Speech recognition method and device, equipment and storage medium

Abstract

The invention discloses a speech recognition method and device, equipment and a storage medium. The method comprises the following steps: providing a display interface of a plurality of data options; the plurality of data options comprise scene options and dialect options; in response to a first selection operation of the user on the scene selection item and the dialect selection item through the display interface, providing a target recognition model matched with the selected target scene and target dialect for the user; and recognizing to-be-recognized voice input by the user by using the target recognition model to obtain a corresponding recognition text. In this way, the accuracy of speech recognition can be 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] The role of speech recognition technology in some special fields has become a current research hotspot. Taking the medical field as an example, speech recognition technology is increasingly used in medical scenarios such as medical text information entry, medical equipment interaction, guidance services, and post-diagnosis follow-up, greatly reducing the cost of communication between hospitals, doctors, and patients. However, general speech recognition models are difficult to meet the needs of complex and changing medical application scenarios, various professional vocabulary, and dialects in different regions. Among them, how to improve the recognition effect of speech to be recognized in different scenarios and different dialects is particularly important. Summary of the invention

[0003] The main technical problem solved by the present application is to provide a speech recognition method, apparatus, device and storage medium, which can improve the accuracy of speech recognition.

[0004] In order to solve the above technical problems, a technical solution adopted in the present application is: to provide a speech recognition method, the method comprising: providing a display interface for several data selection items; the several data selection items include scene selection items and dialect selection items; in response to the user's first selection operation on the scene selection items and the dialect selection items through the display interface, providing the user with a target recognition model matching the selected target scene and target dialect; using the target recognition model to recognize the speech to be recognized input by the user to obtain the corresponding recognition text.

[0005] In order to solve the above technical problems, another technical solution adopted by the present application is: to provide a speech recognition device, including: an interface providing module, a model providing module and a speech recognition module. The interface providing module is used to provide a display interface for a number of data selection items; the number of data selection items include scene selection items and dialect selection items; the model providing module is used to respond to the user's first selection operation of the scene selection item and the dialect selection item through the display interface, and provide the user with a target recognition model that matches the selected target scene and target dialect; the speech recognition module is used to use the target recognition model to recognize the speech to be recognized input by the user, and obtain the corresponding recognition text.

[0006] To solve the above technical problems, another technical solution adopted in the present application is: to provide an electronic device, comprising a memory and a processor coupled to each other, the memory storing program instructions; the processor is used to execute the program instructions stored in the memory to implement the above method.

[0007] In order to solve the above technical problem, another technical solution adopted by the present application is: providing a computer-readable storage medium for storing program instructions, which can be executed to implement the above method.

[0008] The above scheme, after providing a display interface of several data selection items including scene selection items and dialect selection items, can respond to the user's selection operation of the scene selection items and the dialect selection items through the display interface, provide the user with a target recognition model that matches the selected target scene and target dialect, and after obtaining the speech to be recognized input by the user, use the target recognition model to recognize the speech to be recognized input by the user to obtain the corresponding recognition text. Compared with the method of performing speech recognition using the same speech recognition model without distinguishing between scenes and dialects, the above method of the present application can provide users with a speech recognition model suitable for recognizing speech in the target scene and target dialect, which can effectively reduce the phenomenon of misrecognition caused by using an unsuitable speech recognition model, and thus can improve the recognition accuracy of speech. Furthermore, users can customize a suitable target recognition model by selection, and this customized experience also enhances user satisfaction and trust. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 It is a flowchart of an embodiment of a speech recognition method provided by the present application;

[0010] Figure 2 This is a flowchart of an embodiment of customizing a second recognition model using first audio data provided by the present application;

[0011] Figure 3 is a schematic diagram of a flow chart of an embodiment of determining a second recognition model provided by the present application;

[0012] Figure 4 It is a schematic diagram of the framework of an embodiment of a speech recognition device provided by the present application;

[0013] Figure 5 It is a schematic diagram of a framework of an embodiment of an electronic device provided by the present application;

[0014] Figure 6 It is a schematic diagram of the framework of the computer-readable storage medium provided by this application. DETAILED DESCRIPTION

[0015] In order to make the purpose, technical solution and effect of the present application clearer and more specific, the present application is further described in detail below with reference to the accompanying drawings and examples.

[0016] In addition, if there are descriptions involving "first", "second", etc. in the embodiments of the present application, the descriptions of "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or suggesting their relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the technical solutions between the various embodiments can be combined with each other, but they must be based on the ability of ordinary technicians in the field to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such combination of technical solutions does not exist and is not within the scope of protection required by this application.

