confidence assessment model can adopt, but is not limited to, a confidence assessment (Confidence Estimation Module, CEM) model based on deep learning, which is not limited here.

[0061] For example, the network architecture of the confidence assessment model may include a fully connected layer and an activation function layer. The confidence assessment model may calculate the weight and bias of the preset recognition result through the fully connected layer, and determine the confidence score corresponding to the preset recognition result through the application of the activation function layer.

[0062] For example, the computer device may be provided with a preset confidence threshold corresponding to the confidence assessment model. The preset confidence threshold may be preset or may be set by the user, which is not limited here. The computer device may evaluate the speech recognition accuracy of the speech recognition model for the speech information to be recognized by comparing the confidence score corresponding to the preset recognition result output by the confidence assessment model with the preset confidence threshold, so as to determine whether the speech information to be recognized needs to be re-recognized.

[0063] When the confidence score corresponding to the preset recognition result is less than or equal to the preset confidence threshold, the computer device can determine that the speech recognition accuracy of the speech recognition model for the speech information to be recognized is poor and the speech information to be recognized needs to be re-recognized.

[0064] Take the case where the speech type corresponding to the training data of the speech recognition model is a commonly used language as an example. In the case where the speech type of the speech information to be recognized includes at least one of carrying a dialect, carrying an accent, carrying noise, a speech speed greater than or equal to a preset speech speed threshold, and a number of languages ​​greater than or equal to a preset number threshold, since the speech type of the speech information to be recognized does not match the speech type corresponding to the training data of the speech recognition model, it is easy to cause the speech recognition model to perform speech recognition on the speech information to be recognized. The confidence score corresponding to the preset recognition result is less than or equal to the preset confidence threshold. Accordingly, the computer device can determine that the speech recognition accuracy of the speech recognition model to be recognized speech information is low when it is determined that the confidence score corresponding to the preset recognition result is less than or equal to the preset confidence threshold, and then it can be determined that the speech type of the speech information to be recognized includes at least one of carrying a dialect, carrying an accent, carrying noise, a speech speed greater than or equal to a preset speech speed threshold, and a number of languages ​​greater than or equal to a preset number threshold based on the speech type corresponding to the training data of the speech recognition model as carrying a commonly used language.

[0065] When the confidence score corresponding to the preset recognition result is greater than the preset confidence threshold, the computer device can determine that the speech recognition model has a high speech recognition accuracy for the speech information to be recognized, and there is no need to re-recognize the speech information. Accordingly, the computer device can determine that the speech type of the speech information to be recognized includes carrying a common language based on the speech type corresponding to the training data of the speech recognition model as carrying a common language.

[0066] In some embodiments, based on a text analysis model, the semantic relevance of adjacent texts in a preset recognition result is analyzed to obtain the correlation between adjacent texts; when the ratio between the number of adjacent texts whose correlation is less than or equal to a preset correlation threshold and the total number of texts in the preset recognition result is greater than or equal to a preset ratio threshold, it is determined that the voice type of the voice information to be recognized includes at least one of carrying a dialect, carrying an accent, carrying noise, and a speaking speed greater than or equal to a preset speaking speed threshold.

[0067] When a preset recognition result is obtained, the computer device may input the preset recognition result into a text analysis model, and the text analysis model may analyze the semantic relevance of adjacent texts in the preset recognition result to obtain the relevance between adjacent texts.

[0068] For example, the computer device may be provided with a preset relevance threshold corresponding to the text analysis model. The preset relevance threshold may be pre-set or may be set by the user, which is not limited here. The computer device may evaluate the speech recognition accuracy of the speech recognition model for the speech information to be recognized based on the relationship between the relevance between adjacent texts corresponding to the preset recognition result output by the text analysis model and the preset relevance threshold, so as to determine whether the speech information to be recognized needs to be re-recognized.

