Speech processing method and apparatus, computer device, and storage medium

Abstract

The application relates to a voice processing method and device, computer equipment and a storage medium. The method comprises the following steps: in response to a received voice signal, displaying a first recognition result of processing the voice signal by a terminal; sending the voice signal to a server to instruct the server to process the voice signal again according to a preset voice recognition model and an association relationship between the accuracy of the voice recognition model and an intent category, and obtain a second recognition result; and generating a task control instruction according to the first recognition result and the second recognition result. The embodiment of the present application can guarantee the interactive experience of a user and improve the accuracy of a recognition result.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent vehicles, and in particular to a voice processing method and device, computer equipment, a storage medium and a computer program product. BACKGROUND

[0002] With the development of intelligent vehicles, voice interaction between users and vehicles becomes more convenient. However, in the related art, in response to a voice signal issued by a user, the response to the voice signal is not timely, or the understanding accuracy of the voice signal intent is too poor, so that the level of voice interaction in the related art is not high, and it is difficult to meet the use requirements of users. SUMMARY

[0003] Therefore, it is necessary to provide a voice processing method, device, computer equipment, storage medium and computer program product in view of the above technical problems.

[0004] In a first aspect, the present application provides a voice processing method. The method comprises:

[0005] In response to the received voice signal, a first recognition result of processing the voice signal by the terminal is displayed;

[0006] The voice signal is sent to a server to instruct the server to process the voice signal again according to a preset voice recognition model and an association relationship between the accuracy of the voice recognition model and the intent category, to obtain a second recognition result;

[0007] According to the first recognition result and the second recognition result, a task control instruction is generated.

[0008] In a possible implementation, the voice recognition model comprises a plurality of, and the server processes the voice signal again according to the preset voice recognition model and the association relationship between the accuracy of the voice recognition model and the intent category, to obtain a second recognition result, comprising:

[0009] The server inputs the voice signal into a plurality of preset voice recognition models respectively, to obtain a plurality of initial second recognition results corresponding thereto;

[0010] The plurality of initial second recognition results are input into an intent recognition model, and an intent result corresponding to each initial second recognition result is output;

[0011] In a case where a preset number of the intent results belong to a same intent category, the voice recognition model with the highest accuracy matched with the intent category is determined according to the association relationship between the accuracy of the voice recognition model and the intent category;

[0012] The second recognition result is determined as the initial second recognition result of the speech recognition model with the highest accuracy.

[0013] In a possible implementation, after the corresponding multiple initial second recognition results are obtained, the method further includes:

[0014] In a case where the preset number of the intent results do not belong to the same intent category, the second recognition result is determined as the initial second recognition result of the preset primary speech recognition model.

[0015] In a possible implementation, the generating of the task control instruction according to the first recognition result and the second recognition result includes:

[0016] The second recognition result is received, and a receiving time is determined.

[0017] In a case where a time delay between the sending time of the voice signal and the receiving time is less than or equal to a preset threshold, the task control instruction is generated according to the second recognition result.

[0018] In a possible implementation, after the second recognition result is received and the receiving time is determined, the method further includes:

[0019] In a case where a time delay between the sending time of the voice signal and the receiving time is less than or equal to a preset threshold, the first recognition result displayed is replaced by the second recognition result.

[0020] In a possible implementation, the server performs reprocessing on the voice signal according to a preset speech recognition model and an association relationship between an accuracy of a speech recognition model and an intent category, to obtain a second recognition result, including:

[0021] The server determines a target speech recognition model from the preset speech recognition model according to a preset speech recognition model and an association relationship between an accuracy of a speech recognition model and an intent category.

[0022] An intermediate recognition result of the target speech recognition model on the voice signal is obtained.

[0023] According to a preset mapping relationship between an original word and a target word, an original word in the intermediate recognition result is replaced by a corresponding target word, to obtain a second recognition result.

[0024] In a possible implementation, in response to the received voice signal, the first recognition result of the terminal on the voice signal is displayed, including:

[0025] In response to the received voice signal, the voice signal is identified and processed to obtain an initial first recognition result.

[0026] According to the user preference information corresponding to the terminal and the address book information, the initial first recognition result is corrected to obtain a first recognition result.

[0027] In a second aspect, the present application further provides a voice processing device. The device comprises:

[0028] A display module is configured to display a first recognition result of processing the voice signal by the terminal in response to the received voice signal.

[0029] A processing module is configured to send the voice signal to a server to instruct the server to process the voice signal again according to a preset voice recognition model and an association between the accuracy of the voice recognition model and the intent category to obtain a second recognition result.

[0030] A generation module is configured to generate a task control instruction according to the first recognition result and the second recognition result.

[0031] In a possible implementation, the voice recognition model comprises a plurality of, and the processing module comprises:

[0032] A first input sub-module is configured to input the voice signal to a plurality of preset voice recognition models respectively by the server to obtain a plurality of initial second recognition results corresponding to the voice signal.