[0017] It should be noted that although speech recognition technology has made significant progress and can support the recognition of multiple languages, users in different regions are accustomed to using their own dialects when communicating, and existing speech recognition systems are mostly designed and trained based on common languages ​​(such as Mandarin), and have limited ability to recognize dialects, resulting in insufficient accuracy of speech recognition systems in dialect areas. In addition, in some special fields, such as the medical field, due to the complexity of professional terminology and the diversity of regional dialects, existing speech recognition systems often find it difficult to achieve ideal recognition results.

[0018] Furthermore, even in the same field, the requirements for speech recognition systems vary in different scenarios. For example, medical scenarios can generally be divided into telephone follow-up scenarios, doctor Q&A scenarios, data entry scenarios, etc. Among them, telephone follow-up scenarios may require the recognition system to have higher noise suppression capabilities, while doctor Q&A scenarios require the recognition system to be able to accurately recognize and understand complex medical terms. However, existing speech recognition systems often lack customized optimization for these specific scenarios.

[0019] In order to solve the above-mentioned problems, the present application proposes a speech recognition method, which can provide a user with a display interface of several data selection items including scene selection items and dialect selection items, and can respond to the user's selection operations on the scene selection items and the dialect selection items through the display interface, and provide the user with a target recognition model that matches the selected target scene and target dialect, so as to use the target recognition model that matches the target scene and target dialect selected by the user to recognize the speech to be recognized input by the user, and obtain the corresponding recognition text.

[0020] Specifically, see Figure 1 , Figure 1 is a flow chart of an embodiment of the speech recognition method provided by the present application. It should be noted that if there are substantially the same results, this embodiment does not take Figure 1 The process sequence shown is limited. Figure 1 As shown, this embodiment includes:

[0021] S11: providing a display interface for a number of data selection items; the number of data selection items include scene selection items and dialect selection items.

[0022] S12: In response to a first selection operation of a scene selection item and a dialect selection item by a user through a display interface, a target recognition model matching the selected target scene and target dialect is provided to the user.

[0023] S13: Using the target recognition model to recognize the speech to be recognized input by the user, and obtaining the corresponding recognition text.

[0024] This application provides a corresponding human-computer interaction interface, so that the user can select the application scenario of speech recognition and the dialect used by the user, so as to recognize the speech to be recognized by using the target recognition model selected by the user and matching the selected target scenario and target dialect. It can be understood that since the target recognition model selected by the user matches the scenario and dialect corresponding to the speech to be recognized, the phenomenon of misrecognition caused by using an unsuitable speech recognition model can be effectively reduced. Therefore, the recognition text obtained by recognizing the speech to be recognized by using the target recognition model selected by the user is more accurate and more in line with the user's expectations.

[0025] In one embodiment, step S21 provides a display interface for several data selection items and subsequent steps are executed by a voice recognition system, or by a device where the voice recognition system is located. The following description of the various embodiments of the present application will take the execution of the voice recognition system as an example.

[0026] Among them, in order to enable users to select the corresponding speech recognition model according to the dialect of their own area and the scene to be used, the speech recognition system can be set to include speech recognition models that match the scenes and dialects required by several different users, and a display interface that can provide several data selection items including scene selection items and dialect selection items is set, so that users can select a target recognition model that meets their needs. Among them, the speech recognition system of the present application includes target recognition models for identifying different dialects and different scenes, each target recognition model is trained using audio-related data under the corresponding scene and the corresponding dialect. After the user selects the desired target scene and target dialect, the user can be provided with a target recognition model that matches the selected scene and dialect for speech recognition.

[0027] It should be noted that this application can regard a field as a scenario, or several sub-scenarios under a field as a scenario. For example, each field in the medical field, the transportation field, and the teaching field can be regarded as a scenario, or telephone follow-up, doctor Q&A, and data entry in the medical field can be regarded as a scenario.

[0028] Among them, the provided scene selection items include several scene options that can be selected by users, and the provided dialect selection items include several dialect options that can be selected by users who use different dialects. The scene items included in the scene selection items and the dialect items included in the dialect selection items can be set according to actual conditions.

[0029] For example, the scenario options provided include telephone follow-up scenarios, doctor Q&A scenarios, and data entry scenarios, and the dialect options provided include Henan dialect, Sichuan dialect, and Cantonese. If the user using the speech recognition system is from Henan and the content of the communication is about the doctor Q&A scenario, in this case, the user can select the options corresponding to Henan dialect and the doctor Q&A scenario, so that the speech recognition system can respond to the user's selection and provide the user with a target recognition model that matches Henan dialect and the doctor Q&A scenario. The target recognition model is pre-trained using audio-related data corresponding to Henan dialect and the doctor Q&A scenario, and the audio-related data may include, but is not limited to, audio and text.