[0069] When the speech type corresponding to the training data of the speech recognition model carries a commonly used language, the speech recognition model usually performs speech recognition on the speech information to be recognized according to the logic of the commonly used language. If the speech type of the speech information to be recognized is different from the speech type corresponding to the training data of the speech recognition model, it is easy for the correlation between adjacent texts in the preset speech results to be low. Based on this, the computer device can evaluate the speech recognition accuracy of the speech recognition model for the speech information to be recognized based on the correlation between adjacent texts in the preset speech results, so as to determine whether the speech information to be recognized needs to be re-recognized.

[0070] Take the common language as Mandarin and the dialect as Cantonese as an example. The preset recognition result obtained by the speech recognition model for the speech information to be recognized with the speech type of the dialect includes, for example, "Leihou, I want to ask, can you tell me about this product?". The preset recognition result corresponds to the Mandarin text "Hello, I want to ask, can you explain this product to me?". In the case where the speech type corresponding to the training data of the speech recognition model is a common language, the speech recognition model performs speech recognition on the speech information with the dialect according to the logic of Mandarin to obtain the preset recognition result. The correlation between adjacent texts will be low, which can easily lead to poor speech recognition accuracy of the speech recognition model for the speech information to be recognized.

[0071] When a computer device performs speech recognition on speech information to be recognized with an accent through a speech recognition model to obtain a preset recognition result, the presence of the accent may easily lead to a low correlation between adjacent texts in the preset recognition result. Taking Mandarin as an example, the commonly used language may include situations such as the inability to distinguish between front and back nasal sounds, flat and curled tongue sounds, lateral nasal sounds n and l sounds, erhua sounds and special tones, different dialect accents, etc. When the speech type corresponding to the training data of the speech recognition model is a commonly used language, the speech recognition model performs speech recognition on the speech information to be recognized with an accent according to the logic of Mandarin to obtain a low correlation between adjacent texts in the preset recognition result, which may easily lead to poor speech recognition accuracy of the speech recognition model for the speech information.

[0072] When a computer device performs speech recognition on speech information to be recognized with a speech type of noisy speech through a speech recognition model and obtains a preset recognition result, due to the presence of noise, it is easy for the text corresponding to the noise to appear in the preset recognition result. Taking customer service as an example, when a customer service agent answers a customer's call, the agent can record the questions raised by the customer or search for relevant materials for the customer's needs. Then, for example, the agent can record the corresponding questions or search for relevant materials through a keyboard. Correspondingly, if the call between the customer and the customer service agent is recorded to form the speech information to be recognized, the speech information to be recognized may carry noise, such as the sound of keyboard tapping. For example, after performing speech recognition on the speech information to be recognized through a speech recognition model, the preset recognition result may have a situation where the call text corresponding to the customer service agent and the customer is adjacent to the noise text corresponding to the keyboard tapping sound. When the relevance between the call text and the noise text is relatively low, it is easy for the relevance between adjacent texts in the preset recognition result to be relatively low, resulting in poor speech recognition accuracy of the speech recognition model for the speech information.

[0073] When a computer device performs speech recognition on speech information to be recognized with a speech type of speech speed greater than or equal to a speech speed threshold through a speech recognition model and obtains a preset recognition result, due to the relatively fast speech speed, situations such as word omission or substitution, sentence segmentation errors, incorrect word recognition, punctuation errors, word order confusion, etc. are likely to occur in the preset recognition result. Then, it is easy for the relevance between adjacent texts in the preset recognition result to be relatively low, resulting in poor speech recognition accuracy of the speech recognition model for the speech information.

[0074] Exemplarily, the computer device can perform text analysis on the preset recognition result through a text analysis model to determine the speech type of the speech information.

