Speech recognition method, electronic equipment and storage medium

Abstract

The invention provides a voice recognition method, electronic equipment and a storage medium. The method comprises the following steps: acquiring text data generated when a voice recognition model recognizes to-be-recognized voice data; a professional term library of a target industry to which the to-be-recognized voice data belongs is acquired, and the professional term library comprises a plurality of standard professional terms of the target industry; and correcting the text data by using each standard professional term in a professional term library of the target industry. According to the method, each standard professional term in the professional term library of the target industry to which the to-be-recognized voice data belongs is utilized to correct the text data recognized by the voice recognition model, so that the corrected text data is obviously higher in accuracy, and the problems in the prior art are solved.

Description

Technical Field

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

[0002] With the development of science and technology, speech recognition technology has been applied to many industries. In the process of speech recognition, it is often necessary to convert the voice data input by the user into text data.

[0003] At present, the way to convert voice data into text data is usually to use a speech recognition model to process the voice data and then output the corresponding text data. However, the current speech recognition model is usually a broad spectrum model (also called a general model). The disadvantage of this type of model is that it has poor recognition effect on professional terms in specific industries (such as professional vocabulary, special nouns, Chinese abbreviations or English abbreviations in the industry). Therefore, the text data obtained by using the speech recognition model to recognize voice data is prone to low accuracy. Summary of the invention

[0004] The purpose of the embodiments of the present application is to provide a speech recognition method, an electronic device and a storage medium, so as to solve the problem of low speech recognition accuracy in the prior art.

[0005] A first aspect of an embodiment of the present application provides a speech recognition method, the method comprising:

[0006] Obtain text data generated by the speech recognition model for recognizing the speech data to be recognized;

[0007] Acquire a professional term library of the target industry to which the voice data to be recognized belongs, wherein the professional term library includes a plurality of standard professional terms of the target industry;

[0008] The text data is modified by using various standard professional terms in the professional term library of the target industry.

[0009] Preferably, before modifying the text data using each standard professional term in the professional term library of the target industry, the method further comprises: determining whether the semantics of the text data are smooth; and,

[0010] Using various standard professional terms in the professional term library of the target industry, the text data is modified, specifically including:

[0011] In the case where the semantics of the text in the text data are not smooth, the text data is corrected using various standard professional terms in the professional term library of the target industry.

[0012] Preferably, judging whether the semantics of the text data are smooth specifically includes:

[0013] Obtaining each sentence containing professional terms in the text data;

[0014] The semantic big model is used to determine whether the semantics of each sentence containing professional terms are smooth. If the semantics of each sentence containing professional terms are smooth, the semantics of the text data are smooth; or, if the semantics of at least one sentence containing professional terms are not smooth, the semantics of the text data are not smooth.

[0015] Preferably, when the semantics of the text in the text data are not smooth, the text data is corrected using various standard professional terms in the professional term library of the target industry, specifically including:

[0016] From each sentence containing professional terms, obtain sentences that are semantically incoherent;

[0017] The professional terms in the semantically incoherent sentence are matched with each standard professional term in the professional term library to obtain the standard professional term with the highest similarity to the professional term;

[0018] The professional terms in the semantically incoherent sentences are corrected by using the standard professional terms with the highest similarity, so as to achieve correction of the text data.

[0019] Preferably, the professional terms in the semantically incoherent sentence are matched with each standard professional term in the professional term library respectively by similarity, specifically including:

[0020] Converting the professional terms in the semantically incoherent sentence into a first vector;

[0021] The first vector is matched with each second vector in the vector database respectively in terms of similarity, wherein each second vector in the vector database is generated by using corresponding standard professional terms in the professional term library respectively.

[0022] Preferably, the method further comprises:

[0023] Generate corresponding second vectors for each standard professional term in the professional term library in advance;

[0024] Each standard professional term and the generated second vector are stored in the Milvus vector database accordingly;

[0025] A vector index is created in the Milvus vector database.