[0030] The user in this embodiment is a user of the speech recognition system. For example, in a doctor question and answer scenario, the user is a doctor or a patient.

[0031] The above scheme, after providing a display interface of several data selection items including scene selection items and dialect selection items, can respond to the user's selection operation of the scene selection items and the dialect selection items through the display interface, provide the user with a target recognition model that matches the selected target scene and target dialect, and after obtaining the speech to be recognized input by the user, use the target recognition model to recognize the speech to be recognized input by the user to obtain the corresponding recognition text. Compared with the method of performing speech recognition using the same speech recognition model without distinguishing between scenes and dialects, the above method of the present application can provide users with a speech recognition model suitable for recognizing speech in the target scene and target dialect, which can effectively reduce the phenomenon of misrecognition caused by using an unsuitable speech recognition model, and thus can improve the recognition accuracy of speech. Furthermore, users can customize a suitable target recognition model by selection, and this customized experience also enhances user satisfaction and trust.

[0032] In some embodiments, the target recognition model provided to the user is a first recognition model or a second recognition model in a preset recognition model library that matches the target scene and the target dialect.

[0033] It should be noted that the preset recognition model library of the present application stores several target recognition models for identifying speech to be recognized in different scenarios and dialects. For example, there are target recognition models that match Henan dialect and doctor question and answer scenarios, target recognition models that match Henan dialect and telephone follow-up scenarios, target recognition models that match Sichuan dialect and doctor question and answer scenarios, target recognition models that match Sichuan dialect and telephone follow-up scenarios, etc.

[0034] After the user selects the target scene and target dialect, the speech recognition system can provide the user with a target recognition model that matches the target scene and target dialect for speech recognition.

[0035] Among them, the first recognition model is a basic model (pre-trained) provided by the speech recognition system, and the second recognition model is a recognition model obtained by fine-tuning the first recognition model using the first audio data related to the target scene and target dialect input by the user.

[0036] In one embodiment, the second recognition model is a model customized by the user based on the first recognition model using the first audio data related to the target scene and the target dialect. It can be understood that if the user does not customize the second recognition model, the first recognition model is the target recognition model.

[0037] In one embodiment, in order to facilitate providing the first recognition model or the second recognition model that matches the target scene and the target dialect to the user from the preset recognition model library, the basic model (first recognition model) in the speech recognition system and the second recognition model customized by the user based on the basic model can be stored separately. Exemplarily, the preset recognition model library is set to include a first model library for storing the first recognition model and a second model library for storing the second recognition model.

[0038] Furthermore, after the user selects the target scene and the target dialect, the first recognition model or the second recognition model matching the target scene and the target dialect can be selected as the target recognition model according to the priority order of the first model library and the second model library.

[0039] Taking into account that the second recognition model is customized by the user based on the first recognition model and is more in line with the user's needs for speech recognition, the priority of the second model library can be set higher than that of the first model library. In the process of providing the user with a target recognition model that matches the selected target scene and target dialect, it is prioritized to determine whether there is a second recognition model that matches the selected target scene and target dialect in the second model library. If so, the second recognition model is selected as the target recognition model. Otherwise, the first recognition model that matches the target scene and target dialect is selected from the first model library as the target recognition model.

[0040] In order to enable the user to select the corresponding data processing method based on the speech recognition requirements of the target scene and the target dialect in the second recognition model customization process, the speech recognition system can be set to display at least one processing option for the user to select. In other words, the plurality of data options provided by the speech recognition system also includes at least one processing option, wherein each processing option corresponds to a data processing method.

[0041] Specifically, see Figure 2 , Figure 2 This is a flow chart of an embodiment of customizing a second recognition model using first audio data provided by the present application. This embodiment is performed before providing the user with a target recognition model matching the selected target scene and target dialect in step S12. This embodiment includes:

[0042] S21: Acquire first audio data input by a user.

[0043] This embodiment is used to process the first audio data input by the user using the data processing method selected by the user to obtain sample audio data, and use the sample audio data as training data to adjust the first recognition model to obtain the corresponding second recognition model, so as to enable the user to customize the second recognition model.

[0044] Each speech recognition model of the present application includes an acoustic model and a language model, wherein the acoustic model is used to extract features of the input audio (or speech) and generate the probability distribution of each phoneme or word, and the language model evaluates the probability of different text sequences based on the output of the acoustic model, and then outputs a text sequence that meets the probability requirements through decoding, and presents the text sequence to the user as a recognized text.

[0045] Specifically, the first recognition model includes a first acoustic model and a first language model, and the second recognition model includes a second acoustic model and a second language model. The second acoustic model and the second language model are obtained by adjusting the first recognition model using the first audio data input by the user.