[0075] For example, when the text analysis model performs text analysis on the preset recognition result, it can determine the relevance between every preset number of adjacent texts in the preset recognition result. The preset number can include at least two, and there is no limitation here. For example, when the preset number is two, the text analysis model can determine the relevance between every two adjacent texts, such as the relevance between "lei" and "hou", the relevance between "hou" and "wo", the relevance between "wo" and "xiang", etc. Another example, when the preset number is four, the text analysis model can determine the relevance between every four adjacent texts, such as the relevance between "leihou" and "woxiang", the relevance between "woxiang" and "wenha", the relevance between "wenha" and "yiguo", etc. And so on.

[0076] Of course, it is not limited to this. In some embodiments, the text analysis model can determine the relevance between adjacent texts in combination with the sentence segmentation information corresponding to the preset text result. For example, if there is a preset separator between adjacent texts, the text analysis model may not need to determine the relevance between the adjacent texts. The preset separator may include a comma, a period, a question mark, an exclamation mark, a semicolon, etc., which are not limited here.

[0077] In other implementations, the text analysis model may select some adjacent texts from the preset recognition results in a random selection manner to perform text analysis on the selected adjacent texts to obtain the correlation between the adjacent texts. This is not limited here.

[0078] When determining the correlation between adjacent texts in a preset recognition result, the computer device can determine the voice type of the voice information to be recognized by comprehensively considering the ratio between the number of adjacent texts whose correlation is less than or equal to the preset correlation threshold and the total number of texts in the preset recognition result.

[0079] For example, when the ratio is less than a preset ratio threshold, the computer device can determine that there are more adjacent texts in the preset recognition result with good semantic correlation, and the computer device can determine that the speech recognition model has good speech recognition accuracy for the speech information to be recognized. Accordingly, the computer device can determine that the speech type of the speech information to be recognized is consistent with the speech type corresponding to the training data of the speech recognition model. Taking the example that the training data of the speech recognition model includes the speech type as carrying a common language, the computer device can determine that the speech type of the speech information to be recognized is carrying a common language.

[0080] For example, when the ratio is greater than or equal to a preset ratio threshold, the computer device can determine that there are more adjacent texts with poor semantic correlation in the preset recognition results, and the computer device can determine that the speech recognition accuracy of the speech information by the speech recognition model is poor. Accordingly, the computer device can determine that the speech type of the speech information does not match the speech type corresponding to the training data of the speech recognition model. Taking the example that the training data of the speech recognition model includes speech information with a speech type carrying a commonly used language, the computer device can determine that the speech type of the speech information to be recognized is at least one of carrying an uncommon language, such as carrying a dialect, carrying an accent, carrying noise, and a speech speed greater than or equal to a preset speech speed threshold.

[0081] In some embodiments, when there is text in the preset recognition result that is a regional language text, and the regional language indicated by the regional language text is different from the preset language corresponding to the speech recognition model, determining that the voice type of the speech information to be recognized includes a language number greater than or equal to a preset number threshold.

[0082] Taking customer service as an example, when a customer or customer service needs to switch the language to communicate with the other party, the customer or customer service can ask about the relevant regional language. For example, the customer or customer service can ask the other party "Excuse me, can you speak Cantonese?", "Can you speak Cantonese?", and so on. Among them, "Vernacular" is used to indicate Cantonese, and Cantonese can be used to indicate the language of a certain region. Based on this, when the computer device obtains a preset recognition result corresponding to the voice information to be recognized through the voice recognition model, the computer device can determine that the voice information to be recognized involves a high possibility of language switching based on the regional language text present in the preset recognition result, such as "Vernacular", and thereby determine that the voice type of the voice information to be recognized includes a number of languages ​​greater than or equal to a preset number threshold.