[0033] A second input sub-module is configured to input the plurality of initial second recognition results to an intent recognition model to output an intent result corresponding to each of the initial second recognition results.

[0034] A first determination sub-module is configured to determine a voice recognition model with the highest accuracy that matches an intent category according to the association between the accuracy of the voice recognition model and the intent category when a preset number of the intent results belong to the same intent category.

[0035] A second determination sub-module is configured to determine the second recognition result as the initial second recognition result of the voice recognition model with the highest accuracy.

[0036] In a possible implementation, the processing module further comprises:

[0037] A third determination sub-module is configured to determine the second recognition result as the initial second recognition result of a preset primary voice recognition model when the preset number of the intent results do not belong to the same intent category.

[0038] In a possible implementation, the generation module comprises:

[0039] The receiving sub-module is configured to receive the second recognition result and determine a receiving time point;

[0040] The generating sub-module is configured to generate a task control instruction according to the second recognition result when a time delay between a sending time point of the voice signal and the receiving time point is less than or equal to a preset threshold.

[0041] In a possible implementation, the displaying module further includes:

[0042] The displaying sub-module is configured to replace the displayed first recognition result with the second recognition result when the time delay between the sending time point of the voice signal and the receiving time point is less than or equal to the preset threshold.

[0043] In a possible implementation, the processing module includes:

[0044] The fourth determining sub-module is configured to determine a target voice recognition model from the preset voice recognition models according to a preset voice recognition model and an association relationship between an accuracy of a voice recognition model and an intent category.

[0045] The obtaining sub-module is configured to obtain an intermediate recognition result of the target voice recognition model on the voice signal.

[0046] The first correcting sub-module is configured to replace an original word in the intermediate recognition result with a corresponding target word according to a preset mapping relationship between an original word and a target word, to obtain a second recognition result.

[0047] In a possible implementation, the displaying module includes:

[0048] The processing sub-module is configured to perform recognition processing on the received voice signal to obtain an initial first recognition result.

[0049] The second correcting sub-module is configured to correct the initial first recognition result according to user preference information and address book information corresponding to the terminal, to obtain a first recognition result.

[0050] In a third aspect, a computer device is provided. The computer device includes a memory and a processor. The memory stores a computer program. The processor implements the method in any of the embodiments of the present disclosure when executing the computer program.

[0051] In a fourth aspect, a computer readable storage medium is provided. The computer readable storage medium stores a computer program. The computer program is executed by a processor to implement the method in any of the embodiments of the present disclosure.

[0052] In a fifth aspect, the present application provides a computer program product. The computer program product comprises a computer program which, when executed by a processor, implements the method according to any of the embodiments of the present application.

[0053] In the voice processing method, the terminal displays the first recognition result. Since the first recognition result is processed by the terminal, the response speed of the display is high, and the user has a high interactive experience. Meanwhile, the voice recognition model in the server is used to recognize the voice signal again. The server contains multiple voice recognition models and the corresponding accuracy and intent category association relationship, can integrate the advantages of multiple voice recognition models, and give a second recognition result with higher accuracy. Therefore, the task control instruction generated by using the second recognition result is more accurate. BRIEF DESCRIPTION OF DRAWINGS

[0054] Figure 1 An application environment diagram of the voice processing method in an embodiment is shown in FIG. 1.

[0055] Figure 2 A first flowchart of the voice processing method in an embodiment is shown in FIG. 2.

[0056] Figure 3 A second flowchart of the voice processing method in another embodiment is shown in FIG. 3.

[0057] Figure 4 A third flowchart of the voice processing method in another embodiment is shown in FIG. 4.

[0058] Figure 5 A fourth flowchart of the voice processing method in another embodiment is shown in FIG. 5.

[0059] Figure 6 A fifth flowchart of the voice processing method in another embodiment is shown in FIG. 6.

[0060] Figure 7 A structural block diagram of the voice processing device in an embodiment is shown in FIG. 7.

[0061] Figure 8 An internal structure diagram of the computer device in an embodiment is shown in FIG. 8. DETAILED DESCRIPTION

[0062] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application is further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application.

[0063] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present disclosure described herein can be implemented in an order other than that illustrated or described herein. The implementation described in the following exemplary embodiments does not represent all implementations consistent with the present disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the present disclosure as detailed in the appended claims.

[0064] It should also be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for display, analyzed data, etc.) involved in the present disclosure are all information and data authorized by the user or authorized by all parties. The acquisition, storage, use, processing of data in the technical solutions of the present application comply with the relevant provisions of national laws and regulations.

[0065] The voice processing method provided by the embodiments of the present application can be applied to the application environment as shown in Figure 1 . Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data required to be processed by the server 104. The data storage system can be integrated on the server 104, or placed on the cloud or other network servers. Among them, the terminal 102 can be, but is not limited to, various personal computers, notebook computers, smart phones, tablet computers, Internet of Things devices and portable wearable devices. The Internet of Things device can be a smart speaker, a smart TV, a smart air conditioner, a smart vehicle device, etc. The portable wearable device can be a smart watch, a smart bracelet, a head-mounted device, etc. The server 104 can be implemented by an independent server or a server cluster composed of multiple servers.