[0046] In this embodiment, the first audio data includes at least one of the first audio, the annotated text corresponding to the first audio, and the sample corpus. The second acoustic model is obtained by adjusting the first acoustic model using the first audio, the second audio, the annotated text corresponding to the first audio, and the sample corpus, and the second audio is obtained by performing speech synthesis on the sample corpus using speech synthesis technology; the second language model is obtained by adjusting the first language model using the sample corpus and / or the annotated text of the first audio, wherein the first audio is an audio about the target scene output using the target dialect, and the annotated text corresponding to the first audio is a text in the common language corresponding to the target dialect. The sample corpus is a corpus about the target scene, for example, a text about a medical scene, or a text about a telephone follow-up scene or a doctor's question and answer scene or a data entry scene under a medical scene.

[0047] It should be noted that, in order to make the second acoustic model in the second recognition model finally customized to support audios of multiple audio formats, the input first audio and / or the synthesized second audio can be made to include audios of at least one audio format; similarly, in order to make the second language model in the second recognition model finally customized to support texts of multiple text formats, the input annotated text and / or sample corpus can be made to include corpus of at least one text format.

[0048] S22: In response to a second selection operation of the user on at least one processing selection item, the first audio data is processed using a data processing method associated with the target processing item selected by the user to obtain processed audio data.

[0049] In some embodiments, considering that data acquisition in some special scenarios is difficult or to improve the robustness and generalization of the model to be trained, the speech recognition system can be set to provide the user with at least one processing option for data processing, so that after the user selects the corresponding target processing option, the first audio data is processed using the data processing method associated with the target processing option to obtain processed audio data. It is understandable that in this embodiment, the user can customize the second recognition model using the set at least one processing option.

[0050] In one embodiment, at least one processing option includes at least one of a data enhancement option and a data mixing processing option. The data enhancement option includes a plurality of first selection sub-items corresponding to data enhancement methods, and the data mixing processing option includes a plurality of second selection sub-items regarding a mixing ratio between the first audio data and the second audio data, wherein the second audio data is audio-related data of a common language corresponding to the target dialect.

[0051] For example, considering that medical scenarios are low-resource scenarios involving user privacy information, it is particularly difficult to obtain medical scenario data. Therefore, when customizing the second recognition model, the user may want to increase the amount of data by means of data enhancement to train a model with higher recognition effect. Therefore, at least one processing option may include a data enhancement option.

[0052] In one embodiment, the first selection sub-items corresponding to the above-mentioned data enhancement methods may be: a selection sub-item for data enhancement by using speech synthesis, a selection sub-item for data enhancement by adding noise, or other selection sub-items for data enhancement.

[0053] As for the method of speech synthesis, speech technology can be used to synthesize audio from the text related to the target scene in the first audio data input by the user. When synthesizing audio about the target dialect, it can support multiple speakers, multiple timbres, multiple scenes, different speaking speeds, volumes and pitches, etc., to synthesize a variety of audio data.

[0054] As for the noise adding method, at least one noise adding method such as noise adding, reverberation, adjusting volume and controlling speech speed may be used to perform noise adding processing on the first audio data to obtain the audio data after noise adding processing.

[0055] Noise addition mainly superimposes noise data and the input original audio data to obtain new audio; the noise data can be white noise, human voice in the same scene, environmental noise, etc. The noise type can be selected according to the target application scenario to simulate different external environments; the operation of adding reverberation is similar to the operation of adding noise, which is to superimpose the rebound of sound in a limited space to the original audio; controlling the volume and speaking speed is mainly to simulate different speaking characteristics; through the above-mentioned noise addition method, not only the purpose of data enhancement can be achieved, but also more voice scenes, speaker characteristics, etc. can be taken into account, which plays a very important role in improving the robustness of the target recognition model.

[0056] In one embodiment, at least one processing option includes a data mixing processing item, which includes several second selection sub-items regarding the mixing ratio between the first audio data and the second audio data, so that the user can select the mixing ratio of the first audio data corresponding to the target dialect and the second audio data of the common language corresponding to the target dialect.

[0057] The above examples of data enhancement and data mixing are mainly used to illustrate the relevant processing required for audio. Before training the language model, the sample corpus input by the user can also be processed by data enhancement and / or data mixing according to the user's selection, and the steps of user selection processing of the audio and the steps of user selection processing of the sample corpus can be executed separately.

[0058] Among them, users can customize the generated text content according to the needs of the target scenario, dialogue framework or question set, and can input the collected multi-format text corpus of real scenarios and the generated text content as sample corpus into the speech recognition system. The speech recognition system can provide at least one display interface of processing options for users to make selections, such as providing an option of whether to mix the sample corpus with the general corpus, and providing an option of the mixing ratio of the sample corpus and the general corpus, or providing an option of whether to clean the sample corpus, etc., and even after obtaining the models obtained from each training in the training stage, it provides an option of whether to trim the model, and a selection operation of the target threshold for trimming, etc.