[0083] For example, the computer device can determine whether the voice type of the voice information to be recognized is a language number greater than or equal to a preset number threshold value in combination with the preset language corresponding to the voice recognition model. The preset language corresponding to the voice recognition model can be determined based on the training data of the voice recognition model. In the case where the training data of the voice recognition model includes voice information whose voice type is a commonly used language, and the commonly used language includes Mandarin, the preset language corresponding to the voice recognition model may include Mandarin. In the case where the regional language indicated by the regional language text is Cantonese and the preset language corresponding to the voice recognition model is Mandarin, the computer device can determine that the regional language is different from the preset language. Accordingly, the computer device can determine that the voice information to be recognized includes voice information whose voice type is a language number greater than or equal to a preset number threshold value.

[0084] In some embodiments, text tone change information corresponding to a preset recognition result is determined; when the degree of match between the text tone change information and the preset tone change information is greater than or equal to a preset matching degree threshold, it is determined that the voice information includes at least one of a dialect and an accent.

[0085] For example, for the speech information to be recognized with a speech type of a dialect and / or an accent, at least one of the dialect and the accent may have a certain tone change rule. The computer device may determine the corresponding text tone change information based on the preset recognition result corresponding to the speech information to be recognized, so as to determine whether the text tone change information conforms to the tone change rule of at least one of the corresponding dialect and the accent, such as determining whether the text tone change information conforms to the preset tone change information. The preset tone change information may be used to indicate the tone change rule corresponding to each dialect and the accent.

[0086] When the matching degree between the text tone change information and the preset tone change information is greater than or equal to the preset matching degree threshold, the computer device may determine that the voice type of the to-be-recognized voice information includes at least one of carrying a dialect and carrying an accent. When the matching degree between the text tone change information and the preset tone change information is less than the preset matching degree threshold, the computer device may determine that the voice type of the to-be-recognized voice information includes carrying a non-dialect and carrying a non-accent, such as carrying a common language.

[0087] When the voice type of the voice information to be recognized is determined, the computer device can determine whether the voice information to be recognized meets the re-recognition condition according to the voice type, so as to subsequently determine whether the voice information to be recognized needs to be re-recognized.

[0088] Step S104: When it is determined according to the voice type that the voice information meets the preset re-recognition conditions, voice recognition is performed on the voice information based on the trained voice recognition large model to obtain a target recognition result corresponding to the voice information; wherein the training data of the voice recognition large model is different from the training data of the voice recognition model.

[0089] In some embodiments, when the voice type includes at least one of carrying a dialect, carrying an accent, carrying noise, a speaking speed greater than or equal to a preset speaking speed threshold, and a number of languages ​​greater than or equal to a preset number threshold, it is determined that the voice information meets the preset re-recognition conditions.

[0090] Take the case where the speech type corresponding to the training data of the speech recognition model is a commonly used language. When the speech type of the speech information to be recognized includes at least one of carrying a dialect, carrying an accent, carrying noise, a speech speed greater than or equal to a preset speech speed threshold, and a number of languages ​​greater than or equal to a preset number threshold, the computer device can determine that the speech type of the speech information to be recognized is different from the speech type corresponding to the training data of the speech recognition model, and the computer device can determine that the speech information to be recognized meets the preset re-recognition condition based on the speech type.

[0091] When it is determined that the voice information to be recognized meets the preset re-recognition conditions, the computer device can input the voice information to be recognized into the voice recognition big model, so that the voice recognition big model performs voice recognition on the voice information to be recognized and obtains the target recognition result corresponding to the voice information to be recognized.

[0092] Exemplarily, the training data of the large speech recognition model is different from the training data of the speech recognition model.

[0093] For example, the speech type corresponding to the training data of the speech recognition big model may include at least one of carrying a dialect, carrying an accent, carrying noise, a speech speed greater than or equal to a preset speech speed threshold, and a number of languages ​​greater than or equal to a preset number threshold. Based on this, the speech recognition big model can be used to recognize various different languages ​​and dialects. Correspondingly, the speech recognition big model can also be used to recognize speech information with a faster speech speed and a noisy environment.