[0066] In one embodiment, as shown in Figure 2 , a voice processing method is provided. Taking the terminal 102 in Figure 1 as an example, the method includes the following steps:

[0067] Step S201, in response to the received voice signal, displaying a first recognition result of processing the voice signal by the terminal.

[0068] Specifically, the received voice signal can include a voice signal received by a terminal, such as a mobile phone terminal, a smart sound terminal, a vehicle, and the like. In a vehicle, a voice signal of a user in the vehicle can be received by a microphone sensor. In an example embodiment, the user initiates a voice interaction function of the terminal by a preset wake-up word. In an example embodiment, after the terminal receives the voice signal, a front-end noise reduction module (ECNR, Echo Cancellation and Noise Reduction) can be used to obtain the position of the voice signal source, to perform noise and echo cancellation on the voice signal, and to enhance the human voice signal. Higher voice interaction quality can be obtained.

[0069] In the embodiments of the present disclosure, the terminal processes the voice signal. In an example embodiment, some general processing can be performed on the voice signal to obtain a first recognition result, such as converting to “You just said…”, or some expressions or symbols, such as a listening expression, to quickly respond and improve user experience. In another example embodiment, the terminal can also store some lightweight voice recognition models to process the voice signal to obtain a first recognition result. The first recognition result can include a relatively lower accuracy than a second recognition result of a cloud server, so that the recognition efficiency is high and the first recognition result can be quickly displayed.

[0070] In step S203, the voice signal is sent to a server to instruct the server to process the voice signal again according to a preset voice recognition model and an association relationship between the accuracy of the voice recognition model and the intent category, to obtain a second recognition result.

[0071] Specifically, the server can include a cloud server. The voice recognition model in the server can include a more complex voice recognition model with higher recognition accuracy. In an example embodiment, if the number of voice recognition models in the server includes one, the recognition accuracy of the voice recognition model is higher than that of the voice recognition model of the terminal. In another example embodiment, if the number of voice recognition models in the server includes multiple, an association relationship between the accuracy of the voice recognition model and the intent category can be established, such as for the intent category X, the accuracy of model A is 60%, the accuracy of model B is 70%, and the accuracy of model C is 80%; for example, 800 sub-intents are preset in the vehicle field, the recognition accuracy of model A for (0-300) is higher than that of other models, for example, 80%, the recognition accuracy of model B for (200-600) is higher than that of other models, for example, 95%, and the recognition accuracy of model C for (600-800) is higher than that of other models, for example, 90%. In this way, according to the intent category of the recognition result, it can be determined which model has higher accuracy, and the second recognition result of the voice signal processed by the model is selected.

[0072] In step S205, a task control instruction is generated according to the first recognition result and the second recognition result.

[0073] Specifically, the task control instruction can include converting the recognition result into a specific machine instruction, such as "car window", "play XX song", etc. In an exemplary embodiment, the corresponding task control instruction can be generated according to the second recognition result with higher accuracy. Since the second recognition result has higher accuracy, the generated task control instruction is also more accurate. In another exemplary embodiment, a preset time delay can be set. If the calculation time of the second recognition result is within the preset time delay, the corresponding task control instruction is generated according to the second recognition result with higher accuracy.

[0074] In the above voice processing method, the terminal displays the first recognition result. Since the first recognition result is processed by the terminal, the processing time period is short, the response speed is high, and the user has a high interactive experience. At the same time, the voice recognition model in the server is used to recognize the voice signal again. The server includes multiple voice recognition models and the corresponding accuracy and intent category association relationship, can integrate the advantages of multiple voice recognition models, and give the second recognition result with higher accuracy, so that the task control instruction generated by the second recognition result is also more accurate.

[0075] In a possible implementation, the voice recognition model includes multiple, and the server re-processes the voice signal according to the preset voice recognition model and the association relationship between the accuracy of the voice recognition model and the intent category, to obtain a second recognition result, including:

[0076] The server inputs the voice signal into the preset multiple voice recognition models to obtain a plurality of initial second recognition results corresponding to the voice signal;

[0077] The multiple initial second recognition results are input into an intent recognition model to output an intent result corresponding to each initial second recognition result;

[0078] In a case where a preset number of the intent results belong to the same intent category, an accuracy highest voice recognition model matched with the intent category is determined according to the association relationship between the accuracy of the voice recognition model and the intent category;

[0079] The second recognition result is determined as the initial second recognition result of the accuracy highest voice recognition model.