[0059] It should be noted that the at least one processing option provided above is used for the user to select a corresponding processing method when necessary. Of course, when the user does not select the corresponding processing option, the corresponding processing will not be performed.

[0060] S23: Using the processed audio data as sample audio data to adjust the first recognition model using the sample audio data to obtain a corresponding second recognition model.

[0061] Among them, the existing method can be used to adjust the first recognition model using the sample audio data to obtain the corresponding second recognition model.

[0062] In a specific embodiment, in the process of adjusting the first acoustic model in the first recognition model using sample audio data to obtain the corresponding second acoustic model, the FBank method can be used to extract audio features in the sample audio data to retain more original features. The main process is: pre-emphasize, frame, and add Hamming window processing to the audio signal, and then obtain the spectrum through short-time Fourier transform, and obtain the energy spectrum after squaring; superimpose the energy in each filter band, and take the logarithm of the filter output to obtain the logarithmic power spectrum of the corresponding frequency band, and align the obtained audio features with the features at the word level of the annotated text, and then train or adjust the acoustic model. Specifically: the audio features and annotated text are processed by the ctc decoding tool to obtain the text phone features and align them with the audio features.

[0063] In a specific embodiment, before and during the process of adjusting the first language model in the first recognition model using the sample audio data to obtain the corresponding second language model, the following processing is also included: first, generating and uploading corpus according to the specified scene, dialogue outline or specified question input by the user, and the number of texts generated by the user; second, determining the final training text according to the user's selection of the corpus required for this model adjustment, the number of copies of each corpus, whether to mix with the general corpus and the mixing ratio, whether to clean the text, and other options; third, generating 3gram and 4gram models according to the final training text and the internally integrated dictionary, and merging the 3gram and 4gram models with the general n-gram model according to the set ratio; fourth, serializing and packaging the model; for the first-pass language model, since the generated 4gram resources are too large, it will affect the CPU occupancy when the model is running, so the fused 4-gram is reduced to 3-gram. The 3gram model is packaged based on the dictionary and pronunciation dictionary to generate resources such as L and G networks, and the first-pass language model is obtained after serialization; for the second-pass language model, the 4-gra model and dictionary are used to generate the second-pass language model.

[0064] As can be seen from the above, the second recognition model is obtained by adjusting the first recognition model using the first audio data, wherein the first recognition model needs to be adjusted multiple times in the process of customizing the second recognition model. The recognition model obtained by adjusting the first recognition model each time is called the third recognition model. The final second recognition model is not necessarily the third recognition model obtained by the last adjustment, but may be the third recognition model obtained by an intermediate adjustment. However, it can be determined that the second recognition model is a recognition model selected from the third recognition models.

[0065] In one embodiment, in the process of adjusting the first recognition model using the first audio data input by the user to obtain the corresponding second recognition model, Figure 3 The second recognition model is determined in the manner of the illustrated embodiment.

[0066] Specifically, see Figure 3 , Figure 3 This is a flow chart of an embodiment of determining a second recognition model provided by the present application. This embodiment specifically includes:

[0067] S31: retaining a third recognition model obtained by adjusting the first recognition model in various ways.

[0068] S32: Select a model to be compared from each third recognition model in turn according to a preset priority order, and use the second audio data input by the user to perform effect tests on the model to be compared and the first recognition model respectively, to obtain a first recognition result of the model to be compared and a second recognition result of the first recognition model.

[0069] S33: Based on the comparison result of the first recognition result and the second recognition result, determine whether to use the model to be compared as the second recognition model.

[0070] It should be noted that although the model may be overfitted after undergoing a large number of adjustments, generally speaking, the more times the model is adjusted, the higher the recognition result of the corresponding adjusted model is. Therefore, in order to quickly obtain a second recognition model that meets the recognition effect requirements, the preset priority order can be set to be negatively correlated with the order of performing each adjustment. That is, after arranging each third recognition model in order from small to large according to the number of adjustments, each third recognition model can be selected from back to front as the model to be compared until it is determined that the currently selected model to be compared is the second recognition model.

[0071] In addition, it should be noted that the user customizes the second recognition model in the hope of adjusting it to obtain a model that is better than the corresponding first recognition model, so as to use the better second recognition model to better recognize the speech to be recognized. Therefore, in step S33, based on the comparison result of the first recognition result and the second recognition result, in the process of determining whether to use the model to be compared as the second recognition model, it can be determined whether the first recognition result of the model to be compared is better than the second recognition result of the first recognition model. If the first recognition result is better than the second recognition result, it is determined that the corresponding model to be compared is used as the second recognition model; if the first recognition result is worse than the second recognition result of the first recognition model, step S32 and subsequent steps are continued until a second recognition model whose first recognition result is better than the second recognition result is determined.