[0094] Compared with the speech recognition model, the speech recognition model has better robustness and generalization, and the speech recognition accuracy of the speech recognition model is better for speech information in complex scenarios. However, if the speech recognition model is used to perform speech recognition on all speech information to be recognized, the large number of parameters of the speech recognition model will easily increase the computing cost of the computer equipment. Based on this, Figure 2 As shown, the computer device can preferentially perform speech recognition on the speech information to be recognized according to the speech recognition model to obtain the corresponding preset recognition result. When it is determined that the speech information to be recognized meets the preset re-recognition condition according to the speech type corresponding to the preset recognition result, the speech information to be recognized is then re-recognized through the speech recognition large model to obtain the corresponding target recognition result.

[0095] Accordingly, when it is determined that the voice information to be recognized does not meet the preset re-recognition conditions according to the voice type corresponding to the preset recognition result, the computer device does not need to perform voice re-recognition on the voice information to be recognized through the large voice recognition model, and can determine the preset recognition result corresponding to the voice information to be recognized determined by the voice recognition model as the target recognition result corresponding to the voice information to be recognized.

[0096] Exemplarily, a first training sample set is obtained, the first training sample set includes multiple first voice information; based on the first training sample set, unsupervised pre-training is performed on the trained voice recognition large model to obtain the pre-trained voice recognition large model; a second training sample set is obtained, the second training sample set includes multiple second voice information and first recognition result labels corresponding to each second voice information; the number of second voice information is less than the number of first voice information; based on the pre-trained voice recognition large model, voice recognition is performed on the second voice information to obtain the first recognition result corresponding to the second voice information; based on the first recognition result and the first recognition result label, supervised training is performed on the pre-trained voice recognition large model to obtain the trained voice recognition large model.

[0097] For example, when training a large speech recognition model, a first training sample set and a second training sample set may be obtained. The first training sample set may be used for unsupervised pre-training of the large speech recognition model, and the second training sample set may be used for supervised fine-tuning training of the large speech recognition model.

[0098] The first training sample set may include multiple first voice information. The first voice information is unlabeled voice information. The voice type corresponding to the first voice information may include at least one of carrying a dialect, carrying an accent, carrying noise, a speech speed greater than or equal to a preset speech speed threshold, and a number of languages ​​greater than or equal to a preset number threshold.

[0099] For example, when training a large speech recognition model, it is necessary to learn how to extract a better feature expression corresponding to the first speech information. Figure 3 As shown, the computer device can perform unsupervised pre-training on the speech recognition large model according to the first training sample set to obtain the pre-trained speech recognition large model. For example, the computer device can use a self-supervised learning algorithm, such as the HuBERT algorithm, the Wav2Vec algorithm, etc., to pre-train the preset speech recognition large model to obtain the pre-trained speech recognition large model, which is not limited here.

[0100] The second training sample set may include multiple second voice information and first recognition result labels corresponding to each second voice information. The voice type corresponding to the second voice information may include voice information with dialect, accent, noise, speech speed greater than or equal to a preset speech speed threshold, and the number of languages ​​greater than or equal to a preset number threshold. The number of second voice information is less than or equal to the number of first voice information.

[0101] In an exemplary embodiment, when the second voice information is obtained, the first recognition result label corresponding to the second voice information can be determined by a preset annotation method, which is not limited here.

[0102] For example, when training a large speech recognition model, it is necessary to learn how to perform speech recognition on the second speech information. Figure 3 As shown, the computer device can perform supervised training, such as fine-tuning training, on the pre-trained speech recognition large model based on the second training sample set to obtain a trained speech recognition large model. The computer device can determine the first recognition result corresponding to the second voice information through the pre-trained speech recognition large model, and adjust and optimize the model parameters of the speech recognition large model in combination with the first recognition result and the first recognition result label, so as to obtain a trained speech recognition large model. For example, the computer device can use a CTC ASR model or other non-autoregressive model to perform supervised fine-tuning on the pre-trained speech recognition large model, which is not limited here.