[0026] Preferably, the professional terms in the semantically incoherent sentence are matched with each standard professional term in the professional term library respectively by similarity, specifically including:

[0027] For each standard professional term, the intersection and union between the professional term and the standard professional term are obtained, and the ratio of the number of characters in the intersection and the number of characters in the union is calculated, and the ratio is used as the similarity between the professional term and the standard professional term.

[0028] Preferably, the method further comprises:

[0029] Feedback the corrected text data to the client.

[0030] A second aspect of an embodiment of the present application provides an electronic device, including:

[0031] Memory for storing computer programs;

[0032] A processor is used to execute any one of the methods described in the method embodiments of the present application.

[0033] A third aspect of an embodiment of the present application provides a storage medium, including: a program, which, when executed on an electronic device, enables the electronic device to execute any one of the methods in the method embodiments of the present application.

[0034] The method provided by the embodiment of the present application includes obtaining text data generated by the speech recognition model for recognizing speech data to be recognized, then obtaining a professional terminology library of the target industry to which the speech data to be recognized belongs, the professional terminology library including multiple standard professional terms of the target industry, and then using each standard professional terminology in the professional terminology library of the target industry to correct the text data. Since the method uses each standard professional terminology in the professional terminology library of the target industry to which the speech data to be recognized belongs to correct the text data recognized by the speech recognition model, the corrected text data is obviously more accurate, thereby solving the problems in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments of the present application will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0036] Figure 1 A schematic diagram of the structure of an electronic device provided in one embodiment of the present application.

[0037] Figure 2 A schematic diagram of a specific flow chart of a speech recognition method provided in one embodiment of the present application;

[0038] Figure 3 A schematic diagram of the specific structure of a speech recognition device provided in another embodiment of the present application. DETAILED DESCRIPTION

[0039] The technical solutions in the embodiments of the present application will be described below in conjunction with the accompanying drawings in the embodiments of the present application. In the description of the present application, terms such as "first", "second", "third", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance or sequence.

[0040] As mentioned above, the current method of converting voice data into text data is usually to use a speech recognition model to process the voice data and then output the corresponding text data. However, the current speech recognition model is usually a broad spectrum model (also called a general model). The disadvantage of this type of model is that the recognition effect of professional terms in specific industries is poor. Therefore, the text data obtained by using the speech recognition model to recognize voice data is prone to low accuracy.

[0041] For example, the same English abbreviations, Chinese abbreviations, etc. usually have completely different meanings in different industries. The current wide-spectrum models are usually difficult to distinguish, resulting in poor recognition of professional terms in specific industries. This is a shortcoming of the wide-spectrum model itself. Since this type of model is currently used directly to recognize voice data to obtain text data, it is easy to have a low accuracy rate in voice recognition.

[0042] In view of this, the embodiments of the present application provide a speech recognition method, device, electronic device and storage medium, which can be used to solve the technical problem. Figure 1 FIG. 1 is a schematic diagram of a specific structure of an electronic device provided in this embodiment. The electronic device 1 includes: at least one processor 11 and a memory 12. Figure 1 A processor is used as an example. The processor 11 and the memory 12 may be connected via a bus 10, and the memory 12 stores instructions that can be executed by the processor 11, and the instructions are executed by the processor 11, so that the electronic device 1 can execute all or part of the process of the method in the embodiment of the present application.

[0043] The electronic device 1 may be a client electronic device, such as a user's mobile phone, computer, etc. The electronic device 1 may also be a server electronic device, such as a server node, etc.

[0044] In one embodiment, the electronic device 1 serves as a client electronic device (hereinafter referred to as the client), so that the user can input voice data to the client, so that the client can execute the method provided in the embodiment of the present application to recognize the voice data.

[0045] In one embodiment, the electronic device 1 serves as a server-side electronic device (hereinafter referred to as a server). At this time, the server can connect to multiple clients, thereby being able to obtain voice data sent by the user through the client and execute the method provided in the embodiment of the present application.

[0046] like Figure 2 The figure shows a specific flow chart of the speech recognition method provided by the embodiment of the present application. The method can be performed by Figure 1 The method is performed by the electronic device 1 shown in the figure. Here, the method is described by taking the electronic device 1 as a server to perform the method. In this case, the method includes the following steps:

[0047] Step S21: Acquire text data generated by the speech recognition model through recognition of the speech data to be recognized.