[0072] Furthermore, in one embodiment, the determination conditions of the second recognition model can be added to select a better second recognition model. Specifically, after determining that the first recognition result is better than the second recognition result, it is further determined whether the difference between the first recognition result and the second recognition result meets the preset difference requirement. If so, it is determined that the model to be compared will be used as the second recognition model.

[0073] In some embodiments, after determining the second recognition model that meets the above requirements, the first recognition model corresponding to the target scene and the target dialect may be deleted to reduce the space occupied by the first recognition model in the speech recognition system.

[0074] See also Figure 4 , Figure 4 It is a schematic diagram of the framework of an embodiment of a speech recognition device provided by the present application. In this embodiment, the speech recognition device 40 includes: an interface providing module 41, a model providing module 42 and a speech recognition module 43. The interface providing module 41 is used to provide a display interface for a number of data selection items; the several data selection items include scene selection items and dialect selection items; the model providing module 42 is used to respond to the user's first selection operation of the scene selection item and the dialect selection item through the display interface, and provide the user with a target recognition model that matches the selected target scene and target dialect; the speech recognition module 43 is used to use the target recognition model to recognize the speech to be recognized input by the user, and obtain the corresponding recognition text.

[0075] In some embodiments, the target recognition model provided by the model providing module 42 is a first recognition model or a second recognition model in a preset recognition model library that matches the target scene and the target dialect; the second recognition model is obtained by adjusting the first recognition model using the first audio data input by the user, and the first audio data is audio-related data about the target scene and the target dialect.

[0076] In some embodiments, the several data selection items provided by the interface providing module 41 also include at least one processing selection item; before the model providing module 42 provides the user with a target recognition model that matches the selected target scene and target dialect, it also includes: obtaining first audio data input by the user; in response to the user's second selection operation on at least one processing selection item, processing the first audio data using a data processing method associated with the target processing item selected by the user to obtain processed audio data; using the processed audio data as sample audio data to adjust the first recognition model using the sample audio data to obtain a corresponding second recognition model.

[0077] In some embodiments, at least one processing option includes at least one of a data enhancement option and a data mixing processing option; the data enhancement option includes a plurality of first selection sub-items corresponding to data enhancement methods, the data mixing processing option includes a plurality of second selection sub-items regarding a mixing ratio between the first audio data and the second audio data, and the second audio data is audio-related data of a common language corresponding to the target dialect;

[0078] In some embodiments, the first recognition model includes a first acoustic model and a first language model, and the first audio data includes a first audio, an annotated text corresponding to the first audio, and at least one of a sample corpus; the second recognition model includes a second acoustic model and a second language model, and the second acoustic model is obtained by adjusting the first acoustic model using the first audio, the second audio, the annotated text corresponding to the first audio, and at least one of the sample corpus, and the second audio is obtained by performing speech synthesis on the sample corpus; the second language model is obtained by adjusting the first language model using the sample corpus, wherein the sample corpus is corpus about the target scene.

[0079] In some embodiments, the first audio data includes at least one of the first audio, the annotated text corresponding to the first audio, and the sample corpus; the first audio is audio including at least one audio format, and the sample corpus is corpus including at least one text format.

[0080] In some embodiments, the second recognition model is obtained by adjusting the first recognition model multiple times using the first audio data; in the process of adjusting the first recognition model using the first audio data input by the user to obtain the corresponding second recognition model, it also includes: retaining the third recognition model obtained by adjusting the first recognition model each time; selecting the model to be compared from each third recognition model in turn according to the preset priority order, and using the second audio data input by the user to test the recognition effect of the model to be compared and the first recognition model respectively, to obtain the first recognition result of the model to be compared and the second recognition result of the first recognition model; based on the comparison result of the first recognition result and the second recognition result, determining whether to use the model to be compared as the second recognition model.

[0081] In some embodiments, the preset priority order is negatively correlated with the order in which each adjustment is performed, and based on the comparison result of the first recognition result and the second recognition result, it is determined whether to use the model to be compared as the second recognition model, including: in response to the first recognition result being better than the second recognition result, and the difference between the first recognition result and the second recognition result meeting the preset difference requirement, determining that the model to be compared is used as the second recognition model.

[0082] In some embodiments, the preset recognition model library includes a first model library for storing a first recognition model and a second model library for storing a second recognition model; the model providing module 42 provides the user with a target recognition model that matches the selected target scene and target dialect, including: selecting the first recognition model or the second recognition model as the target recognition model according to the priority order of the first model library and the second model library.