[0103] In an exemplary embodiment, when the first recognition result corresponding to the second voice information is determined by the speech recognition large model, the first recognition result is equivalent to the prediction result of the speech recognition large model for the target recognition result corresponding to the second voice information. The first recognition result label corresponding to the second voice information is equivalent to the actual result of the speech recognition large model for the target recognition result corresponding to the second voice information. Based on the first recognition result and the first recognition result label, the loss value between the actual result and the predicted result of the speech recognition large model when determining the target recognition result corresponding to the second voice information can be determined.

[0104] The speech recognition model can determine the model optimization parameters of the speech recognition model according to the loss value. The model optimization parameters can be used to subsequently adjust the model parameters of the speech recognition model, so that the speech recognition model can more accurately move the prediction result of the speech recognition model for the target recognition result corresponding to the second speech information closer to the actual result, thereby achieving the purpose of training.

[0105] In this way, the computer device trains the speech recognition large model through the first training sample set and the second training sample set, which is helpful to improve the speech recognition accuracy and convenience of the speech recognition large model for speech information with at least one of the following speech types: speech information with dialects, speech accents, speech noise, speech speed greater than or equal to a preset speech speed threshold, and language number greater than or equal to a preset number threshold.

[0106] The speech recognition method provided in the above embodiment obtains the speech information to be recognized; based on the trained speech recognition model, the speech information to be recognized is subjected to speech recognition to obtain the preset recognition result corresponding to the speech information to be recognized; according to the preset recognition result, the speech type of the speech information to be recognized is determined; when it is determined according to the speech type that the speech information to be recognized meets the preset re-recognition condition, the speech information to be recognized is subjected to speech recognition based on the trained speech recognition large model to obtain the target recognition result corresponding to the speech information to be recognized; wherein, the training data of the speech recognition large model is different from the training data of the speech recognition model, so as to improve the speech recognition convenience and speech recognition accuracy of speech information involving complex speech situations such as dialects, accents, and noise. For example, it can be used for customer service business in the financial field. For customer service business, when the customer service receives a call from a customer, the computer device can preferentially perform speech recognition on the speech information to be recognized corresponding to the telephone communication process between the customer service and the customer through the trained speech recognition model to reduce the computing resources required for speech recognition. Accordingly, when it is determined that the voice type of the voice information to be recognized meets the re-recognition conditions according to the preset recognition results output by the voice recognition model, such as involving dialects, accents, noise, etc., the computer device can further perform voice recognition on the voice information to be recognized through the trained voice recognition model to obtain the corresponding target recognition results, so as to more accurately recognize the problems described by the customer through voice, so that the customer service can provide better customer service to the customer.

[0107] See also Figure 4 , Figure 4 1 is a schematic block diagram of a speech recognition device provided in an embodiment of the present application. The speech recognition device can be configured in a server or an electronic device to execute the above-mentioned speech recognition method.

[0108] like Figure 4 As shown, the speech recognition device includes: an acquisition module 110 , a first recognition module 120 , a determination module 130 and a second recognition module 140 .

[0109] The acquisition module 110 is used to acquire the voice information to be recognized.

[0110] The first recognition module 120 is used to perform speech recognition on the speech information to be recognized based on the trained speech recognition model to obtain a preset recognition result corresponding to the speech information to be recognized.

[0111] The determination module 130 is used to determine the voice type of the voice information to be recognized according to the preset recognition result.

[0112] The second recognition module 140 is used to perform speech recognition on the speech information to be recognized based on a trained speech recognition large model when it is determined that the speech information to be recognized meets the preset re-recognition conditions according to the speech type, so as to obtain a target recognition result corresponding to the speech information to be recognized; wherein the training data of the speech recognition large model is different from the training data of the speech recognition model.