[0080] Specifically, in the embodiments of the present disclosure, the speech recognition model can include multiple, for example, different service providers can develop speech recognition models that can be shared in the cloud; for another example, different types of speech recognition models of the same service provider, or a combination of the two, etc. The server inputs the speech signal into a plurality of preset speech recognition models to obtain a plurality of corresponding initial second recognition results. In an exemplary embodiment, the plurality of initial second recognition results are input into an intent recognition model, and an intent result corresponding to each initial second recognition result is output. The intent result can include a classification result of the initial second recognition result, and the intent recognition model can be obtained by training a classification model based on an artificial neural network.

[0081] In the embodiments of the present disclosure, the preset number can include all the numbers of all speech recognition models, or can include a preset proportion of the total number, for example, 80%. The intent results of the preset number belong to the same intent category, for example, the initial second recognition results include: A, turn right; B, park on the right; C, turn left. All initial second recognition results belong to the navigation intent category, and the accuracy of the speech recognition model B of the navigation intent category is higher, so the second recognition result takes the initial second recognition result of the speech recognition model B: park on the right. For another example, the initial second recognition results include: A, play children's songs; B, play two brothers; C, play two songs. All initial second recognition results belong to the leisure and entertainment intent category, and the accuracy of the speech recognition model A of the leisure and entertainment category is higher, so the second recognition result takes the initial second recognition result of the speech recognition model A.

[0082] In an exemplary embodiment, the association between the accuracy of the speech recognition model and the intent category can also be updated. For example, when it is tested that the effect of a certain intent service A is worse than that of B, the intent will be reclassified to service B.

[0083] The above embodiments can select a speech recognition model with higher accuracy from multiple speech recognition models when the preset number of intent results belong to the same intent category, thereby improving the accuracy of speech recognition.

[0084] In a possible implementation, after the plurality of initial second recognition results are obtained, the method further includes:

[0085] When the preset number of intent results do not belong to the same intent category, the second recognition result is determined as the initial second recognition result of the preset primary speech recognition model.

[0086] Specifically, if the differences of the intent results of the plurality of speech recognition models are large, for example, the initial second recognition result corresponding to A is the leisure and entertainment intent class; the initial second recognition result corresponding to B is the navigation intent class; and the initial second recognition result corresponding to C is the Internet of Things intent class, the initial second recognition result of the preset primary speech recognition model can be used as the second recognition result, for example, the speech recognition model B is set as the primary speech recognition model, and then the second recognition result is determined as the navigation intent class. It should be noted that the recognition accuracy of the primary speech recognition model is higher than the accuracy of the terminal recognition. The primary speech recognition model can include a speech recognition model with high comprehensive performance.

[0087] In the above embodiment, in the case where the differences of the initial second recognition results are large, the initial second recognition result of the preset primary speech recognition model is used as the second recognition result, so that the accuracy of the recognition result can be improved to the maximum extent.

[0088] In a possible implementation, the generating the task control instruction according to the first recognition result and the second recognition result comprises:

[0089] receiving the second recognition result and determining a receiving time;

[0090] In the case where the time delay between the sending time of the voice signal and the receiving time is less than or equal to a preset threshold, the task control instruction is generated according to the second recognition result.

[0091] Specifically, the preset threshold can be determined according to a specific application scenario, for example, 1 minute or 1.5 minutes. In an exemplary embodiment, the receiving the second recognition result and determining the receiving time, and in the case where the time delay between the sending time of the voice signal and the receiving time is less than or equal to a preset threshold, the task control instruction is generated according to the second recognition result. In another exemplary embodiment, in the case where the time delay between the sending time of the voice signal and the receiving time is greater than the preset threshold, the task control instruction is generated according to the first recognition result.

[0092] In the embodiments of the present disclosure, considering that the network condition and the model calculation speed may need some time, for example, in the case that the network condition or the model calculation is time-consuming, a timeout mechanism is set, when the return time of the second recognition result is longer, the first recognition result is selected, and when the return time of the second recognition result is shorter, the second recognition result is selected, thereby ensuring the user experience. It should be noted that the timeout mechanism is not limited to the case that the time delay between the sending time and the receiving time of the voice signal is less than or equal to the preset threshold as described in the above embodiments, and the server can also monitor the calculation time delay of the model, and when the time delay is longer and greater than the preset threshold, the timeout reminder is sent to the terminal. Alternatively, the terminal does not receive the second recognition result within the preset threshold time, and the first recognition result can also be selected to generate the task control instruction. Under the inspiration of the technical essence of the present application, other changes can also be made by those skilled in the art, as long as the functions and effects achieved are the same or similar to the present application, which should be covered within the protection scope of the present application.

[0093] In a possible implementation, after the receiving the second recognition result and determining the receiving time, the method further includes:

[0094] In the case that the time delay between the sending time of the voice signal and the receiving time is less than or equal to the preset threshold, the first recognition result displayed is replaced by the second recognition result.