[0048] The speech recognition model here can be an existing speech recognition model, such as an existing speech recognition software, or a large model such as GPT, which can be used to recognize the speech data to be recognized, thereby obtaining the corresponding text data.

[0049] For example, the speech data to be recognized can be input into the speech recognition model, so that the speech recognition model can process the speech data to be recognized and then output the text data. Of course, as mentioned above, the existing speech recognition model is a wide-spectrum model (also called a general model), and its recognition effect of professional terms in specific industries is poor. The method provided in the embodiment of the present application is based on the text data recognized by the speech recognition model, and then further corrects the text data to improve its accuracy.

[0050] As for the specific implementation method of step S21, for example, the voice data input by the user may be obtained, and the voice data may be used as the voice data to be recognized, and then input into the voice recognition model for recognition, so as to obtain the text data output by the voice recognition model.

[0051] For example, in one embodiment, a user sends voice data to a server through a client. After receiving the voice data, the server uses the voice data as voice data to be recognized and inputs it into a voice recognition model for recognition, thereby obtaining text data output by the voice recognition model.

[0052] Of course, in another embodiment, since the server is usually connected to multiple clients, after receiving the voice data sent by multiple clients, the server can input them into the voice recognition model for recognition, and then obtain the text data output by the voice recognition model. At this time, since there are text data corresponding to multiple voice data, a text data queue can be constructed, and the text data corresponding to these voice data can be stored in the text data queue in sequence. In this way, in step S21, the corresponding text data can be obtained from the text data queue in sequence. In this way, since the recognition of the voice data by the voice recognition model and the acquisition of the text data by the server are asynchronous processing, the computing pressure of the server can be reduced when the number of clients is large.

[0053] Step S22: Obtain a professional terminology library of the target industry to which the voice data to be recognized belongs.

[0054] The professional term library includes a plurality of standard professional terms of the target industry.

[0055] In actual applications, corresponding professional terminology libraries can be established for multiple industries in advance, and the professional terminology library includes multiple standard professional terms for the corresponding industries. For example, the standard professional terms can be professional vocabulary, special terms, Chinese abbreviations or English abbreviations of the industry standards. For example, for a certain industry, the professional vocabulary, special terms, Chinese abbreviations or English abbreviations of the industry standards can be obtained, and then the professional vocabulary, special terms, Chinese abbreviations or English abbreviations of the industry standards can be used to establish the professional terminology library corresponding to the industry.

[0056] In step S22, the industry to which the voice data to be recognized belongs (ie, the target industry) may be determined first, and then a professional term library corresponding to the target industry may be obtained from professional term libraries corresponding to a plurality of industries.

[0057] Among them, for the method of determining the target industry to which the voice data to be recognized belongs, the target industry can be determined by interaction with the user. For example, before or after recognizing the voice data to be recognized, the user can be prompted to select an industry as the target industry to which the voice data to be recognized belongs. In this way, the target industry to which the voice data to be recognized belongs can be determined based on the industry selected by the user. Another method is to determine the target industry based on the user's portrait. For example, considering that the user's daily behavior has a certain relevance and continuity, such as football fans tend to search (such as voice search) for football-related information, and the user portrait can reflect this relevance and continuity to a certain extent. Therefore, the target industry is determined based on the user's portrait. Of course, after obtaining the text data of the voice data to be recognized through the above-mentioned step S21, the target industry to which the voice data to be recognized belongs can also be determined based on the text data, such as identifying multiple professional terms in the text data, and then determining the target industry based on these professional terms.

[0058] Step S23: using various standard professional terms in the professional term library of the target industry to modify the text data.