[0113] It should be noted that technicians in the relevant field can clearly understand that for the convenience and simplicity of description, the specific working processes of the above-described devices and modules and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0114] The method of the present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld devices or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronic devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0115] Exemplarily, the above method and apparatus may be implemented in the form of a computer program, which may be run on a computer device.

[0116] See also Figure 5 , Figure 5 1 is a schematic block diagram of a computer device provided in an embodiment of the present application. The computer device may be a server or an electronic device.

[0117] like Figure 5 As shown, the computer device includes a processor, a memory, and a network interface connected via a system bus, wherein the memory may include a storage medium and an internal memory.

[0118] The storage medium can store an operating system and a computer program. The computer program includes program instructions, and when the program instructions are executed, the processor can execute any speech recognition method.

[0119] The processor is used to provide computing and control capabilities and support the operation of the entire computer equipment.

[0120] The internal memory provides an environment for the operation of the computer program in the storage medium. When the computer program is executed by the processor, the processor can execute any speech recognition method.

[0121] The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art will understand that Figure 5 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0122] It should be understood that the processor may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0123] In one embodiment, the processor is used to execute a computer program and implement the following steps when executing the computer program:

[0124] Obtaining voice information to be recognized;

[0125] Based on the trained speech recognition model, speech recognition is performed on the speech information to be recognized to obtain a preset recognition result corresponding to the speech information to be recognized;

[0126] Determining the voice type of the voice information to be recognized according to the preset recognition result;

[0127] When it is determined according to the voice type that the voice information to be recognized meets the preset re-recognition conditions, voice recognition is performed on the voice information to be recognized based on the trained voice recognition large model to obtain a target recognition result corresponding to the voice information to be recognized; wherein the training data of the voice recognition large model is different from the training data of the voice recognition model.

[0128] It should be noted that those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of speech recognition described above can refer to the corresponding process in the aforementioned speech recognition method embodiment, and will not be repeated here.

[0129] The embodiment of the present application also provides a computer-readable storage medium, on which a computer program is stored. The method implemented when the computer program is executed by a processor can refer to the various embodiments of the speech recognition method of the present application.

[0130] The computer-readable storage medium may be an internal storage unit of the computer device described in the foregoing embodiment, such as a hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, a smart memory card (SmartMedia Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc., equipped on the computer device.

[0131] It should be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0132] It should also be understood that the term "and / or" used in the specification of this application and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, including these combinations. It should be noted that, in this article, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "including a..." does not exclude the presence of other identical elements in the process, method, article or system including the element.

[0133] It should also be understood that the non-Company software tools or components appearing in the embodiments of the present application are merely examples and do not represent actual use.

[0134] The serial numbers of the embodiments of the present application are for description only and do not represent the advantages and disadvantages of the embodiments. The above description is only a specific implementation mode of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present application, and these modifications or replacements should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A speech recognition method, characterized in that: include: Obtaining voice information to be recognized; Based on the trained speech recognition model, speech recognition is performed on the speech information to be recognized to obtain a preset recognition result corresponding to the speech information to be recognized; Determining the voice type of the voice information to be recognized according to the preset recognition result; When it is determined according to the voice type that the voice information to be recognized meets the preset re-recognition conditions, voice recognition is performed on the voice information to be recognized based on the trained voice recognition large model to obtain a target recognition result corresponding to the voice information to be recognized; wherein the training data of the voice recognition large model is different from the training data of the voice recognition model.

2. The speech recognition method according to claim 1, characterized in that: The determining, according to the voice type, that the voice information to be recognized meets a preset re-recognition condition includes: When the voice type includes at least one of carrying a dialect, carrying an accent, carrying noise, a speaking speed greater than or equal to a preset speaking speed threshold, and a number of languages ​​greater than or equal to a preset number threshold, it is determined that the voice information to be recognized meets the preset re-recognition condition.