[0095] Specifically, in the case that the time delay between the sending time of the voice signal and the receiving time is less than or equal to the preset threshold, the first recognition result displayed is replaced by the second recognition result, and the second recognition result is displayed. For example, the first recognition result can be dynamically updated to replace a certain word or punctuation, and finally the second recognition result is stably displayed.

[0096] In the above embodiments, by replacing the first recognition result with the second recognition result and displaying the second recognition result, the user can know the accuracy of the second recognition result in time.

[0097] In a possible implementation, the server reprocesses the voice signal according to the preset voice recognition model and the association relationship between the accuracy of the voice recognition model and the intent category, to obtain the second recognition result, including:

[0098] The server determines a target voice recognition model from the preset voice recognition model according to the preset voice recognition model and the association relationship between the accuracy of the voice recognition model and the intent category;

[0099] Obtaining an intermediate recognition result of the target voice recognition model on the voice signal;

[0100] According to a preset mapping relationship between original words and target words, the original words in the intermediate recognition result are replaced by corresponding target words to obtain a second recognition result.

[0101] Specifically, the target speech recognition model can include a speech recognition model with higher accuracy described in the above embodiments, or a main speech recognition model described in the above embodiments. The intermediate recognition result of the target speech recognition model for the speech signal is obtained, and according to a preset mapping relationship between original words and target words, the original words in the intermediate recognition result are replaced by corresponding target words, for example, new radio is replaced by XXX radio; to new power is replaced by to XXX power. Wherein, the target words can include naming words associated with components of the terminal. In an exemplary embodiment, a preset filter word can also be set, for example, when "XX" appears in the intermediate recognition result, the word is filtered.

[0102] The above embodiments, by establishing the mapping relationship between the original words and the target words, replacing the original words in the intermediate recognition result with the corresponding target words, help to improve the accuracy of the recognition result of the specific terminal.

[0103] In a possible implementation, in response to the received speech signal, a first recognition result of the terminal processing the speech signal is displayed, including:

[0104] In response to the received speech signal, the speech signal is recognized and processed to obtain an initial first recognition result.

[0105] According to the user preference information and the address book information corresponding to the terminal, the initial first recognition result is corrected to obtain the first recognition result.

[0106] Specifically, the user preference information can include user preference data obtained by analyzing historical use data of the user. The address book information can include the association relationship between the user and other users, such as telephone address book, instant chat address book, etc. In the embodiments of the present disclosure, according to the terminal processing method described in the above embodiments, in response to the received speech signal, the speech signal is recognized and processed to obtain an initial first recognition result.

[0107] In an example embodiment, the terminal can store the preference information and the address book information of the user corresponding to the terminal, and correct the initial first recognition result according to the user preference information and the address book information of the terminal. For example, the address book of the user includes "Chang'an", and the initial first recognition result includes: calling Chang'an. The initial first recognition result is corrected according to the address book information, including replacing "Chang'an" in the initial first recognition result with "Chang'an". For another example, the preference information of the user includes listening to Cantonese version of song A, and the initial first recognition result is English version of song A. The initial first recognition result is corrected according to the preference information, including replacing the Cantonese version of song A in the initial first recognition result with the English version of song A. In another example embodiment, the preference information and the address book information of the user can also be uploaded to the server under the authorization of the user, and the server corrects the second recognition result according to the preference information and the address book information.

[0108] The above embodiment can obtain more accurate first recognition result by correcting the initial first recognition result according to the user preference information and the address book information.

[0109] Figure 3 For a second flowchart of the voice processing method in another example embodiment, referring to FIG. 3, the method includes: Figure 3

[0110] In step S201, a first recognition result of processing the voice signal by the terminal is displayed in response to the received voice signal.

[0111] In the embodiment of the present disclosure, the terminal processes the voice signal, which in an example embodiment can include some general processing of the voice signal to obtain a first recognition result, such as converting into "You just said…", or some expressions or symbols, such as a listening expression, to quickly respond and improve user experience. In another example embodiment, the terminal can also store some lightweight voice recognition models to process the voice signal to obtain a first recognition result, which can include a relatively low accuracy compared to the second recognition result of the cloud server, thereby improving the recognition efficiency and displaying the first recognition result.

[0112] In step S301, the server inputs the voice signal into a plurality of preset voice recognition models respectively to obtain a plurality of initial second recognition results corresponding to the voice signal. The plurality of initial second recognition results are input into an intent recognition model to output an intent result corresponding to each of the initial second recognition results.

[0113] ​In a case where the preset number of the intent results belong to the same intent category, a speech recognition model with the highest accuracy that matches the intent category is determined according to an association between the accuracy of the speech recognition model and the intent category.

[0114] In step S305, the second recognition result is determined as the initial second recognition result of the speech recognition model with the highest accuracy.

[0115] Specifically, in the embodiments of the present disclosure, the speech recognition model can include multiple, for example, different service providers' speech recognition models that can be shared in the cloud; for another example, different types of speech recognition models of the same service provider, or a combination of the two, etc. The server inputs the speech signal into a plurality of preset speech recognition models to obtain a plurality of initial second recognition results corresponding thereto.