[0059] There are many specific implementation methods for step S23 in actual applications. For example, each professional term in the text data can be directly obtained. At this time, since the text data is generated by the speech recognition model to recognize the speech data to be recognized, each professional term in the text data often has inaccurate recognition, such as misunderstanding of the meaning of the professional term. In this way, in the first implementation method of step S23, each professional term in the text data is directly obtained, and then for each professional term, the professional term is matched with each standard professional term in the professional term library for similarity, so as to obtain the standard professional term with the highest similarity to the professional term, and then use the standard professional term with the highest similarity to correct the professional term, such as replacing the professional term with the standard professional term with the highest similarity, so as to correct the text data.

[0060] For example, the text data includes three professional terms, namely professional term 1, professional term 2 and professional term 3. At this time, the following operations can be performed on these three professional terms (professional term 1 is taken as an example here). The professional term 1 can be matched with each standard professional term in the professional term library for similarity, so as to obtain the standard professional term with the highest similarity to the professional term 1, and then the professional term 1 can be replaced with the standard professional term with the highest similarity; similarly, the same processing can be performed on the professional term 2 and the professional term 3, so as to finally realize the correction of the text data.

[0061] Herein, how to perform similarity matching between the professional term and each standard professional term in the professional term library, and then obtain the standard professional term with the highest similarity to the professional term, will be described later.

[0062] Of course, after the text data is corrected to obtain the corrected text data, the method may further include feeding back the corrected text data to the client, so that the user can view the corrected text data through the client.

[0063] For step S23, in the first implementation method mentioned above, each professional term in the text data is directly obtained. However, in actual applications, for the text data generated by the speech recognition model recognizing the speech data to be recognized, even if the speech recognition model is a wide-spectrum model, there may be a situation where the recognition is accurate, or the situation where some sentences in the text data are accurately recognized. In this first implementation method, no distinction is made. In this way, the situation where the speech recognition model recognizes accurately or some sentences are accurately recognized is processed through the first implementation method, which is likely to lead to a decrease in recognition efficiency.

[0064] In view of this situation, a second implementation method can be further provided for step S23. In the second implementation method, before using the various standard professional terms in the professional term library of the target industry to correct the text data in step S23, it can be determined whether the semantics of the text data are smooth. If it is smooth, it means that the text data generated by the speech recognition model for the recognition of the speech data to be recognized is basically accurate in semantics, so there is no need to execute step S23, and the text data can be directly used as the final recognition result and fed back to the client; on the contrary, if it is determined that the semantics are not smooth, it means that the text data generated by the speech recognition model for the recognition of the speech data to be recognized is inaccurate or not completely accurate in semantics, so step S23 can be executed, that is, when it is determined that the semantics of the text in the text data are not smooth, step S23 is executed, so as to use the various standard professional terms in the professional term library of the target industry to correct the text data.

[0065] It should be noted that, for the method of judging whether the semantics of text data are smooth, for example, the sentences containing professional terms in the text data can be first obtained, for example, the sentences of the text data can be split to obtain the sentences in the text data, and then each sentence can be identified to see whether it contains professional terms, so as to extract each sentence containing professional terms; after obtaining each sentence containing professional terms in the text data, the semantic big model can be used to judge whether the semantics of each sentence containing professional terms are smooth, wherein the semantic big model can be, for example, GPT, Wenxinyiyan and other big models, and these sentences containing professional terms can be input into the semantic big model, so as to use the semantic big model to judge whether the semantics of the sentence containing professional terms are smooth.

[0066] In the embodiment of the present application, the reason why the sentence containing professional terms is directly input into the semantic big model instead of the text data is that the text length of the sentence containing professional terms is relatively short (relative to the text data), so the recognition accuracy is higher. Of course, the text data also includes sentences without professional terms. In practical applications, the accuracy of these sentences without professional terms is relatively high. Therefore, in the embodiment of the present application, the semantic big model is used to identify whether the semantics of the sentence containing professional terms are smooth.

[0067] For the recognition results, if the semantics of each sentence containing professional terms are smooth, it means that the semantics of the text data are smooth, and there is no need to correct the text data at this time; or, if the semantics of at least one sentence containing professional terms are not smooth, it means that the semantics of the text data are not smooth, and the text data needs to be corrected.