3. The speech recognition method according to claim 2, characterized in that: The step of determining the voice type of the voice information to be recognized according to the preset recognition result includes: Based on the confidence evaluation model, a confidence evaluation is performed on the preset recognition result to obtain a corresponding confidence score; When the confidence score is less than or equal to a preset confidence threshold, it is determined that the voice type of the voice information to be recognized includes at least one of carrying a dialect, carrying an accent, carrying noise, a speaking speed greater than or equal to a preset speaking speed threshold, and a number of languages ​​greater than or equal to a preset number threshold.

4. The speech recognition method according to claim 2, characterized in that: The step of determining the voice type of the voice information to be recognized according to the preset recognition result includes: Based on the text analysis model, the semantic relevance of adjacent texts in the preset recognition result is analyzed to obtain the relevance between the adjacent texts; When the ratio between the number of adjacent texts whose correlation is less than or equal to a preset correlation threshold and the total number of texts in a preset recognition result is greater than or equal to a preset ratio threshold, it is determined that the voice type of the voice information to be recognized includes at least one of carrying a dialect, carrying an accent, carrying noise, and a speaking speed greater than or equal to a preset speaking speed threshold.

5. The speech recognition method according to claim 2, characterized in that: The step of determining the voice type of the voice information to be recognized according to the preset recognition result includes: When there is text in the preset recognition result that is a regional language text, and the regional language indicated by the regional language text is different from the preset language corresponding to the speech recognition model, it is determined that the voice type of the speech information to be recognized includes a language quantity greater than or equal to a preset quantity threshold.

6. The speech recognition method according to any one of claims 1 to 5, characterized in that: Before performing speech recognition on the speech information to be recognized based on the trained speech recognition large model to obtain a target recognition result corresponding to the speech information to be recognized, the method further includes: Acquire a first training sample set, where the first training sample set includes a plurality of first voice information; According to the first training sample set, unsupervised pre-training is performed on the trained speech recognition large model to obtain a pre-trained speech recognition large model; Acquire a second training sample set, wherein the second training sample set includes a plurality of second voice information and a first recognition result label corresponding to each of the second voice information; the number of the second voice information is less than the number of the first voice information; Based on the pre-trained speech recognition large model, perform speech recognition on the second speech information to obtain a first recognition result corresponding to the second speech information; According to the first recognition result and the first recognition result label, supervised training is performed on the pre-trained speech recognition large model to obtain a trained speech recognition large model.

7. The speech recognition method according to any one of claims 1 to 5, characterized in that: Before performing speech recognition on the speech information to be recognized based on the trained speech recognition model to obtain a preset recognition result corresponding to the speech information to be recognized, the method further includes: Acquire a third training sample set, wherein the third training sample set includes a plurality of third voice information and a second recognition result label corresponding to each of the third voice information; Based on the trained speech recognition model, performing speech recognition on the third speech information to obtain a second recognition result corresponding to the third speech information; Determining model optimization parameters of the speech recognition model according to the second recognition result and the second recognition result label; According to the model optimization parameters, the model parameters of the speech recognition model are adjusted to obtain a trained speech recognition model.

8. A speech recognition device, characterized in that: The speech recognition device comprises: An acquisition module, used to acquire voice information to be recognized; A first recognition module, configured to perform speech recognition on the speech information to be recognized based on a trained speech recognition model, and obtain a preset recognition result corresponding to the speech information to be recognized; A determination module, used to determine the voice type of the voice information to be recognized according to the preset recognition result; The second recognition module is used to perform speech recognition on the speech information to be recognized based on a trained speech recognition large model when it is determined that the speech information to be recognized meets the preset re-recognition conditions according to the speech type, so as to obtain a target recognition result corresponding to the speech information to be recognized; wherein the training data of the speech recognition large model is different from the training data of the speech recognition model.

9. A computer device, characterized in that: The computer device includes a memory and a processor; The memory is used to store computer programs; The processor is used to execute the computer program and implement the speech recognition method according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the speech recognition method according to any one of claims 1 to 7 are implemented.

Images