[0116] In the embodiments of the present disclosure, the preset number can include all the numbers of all speech recognition models, or can include a preset proportion of the total number, for example, 80%. In a case where the preset number of the intent results belong to the same intent category, a speech recognition model with higher accuracy can be selected from the plurality of speech recognition models, thereby improving the accuracy of speech recognition.

[0117] In step S307, in a case where the preset number of the intent results do not belong to the same intent category, the second recognition result is determined as the initial second recognition result of a preset primary speech recognition model.

[0118] Specifically, if the intent results of the plurality of speech recognition models are significantly different, for example, the initial second recognition result corresponding to A is a leisure and entertainment intent category; the initial second recognition result corresponding to B is a navigation intent category; and the initial second recognition result corresponding to C is an Internet of Things intent category, the initial second recognition result of the preset primary speech recognition model can be used as a reference, for example, speech recognition model B is set as the primary speech recognition model, and then the second recognition result is determined as the navigation intent category. It should be noted that the recognition accuracy of the primary speech recognition model is higher than the accuracy of terminal recognition. The primary speech recognition model can include a speech recognition model with higher comprehensive performance.

[0119] In step S309, the second recognition result is received and a receiving time is determined.

[0120] In step S311, in a case where a time delay between the sending time of the speech signal and the receiving time is less than or equal to a preset threshold, a task control instruction is generated according to the second recognition result.

[0121] Specifically, the preset threshold can be determined according to a specific application scenario, for example, 1 minute, 1.5 minutes. In an exemplary embodiment, the second recognition result is received, and a time delay between a sending time of the voice signal and a receiving time is determined. In a case where the time delay is less than or equal to the preset threshold, a task control instruction is generated according to the second recognition result. In another exemplary embodiment, in a case where the time delay is greater than the preset threshold, a task control instruction is generated according to the first recognition result.

[0122] In the embodiments of the present disclosure, considering that network conditions and model calculation speed may require some time, for example, for a case where network conditions or model calculation is time-consuming, a timeout mechanism is set. When the return time of the second recognition result is longer, the first recognition result is selected; when the return time of the second recognition result is shorter, the second recognition result is selected, thereby ensuring the user experience.

[0123] Figure 4 to Figure 6 For a flowchart of a voice processing method in another embodiment, as shown in Figure 4 The terminal can include a car machine system. The car machine system performs end-side writing on the received voice signal, and obtains a first recognition result in combination with user preferences (information), and displays the first recognition result on the screen in real time for the user to view. The car machine system sends the received voice signal to a cloud server, and simultaneously sends the user preference information to the cloud server in a case where vehicle information authorization is performed. As shown in Figure 5 The cloud server identifies the voice signal by using multiple voice recognition models to obtain multiple initial second recognition results, inputs the multiple initial second recognition results into an intent recognition model, and outputs an intent result corresponding to each initial second recognition result, for example, (text A, intent A), (text B, intent B), and the like. The cloud server selects a second recognition result of a voice recognition model with higher accuracy according to a category of the recognized intent result. The text in the second recognition result is sent to the terminal, and downstream task processing is performed. As shown in Figure 6 A timeout mechanism is established to calculate whether the second recognition result returned by the cloud is timed out. If the second recognition result is timed out, a downstream task corresponding to the first recognition result is selected. If the second recognition result is not timed out, a downstream task corresponding to the second recognition result is selected.

[0124] It should be understood that although the steps in the flowcharts involved in the embodiments described above are shown in sequence according to the arrows, the steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, the execution of the steps is not strictly limited in sequence, and the steps can be executed in other orders. Moreover, at least some of the steps in the flowcharts involved in the embodiments described above can include multiple steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution order of the steps or stages is not necessarily sequential, but can be alternately executed with other steps or steps or stages in at least part of other steps.

[0125] Based on the same inventive concept, the embodiments of the present application also provide a voice processing device for implementing the voice processing method described above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme described in the above method, so the specific limitations in one or more voice processing device embodiments provided below can refer to the limitations of the voice processing method described above, which will not be repeated here.

[0126] In one embodiment, as shown in Figure 7 A voice processing device is provided, and the device 700 includes:

[0127] The display module 701 is configured to display a first recognition result of processing the voice signal by the terminal in response to the received voice signal.

[0128] The processing module 703 is configured to send the voice signal to a server to instruct the server to process the voice signal again according to a preset voice recognition model and an association relationship between a voice recognition model accuracy and an intent category, to obtain a second recognition result.

[0129] The generation module 705 is configured to generate a task control instruction according to the first recognition result and the second recognition result.

[0130] In a possible implementation, the voice recognition model includes multiple, and the processing module includes:

[0131] The first input sub-module is configured to input the voice signal to the preset multiple voice recognition models by the server, to obtain corresponding multiple initial second recognition results.