[0068] In addition, in the case where the semantics of the text in the text data are not smooth, the method provided in the embodiment of the present application can further utilize the various standard professional terms in the professional terminology library of the target industry to correct the text data. The specific method can be to first obtain the semantically incoherent sentences from each sentence containing the professional terminology. For example, as mentioned in the above-mentioned second implementation method of step S23, the sentences of the text data can be split to obtain each sentence in the text data, and then each sentence can be identified to see whether it contains professional terms, so as to extract each sentence containing professional terms. Then, the semantic big model can be used to determine whether the semantics of each sentence containing professional terms are smooth, and then the semantically incoherent sentences can be obtained from each sentence containing professional terms.

[0069] Then, the professional terms in the semantically incoherent sentence are further matched with each standard professional term in the professional term library for similarity, so as to obtain the standard professional term with the highest similarity to the professional term; specifically, for example, the professional terms in the semantically incoherent sentence can be converted into a first vector, and then the first vector is matched with each second vector in the vector database for similarity, wherein each second vector in the vector database is generated by the corresponding standard professional term in the professional term library. Among them, the first vector is matched with each second vector in the vector database for similarity, one way can be to calculate the pre-similarity of the two, and another processing method can be to obtain the intersection and union between the professional term and the standard professional term text for each standard professional term, and calculate the ratio of the number of characters in the intersection and the union, and use the ratio as the similarity between the professional term and the standard professional term. At this time, the higher the similarity, the greater the overlap between the professional term and the standard professional term text, which also reflects that the two are more similar, and vice versa.

[0070] In practical applications, the vector database may be a Milvus vector database, in which case corresponding second vectors may be generated in advance for each standard professional term in the professional term library, and then each standard professional term and the generated second vector may be stored in the Milvus vector database accordingly, and a vector index may be created in the Milvus vector database.

[0071] After matching the professional terms in the semantically incoherent sentences with the various standard professional terms in the professional term library to obtain the standard professional terms with the highest similarity to the professional terms, the standard professional terms with the highest similarity can be further used to correct the professional terms in the semantically incoherent sentences. For example, the standard professional terms with the highest similarity can be used to replace the professional terms in the semantically incoherent sentences, thereby achieving correction of the text data.

[0072] The method provided by the embodiment of the present application includes obtaining text data generated by the speech recognition model for recognizing speech data to be recognized, then obtaining a professional terminology library of the target industry to which the speech data to be recognized belongs, the professional terminology library including multiple standard professional terms of the target industry, and then using each standard professional terminology in the professional terminology library of the target industry to correct the text data. Since the method uses each standard professional terminology in the professional terminology library of the target industry to which the speech data to be recognized belongs to correct the text data recognized by the speech recognition model, the corrected text data is obviously more accurate, thereby improving the accuracy of speech recognition and solving the problems in the prior art.

[0073] The method provided in the embodiment of the present application may further include, after obtaining the salesperson's comprehensive communication skills score through the above-mentioned step S23, determining whether the salesperson needs to receive communication skills training based on the salesperson's comprehensive communication skills score. For example, a threshold may be set, and the salesperson's comprehensive communication skills score may be compared with the threshold to determine whether the salesperson's comprehensive communication skills score is greater than the threshold. If it is greater than the threshold, it means that the salesperson's communication skills meet the requirements and no communication skills training is required. Otherwise, if it is less than or equal to the threshold, it means that the salesperson's communication skills do not meet the requirements and communication skills training is required accordingly.

[0074] In addition, in an embodiment of the present application, when it is determined that the salesperson needs to be trained in communication skills, the type of communication skills training for the salesperson can be further determined based on the salesperson's responsiveness score, affinity score, professionalism score, and solution integrity score. In an embodiment of the present application, the types of communication skills training may include responsiveness training, affinity training, professionalism training, and solution integrity training, wherein the responsiveness training is used to train the salesperson's responsiveness, such as training the salesperson on how to respond quickly to customer questions; the affinity training is used to train the salesperson's affinity, such as training the salesperson to use more encouraging words during conversations with customers; the professionalism training is used to train the salesperson's professionalism, such as requiring the salesperson to use professional terms more accurately; and the solution integrity training is used to train the salesperson on how to better present the key points of the solution to the user when communicating with customers.