[0083] In some embodiments, the priority of the second model library is higher than that of the first model library; according to the priority order of the first model library and the second model library, the first recognition model or the second recognition model is selected as the target recognition model, including: determining whether there is a second recognition model matching the target scene and the target dialect in the second model library; in response to the existence of the second recognition model, the second recognition model is selected as the target recognition model, otherwise, the first recognition model matching the target scene and the target dialect is selected from the first model library as the target recognition model.

[0084] In some embodiments, the target scenario includes a medical scenario, or the target scenario includes one of a telephone follow-up scenario, a doctor question and answer scenario, and a data entry scenario under the medical scenario; and / or, a display interface for several data selection items is provided and subsequent steps are executed by a voice recognition system.

[0085] See also Figure 5 , Figure 5 1 is a schematic diagram of a framework of an electronic device according to an embodiment of the present application. In this embodiment, the electronic device 50 includes a memory 51 and a processor 52 coupled to each other.

[0086] The memory 51 stores program instructions, and the processor 52 is used to execute the program instructions stored in the memory 51 to implement the steps of any of the above method implementations. In a specific implementation scenario, the electronic device 50 may include, but is not limited to: a microcomputer, a server, and in addition, the electronic device 50 may also include a mobile device such as a laptop computer and a tablet computer, which is not limited here.

[0087] Specifically, the processor 52 is used to control itself and the memory 51 to implement the steps of any of the above-mentioned embodiments. The processor 52 can also be called a CPU (Central Processing Unit). The processor 52 may be an integrated circuit chip with signal processing capabilities. The processor 52 can also be a general-purpose processor, a digital signal processor (Digital Signal Processor, DSP), an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), a field-programmable gate array (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. In addition, the processor 52 can be implemented by an integrated circuit chip.

[0088] See also Figure 6 , Figure 6 It is a schematic diagram of the framework of the computer-readable storage medium provided by the present application. The computer-readable storage medium 60 of the embodiment of the present application stores a program instruction 61, and when the program instruction 61 is executed, the method provided by any embodiment of the above method and any non-conflicting combination is implemented. Among them, the program instruction 61 can form a program file and be stored in the above-mentioned computer-readable storage medium 60 in the form of a software product, so that a computer device (which can be a personal computer, a server, or a network device, etc.) executes all or part of the steps of each implementation method of the present application. The aforementioned computer-readable storage medium 60 includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a disk or an optical disk, or a terminal device such as a computer, a server, a mobile phone, and a tablet.

[0089] The above scheme, after providing a display interface of several data selection items including scene selection items and dialect selection items, can respond to the user's selection operation of the scene selection items and the dialect selection items through the display interface, provide the user with a target recognition model that matches the selected target scene and target dialect, and after obtaining the speech to be recognized input by the user, use the target recognition model to recognize the speech to be recognized input by the user to obtain the corresponding recognition text. Compared with the method of performing speech recognition using the same speech recognition model without distinguishing between scenes and dialects, the above method of the present application can provide users with a speech recognition model suitable for recognizing speech in the target scene and target dialect, which can effectively reduce the phenomenon of misrecognition caused by using an unsuitable speech recognition model, and thus can improve the recognition accuracy of speech. Furthermore, users can customize a suitable target recognition model by selection, and this customized experience also enhances user satisfaction and trust.

[0090] In some embodiments, the functions or modules included in the device provided by the embodiments of the present disclosure can be used to execute the method described in the above method embodiments. The specific implementation can refer to the description of the above method embodiments, and for the sake of brevity, it will not be repeated here.

[0091] The above description of various embodiments tends to emphasize the differences between the various embodiments. The same or similar aspects can be referenced to each other, and for the sake of brevity, they will not be repeated herein.

[0092] In the several embodiments provided in the present application, it should be understood that the disclosed methods and devices can be implemented in other ways. For example, the device implementation described above is only schematic. For example, the division of modules or units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

[0093] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0094] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0095] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) or a processor (processor) to perform all or part of the steps of each implementation method of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program code.

[0096] The above description is only an implementation method of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly used in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A speech recognition method, characterized in that: The method comprises: Providing a display interface for several data selection items; the several data selection items include scene selection items and dialect selection items; In response to a first selection operation of the scene selection item and the dialect selection item by the user through the display interface, providing the user with a target recognition model matching the selected target scene and target dialect; The target recognition model is used to recognize the speech to be recognized input by the user to obtain the corresponding recognition text.

2. The method according to claim 1, characterized in that The target recognition model is a first recognition model or a second recognition model in a preset recognition model library that matches the target scene and the target dialect; The second recognition model is obtained by adjusting the first recognition model using first audio data input by a user, and the first audio data is audio-related data about the target scene and the target dialect.