[0132] The second input sub-module is configured to input the multiple initial second recognition results to an intent recognition model, and output an intent result corresponding to each of the initial second recognition results.

[0133] The first determining sub-module is configured to, when a preset number of the intent results belong to a same intent category, determine, according to an association between the accuracy of the speech recognition model and the intent category, a speech recognition model with the highest accuracy that matches the intent category.

[0134] The second determining sub-module is configured to determine that the second recognition result is an initial second recognition result of the speech recognition model with the highest accuracy.

[0135] In a possible implementation, the processing module further includes:

[0136] The third determining sub-module is configured to, when the preset number of the intent results do not belong to the same intent category, determine that the second recognition result is an initial second recognition result of a preset primary speech recognition model.

[0137] In a possible implementation, the generating module includes:

[0138] The receiving sub-module is configured to receive the second recognition result and determine a receiving time.

[0139] The generating sub-module is configured to, when a time delay between a sending time of the voice signal and the receiving time is less than or equal to a preset threshold, generate a task control instruction according to the second recognition result.

[0140] In a possible implementation, the processing module further includes:

[0141] The displaying sub-module is configured to, when the time delay between the sending time of the voice signal and the receiving time is less than or equal to the preset threshold, replace the first recognition result displayed with the second recognition result.

[0142] In a possible implementation, the processing module includes:

[0143] The fourth determining sub-module is configured to determine, according to a preset speech recognition model and an association between the accuracy of the speech recognition model and the intent category, a target speech recognition model from the preset speech recognition model.

[0144] The obtaining sub-module is configured to obtain an intermediate recognition result of the target speech recognition model on the voice signal.

[0145] The first correcting sub-module is configured to replace original words in the intermediate recognition result with corresponding target words according to a preset mapping relationship between the original words and the target words, to obtain a second recognition result.

[0146] In a possible implementation, the processing module includes:

[0147] The processing submodule is configured to perform identification processing on the received voice signal to obtain an initial first identification result.

[0148] The second correction submodule is configured to correct the initial first identification result according to the user preference information and the address book information corresponding to the terminal to obtain a first identification result.

[0149] The modules in the voice processing apparatus can be implemented by software, hardware, or a combination thereof. The modules can be embedded in or independent of a processor in a computer device in hardware form, or stored in a memory in a computer device in software form, so as to be called and executed by a processor to perform operations corresponding to the modules.

[0150] In one embodiment, a computer device is provided, which can be a terminal. An internal structure diagram of the computer device can be as shown in FIG. 1. Figure 8 The computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is configured to perform wired or wireless communication with an external terminal. The wireless communication can be achieved through WIFI, mobile cellular network, NFC (near field communication), or other technologies. The computer program is executed by the processor to implement a voice processing method. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball, or touchpad arranged on the shell of the computer device. The input device can also be an external keyboard, touchpad, or mouse, etc.

[0151] Those skilled in the art can understand that the structure shown in FIG. 1 is only a block diagram of part of the structure related to the scheme of the present application, and does not limit the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different component arrangement. Figure 8

[0152] ​Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when the computer program is executed, the processes of the above-mentioned embodiments of the methods can be included. Any reference to memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (Read-Only Memory, ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive memory (ReRAM), magnetoresistive random access memory (Magnetoresistive Random Access Memory, MRAM), ferroelectric memory (Ferroelectric Random Access Memory, FRAM), phase change memory (Phase Change Memory, PCM), graphene memory, etc. Volatile memory can include random access memory (Random Access Memory, RAM) or external cache memory, etc. As an illustration but not limitation, RAM can be in various forms, such as static random access memory (Static Random Access Memory, SRAM) or dynamic random access memory (Dynamic Random Access Memory, DRAM), etc. The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., without being limited thereto. The processor involved in the embodiments provided in the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., without being limited thereto.

[0153] Any combination of the technical features of the above embodiments can be made. In order to make the description simple, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combination of the technical features does not exist contradictory, it should be considered as the scope of the present application.

[0154] The above embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the patent of the present application. It should be pointed out that for ordinary skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are within the scope of protection of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. A speech processing method, characterized in that, Applied to terminals, including: In response to the received voice signal, the terminal displays a first recognition result of processing the voice signal; The voice signal is sent to a server to instruct the server to process the voice signal again according to a preset voice recognition model and the correlation between the accuracy of the voice recognition model and the intent category, to obtain a second recognition result. The process of processing the voice signal again according to the preset voice recognition model and the correlation between the accuracy of the voice recognition model and the intent category to obtain a second recognition result includes: the server inputting the voice signal into multiple preset voice recognition models to obtain multiple initial second recognition results; inputting the multiple initial second recognition results into an intent recognition model and outputting the intent result corresponding to each initial second recognition result; when a preset number of intent results belong to the same intent category, determining the voice recognition model with the highest accuracy that matches the intent category according to the correlation between the accuracy of the voice recognition model and the intent category; and determining the second recognition result as the initial second recognition result of the voice recognition model with the highest accuracy. Based on the first identification result and the second identification result, a task control instruction is generated; the generation of the task control instruction based on the first identification result and the second identification result includes: if the calculation time of the second identification result is within a preset delay, then a corresponding task control instruction is generated based on the second identification result with higher accuracy; wherein, the preset delay is obtained by pre-setting.