[0075] For determining the type of communication skills training for a salesperson based on his / her responsiveness score, affinity score, professionalism score, and solution integrity score, for example, the responsiveness score, affinity score, professionalism score, and solution integrity score can be compared with the corresponding preset values, respectively, so as to screen out scores that are less than the corresponding preset values, and then determine the corresponding training type to train for the item. For example, if the responsiveness score is less than the corresponding preset value, it means that the salesperson needs responsiveness training, so responsiveness training is determined as the type of communication skills training for the salesperson, and similarly, other training types for the salesperson can be determined.

[0076] In actual applications, for each type of training, sales personnel can be provided with responsiveness training, affinity training, professional training, and solution integrity training through on-site guidance by experienced sales personnel, watching training videos, or other methods.

[0077] In one application scenario of the present application, for example, for newly hired sales personnel, the method provided in the embodiment of the present application can be used to determine the comprehensive score of the sales personnel's communication skills, and then based on the comprehensive score of the sales personnel's communication skills, it can be determined whether communication skills training is needed for the sales personnel, and the type of communication skills training for the sales personnel can be further determined.

[0078] Based on the same inventive concept as the communication skills assessment method provided in the embodiment of the present application, the embodiment of the present application also provides a speech recognition device. If there is any unclear point, please refer to the corresponding content of the method embodiment. Figure 3 FIG. 3 is a schematic diagram of a specific structure of the device 30, which includes: a first acquisition unit 301, a second acquisition unit 302 and a correction unit 303, wherein:

[0079] The first acquisition unit 301 is used to acquire text data generated by the speech recognition model when recognizing the speech data to be recognized;

[0080] A second acquisition unit 302 is used to acquire a professional term library of the target industry to which the to-be-recognized voice data belongs, wherein the professional term library includes a plurality of standard professional terms of the target industry;

[0081] The correction unit 303 is used to correct the text data by using various standard professional terms in the professional term library of the target industry.

[0082] The device 30 provided in the embodiment of the present application is adopted. Since the device 30 adopts the same inventive concept as the method provided in the embodiment of the present application, on the premise that the method can solve the technical problem, the device 30 can also solve the technical problem, which will not be repeated here.

[0083] The device 30 may further include a judgment unit for judging whether the semantics of the text data are smooth before the text data is modified using each standard professional term in the professional term library of the target industry; and

[0084] Using various standard professional terms in the professional term library of the target industry to correct the text data may specifically include: when the semantics of the text in the text data are not smooth, using various standard professional terms in the professional term library of the target industry to correct the text data.

[0085] Wherein, judging whether the semantics of the text data are smooth may specifically include:

[0086] Obtaining each sentence containing professional terms in the text data;

[0087] The semantic big model is used to determine whether the semantics of each sentence containing professional terms are smooth. If the semantics of each sentence containing professional terms are smooth, the semantics of the text data are smooth; or, if the semantics of at least one sentence containing professional terms are not smooth, the semantics of the text data are not smooth.

[0088] Wherein, in the case where the semantics of the text in the text data are not smooth, the text data is corrected by using various standard professional terms in the professional term library of the target industry, which may specifically include:

[0089] From each sentence containing professional terms, obtain sentences that are semantically incoherent;

[0090] The professional terms in the semantically incoherent sentence are matched with each standard professional term in the professional term library to obtain the standard professional term with the highest similarity to the professional term;

[0091] The professional terms in the semantically incoherent sentences are corrected by using the standard professional terms with the highest similarity, so as to achieve correction of the text data.

[0092] The similarity matching of the professional terms in the semantically incoherent sentence with each standard professional term in the professional term library may specifically include:

[0093] Converting the professional terms in the semantically incoherent sentence into a first vector;

[0094] The first vector is matched with each second vector in the vector database respectively in terms of similarity, wherein each second vector in the vector database is generated by using corresponding standard professional terms in the professional term library respectively.