3. The method according to claim 2, characterized in that Several of the data options also include at least one processing option; Before providing the user with a target recognition model that matches the selected target scene and target dialect, the method further includes: Acquire first audio data input by a user; In response to a second selection operation of the user on at least one of the processing selection items, the first audio data is processed using a data processing method associated with the target processing item selected by the user to obtain processed audio data; The processed audio data is used as sample audio data, so as to adjust the first recognition model using the sample audio data to obtain a corresponding second recognition model.

4. The method according to claim 3, characterized in that At least one of the processing options includes at least one of a data enhancement item and a data mixing processing item; the data enhancement item includes a plurality of first selection sub-items corresponding to the data enhancement methods, and the data mixing processing item includes a plurality of second selection sub-items regarding the mixing ratio between the first audio data and the second audio data, and the second audio data is audio-related data of a common language corresponding to the target dialect.

5. The method according to claim 2, characterized in that: The first recognition model includes a first acoustic model and a first language model, and the first audio data includes at least one of a first audio, an annotated text corresponding to the first audio, and a sample corpus; The second recognition model includes a second acoustic model and a second language model, the second acoustic model is obtained by adjusting the first acoustic model using at least one of the first audio, the second audio, the annotated text corresponding to the first audio, and a sample corpus, and the second audio is obtained by performing speech synthesis on the sample corpus; The second language model is obtained by adjusting the first language model using the sample corpus, wherein the sample corpus is corpus related to the target scenario.

6. The method according to claim 2, characterized in that The first audio data includes at least one of a first audio, annotated text corresponding to the first audio, and a sample corpus; the first audio is audio including at least one audio format, and the sample corpus is corpus including at least one text format.

7. The method according to claim 2, characterized in that The second recognition model is obtained by adjusting the first recognition model multiple times using the first audio data; In the process of adjusting the first recognition model by using the first audio data input by the user to obtain the corresponding second recognition model, the method further includes: retaining a third recognition model obtained by adjusting the first recognition model in various ways; Selecting a model to be compared from each of the third recognition models in sequence according to a preset priority order, and performing effect tests on the model to be compared and the first recognition model respectively using the second audio data input by the user to obtain a first recognition result of the model to be compared and a second recognition result of the first recognition model; Based on the comparison result of the first recognition result and the second recognition result, it is determined whether to use the model to be compared as the second recognition model.

8. The method according to claim 7, characterized in that The preset priority order is negatively correlated with the order in which each adjustment is performed. The determining whether to use the to-be-compared model as the second recognition model based on the comparison result between the first recognition result and the second recognition result includes: In response to the first recognition result being better than the second recognition result and the difference between the first recognition result and the second recognition result meeting a preset difference requirement, it is determined to use the model to be compared as the second recognition model.

9. The method according to claim 2, characterized in that: The preset recognition model library includes a first model library for storing the first recognition model and a second model library for storing the second recognition model; The providing the user with a target recognition model that matches the selected target scene and target dialect includes: According to the priority order of the first model library and the second model library, the first recognition model or the second recognition model is selected as the target recognition model.

10. The method according to claim 9, characterized in that The priority of the second model library is higher than that of the first model library; The selecting the first recognition model or the second recognition model as the target recognition model according to the priority order of the first model library and the second model library comprises: Determining whether there is a second recognition model matching the target scene and the target dialect in the second model library; In response to the existence of the second recognition model, the second recognition model is selected as the target recognition model; otherwise, a first recognition model matching the target scene and the target dialect is selected from the first model library as the target recognition model.

11. The method according to claim 1, characterized in that: The target scenario includes a medical scenario, or the target scenario includes one of a telephone follow-up scenario, a doctor question and answer scenario, and a data entry scenario under the medical scenario; And / or, the display interface providing a plurality of data selection items and subsequent steps are executed by a speech recognition system.

12. A speech recognition device, characterized in that: The device comprises: An interface providing module, used to provide a display interface for a number of data selection items; the number of data selection items includes a scene selection item and a dialect selection item; A model providing module, configured to provide the user with a target recognition model matching the selected target scene and target dialect in response to a first selection operation of the user on the scene selection item and the dialect selection item through the display interface; The speech recognition module is used to use the target recognition model to recognize the speech to be recognized input by the user and obtain the corresponding recognition text.

13. An electronic device, characterized in that: comprising a memory and a processor coupled to each other, The memory stores program instructions; The processor is used to execute the program instructions stored in the memory to implement the method according to any one of claims 1 to 11.

14. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program instructions that can be run by a processor, and the program instructions can be executed by the processor to implement the method according to any one of claims 1 to 11.

Images