2. The method according to claim 1, characterized in that, After obtaining the corresponding multiple initial second identification results, the process further includes: If the preset number of intent results do not belong to the same intent category, the second recognition result is determined as the initial second recognition result of the preset main speech recognition model.

3. The method according to claim 1, characterized in that, The step of generating task control instructions based on the first identification result and the second identification result includes: Receive the second identification result and determine the receiving time; If the time delay between the transmission time and the reception time of the voice signal is less than or equal to a preset threshold, a task control command is generated based on the second recognition result.

4. The method according to claim 3, characterized in that, After receiving the second identification result and determining the reception time, the method further includes: If the time delay between the transmission time and the reception time of the voice signal is less than or equal to a preset threshold, the first recognition result displayed will be replaced with the second recognition result.

5. The method according to claim 1, characterized in that, The server processes the speech signal again based on a preset speech recognition model and the correlation between the accuracy of the speech recognition model and the intent category to obtain a second recognition result, including: The server determines the target speech recognition model from the preset speech recognition models based on the preset speech recognition models and the correlation between the accuracy of the speech recognition models and the intent category. Obtain the intermediate recognition results of the target speech recognition model for the speech signal; Based on the preset mapping relationship between original words and target words, the original words in the intermediate recognition results are replaced with the corresponding target words to obtain the second recognition result.

6. The method according to claim 1, characterized in that, In response to a received voice signal, the terminal displays a first recognition result of processing the voice signal, including: In response to the received voice signal, the voice signal is processed for recognition to obtain an initial first recognition result; Based on the user preference information and contact information corresponding to the terminal, the initial first identification result is corrected to obtain the first identification result.

7. A voice processing device, characterized in that, The device is applied to a terminal and includes: The display module is used to display the first recognition result of the terminal processing the voice signal in response to the received voice signal; A processing module is used to send the voice signal to a server, instructing the server to process the voice signal again according to a preset voice recognition model and the correlation between the accuracy of the voice recognition model and the intent category, to obtain a second recognition result. The processing module includes: a first input submodule, used by the server to input the voice signal into multiple preset voice recognition models respectively to obtain multiple initial second recognition results; a second input submodule, used to input the multiple initial second recognition results into an intent recognition model and output the intent result corresponding to each initial second recognition result; a first determination submodule, used to determine the voice recognition model with the highest accuracy that matches the intent category, based on the correlation between the accuracy of the voice recognition model and the intent category, when a preset number of intent results belong to the same intent category; and a second determination submodule, used to determine that the second recognition result is the initial second recognition result of the voice recognition model with the highest accuracy. The generation module is used to generate task control instructions based on the first recognition result and the second recognition result; the generation of task control instructions based on the first recognition result and the second recognition result includes: if the calculation time of the second recognition result is within a preset delay, then the corresponding task control instruction is generated based on the second recognition result with higher accuracy; wherein, the preset delay is obtained by pre-setting.

8. The apparatus according to claim 7, characterized in that, The processing module further includes: The third determining submodule is used to determine the second recognition result as the initial second recognition result of the preset main speech recognition model when the preset number of the initial second recognition results do not belong to the same intent category.

9. The apparatus according to claim 7, characterized in that, The generation module includes: A receiving submodule is used to receive the second identification result and determine the receiving time; The generation submodule is used to generate task control instructions based on the second recognition result when the time delay between the transmission time and the reception time of the voice signal is less than or equal to a preset threshold.

10. The apparatus according to claim 9, characterized in that, The display module also includes: The display submodule is used to replace the first recognition result with the second recognition result when the time delay between the transmission time and the reception time of the voice signal is less than or equal to a preset threshold.

11. The apparatus according to claim 7, characterized in that, The processing module includes: The fourth determination submodule is used by the server to determine the target speech recognition model from the preset speech recognition models based on the preset speech recognition models and the correlation between the accuracy of the speech recognition models and the intent category. The acquisition submodule is used to acquire the intermediate recognition results of the target speech recognition model for the speech signal; The first correction submodule is used to replace the original words in the intermediate recognition result with the corresponding target words according to the preset mapping relationship between the original words and the target words, so as to obtain the second recognition result.

12. The apparatus according to claim 7, characterized in that, The display module includes: The processing submodule is used to perform recognition processing on the received voice signal in response to obtain an initial first recognition result; The second correction submodule is used to correct the initial first identification result based on the user preference information and address book information corresponding to the terminal, so as to obtain the first identification result.

13. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

14. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

15. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images