[0095] The device 30 may also include a vector database construction unit for generating corresponding second vectors for each standard professional term in the professional term library in advance; storing each standard professional term and the generated second vector in a Milvus vector database; and creating a vector index in the Milvus vector database.

[0096] The similarity matching of the professional terms in the semantically incoherent sentence with each standard professional term in the professional term library may specifically include:

[0097] For each standard professional term, the intersection and union between the professional term and the standard professional term are obtained, and the ratio of the number of characters in the intersection and the number of characters in the union is calculated, and the ratio is used as the similarity between the professional term and the standard professional term.

[0098] The device 30 may further include a feedback unit, which is used to feed back the corrected text data to the client.

[0099] The embodiment of the present invention further provides a storage medium, including: a program, which, when running on an electronic device, enables the electronic device to execute all or part of the process of the method in the above embodiment. The storage medium may be a disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), a flash memory, a hard disk (HDD) or a solid-state drive (SSD). The storage medium may also include a combination of the above types of memories.

[0100] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A speech recognition method, It is characterized in that The method comprises: Obtain text data generated by the speech recognition model for recognizing the speech data to be recognized; Acquire a professional term library of the target industry to which the voice data to be recognized belongs, wherein the professional term library includes a plurality of standard professional terms of the target industry; The text data is modified by using various standard professional terms in the professional term library of the target industry.

2. The method according to claim 1, It is characterized in that Before modifying the text data using each standard professional term in the professional term library of the target industry, the method further includes: determining whether the semantics of the text data are smooth; and, Using various standard professional terms in the professional term library of the target industry, the text data is modified, specifically including: In the case where the semantics of the text in the text data are not smooth, the text data is corrected using various standard professional terms in the professional term library of the target industry.

3. The method according to claim 2, It is characterized in that Determining whether the semantics of the text data are smooth specifically includes: Obtaining each sentence containing professional terms in the text data; The semantic big model is used to determine whether the semantics of each sentence containing professional terms are smooth. If the semantics of each sentence containing professional terms are smooth, the semantics of the text data are smooth; or, if the semantics of at least one sentence containing professional terms are not smooth, the semantics of the text data are not smooth.

4. The method according to claim 3, It is characterized in that In the case where the semantics of the text in the text data are not smooth, the text data is corrected using various standard professional terms in the professional term library of the target industry, specifically including: From each sentence containing professional terms, obtain sentences that are semantically incoherent; The professional terms in the semantically incoherent sentence are matched with each standard professional term in the professional term library to obtain the standard professional term with the highest similarity to the professional term; The professional terms in the semantically incoherent sentences are corrected by using the standard professional terms with the highest similarity, so as to achieve correction of the text data.

5. The method according to claim 4, It is characterized in that The professional terms in the semantically incoherent sentence are matched with each standard professional term in the professional term library respectively by similarity, specifically including: Converting the professional terms in the semantically incoherent sentence into a first vector; The first vector is matched with each second vector in the vector database respectively in terms of similarity, wherein each second vector in the vector database is generated by using corresponding standard professional terms in the professional term library respectively.

6. The method according to claim 5, It is characterized in that The method further comprises: Generate corresponding second vectors for each standard professional term in the professional term library in advance; Each standard professional term and the generated second vector are stored in the Milvus vector database accordingly; A vector index is created in the Milvus vector database.

7. The method according to claim 4, It is characterized in that The professional terms in the semantically incoherent sentence are matched with each standard professional term in the professional term library respectively by similarity, specifically including: For each standard professional term, the intersection and union between the professional term and the standard professional term are obtained, and the ratio of the number of characters in the intersection and the number of characters in the union is calculated, and the ratio is used as the similarity between the professional term and the standard professional term.

8. The method according to claim 1, It is characterized in that The method further comprises: Feedback the corrected text data to the client.

9. An electronic device, It is characterized in that include: Memory for storing computer programs; A processor, configured to execute the method according to any one of claims 1 to 8.

10. A storage medium, It is characterized in that The invention comprises: a program, which, when running on an electronic device, enables the electronic device to execute the method as claimed in any one of claims 1 to 8.

Images