Speech recognition method, device, equipment, system and program product

Abstract

The embodiment of the invention provides a voice recognition method, and the method comprises the steps: carrying out the voice recognition of first voice information, so as to determine a target user and a voice scene corresponding to the first voice information; then configuring a blacklist and a whitelist corresponding to the target user based on the voice scene, wherein the blacklist and the whitelist are used for adjusting a voice end value; in response to the triggering of the second voice information, updating the list resources according to the black and white list; and determining a voice end value corresponding to the second voice information through the updated list resource, so as to recognize the second voice information based on the voice end value. Therefore, the personalized voice end value configuration process is realized, the black list and the white list are configured in combination with the voice features of the user, and the list resources are maintained through the black list and the white list, so that the voice end value can be adapted to different users, the content integrity after the voice input of the user is improved, and the voice recognition accuracy is improved.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to speech recognition methods, devices, equipment, systems and program products. Background Art

[0002] As smart devices become increasingly abundant, controlling devices through voice has become a common method. For example, voice activity detection technology can be used to identify voice activity segments and non-voice activity segments in audio signals for voice recognition.

[0003] Generally, voice activity detection technology divides voice segments by judging whether the silence at the end of the effective voice reaches the corresponding duration of the voice end value. In phrase flow interaction, such as car machine voice recognition, the EOS duration is usually fixed, which leads to early truncation and long waiting problems, which can easily cause the user to lose content during the voice input process and affect the accuracy of voice recognition. Summary of the invention

[0004] In view of this, the present application provides a speech recognition method to improve the accuracy of speech recognition.

[0005] According to a first aspect of an embodiment of the present application, a method for speech recognition is provided, which can be applied to a system or program including a speech recognition function in a terminal device, and specifically includes:

[0006] Performing voice recognition on the first voice information to determine a target user and a voice scene corresponding to the first voice information;

[0007] Based on the voice scenario, a blacklist and a whitelist corresponding to the target user are configured, wherein the blacklist and the whitelist are used to adjust the voice end value;

[0008] In response to the triggering of the second voice information, the list resource is updated according to the black and white list, the second voice information being the voice information in the subsequent conversation of the first voice information;

[0009] The voice end value corresponding to the second voice information is determined through the updated list resource, so as to identify the second voice information based on the voice end value.

[0010] Optionally, in some possible implementations, performing voice recognition on the first voice information to determine a target user and a voice scenario corresponding to the first voice information includes:

[0011] Get voiceprint registration information;

[0012] Comparing the first voice information with the voiceprint registration information to determine the target user;

[0013] Performing voice recognition on the first voice information to determine a voice recognition result associated with the target user;

[0014] The speech recognition result is semantically determined to determine the speech scene corresponding to the first speech information.

[0015] Optionally, in some possible implementations, the voice scene is a truncated scene, and the performing semantic discrimination on the voice recognition result to determine the voice scene corresponding to the first voice information includes:

[0016] Acquire first content and second content whose contents are adjacent to each other in the speech recognition result;

[0017] Determining integrity information corresponding to the first content and the second content, wherein the integrity information is set based on content integrity and semantic integrity;

[0018] If the completeness information indicates that the first content and the second content are complete, and the semantically associated content of the first content and the second content is complete, it is determined that the voice scene corresponding to the first voice information is the truncated scene.

[0019] Optionally, in some possible implementations, the voice scene is a long-wait scene, and the performing semantic discrimination on the voice recognition result to determine the voice scene corresponding to the first voice information includes:

[0020] Determining execution information in the speech recognition result for triggering execution of an action;

[0021] If the semantics indicated by the execution information are complete and the action corresponding to the execution information has not been executed, it is determined that the voice scene corresponding to the first voice information is the long waiting scene.

[0022] Optionally, in some possible implementations, if the semantics indicated by the execution information are complete and the execution information is used to indicate a target action, determining that the voice scene corresponding to the first voice information is the long waiting scene includes:

[0023] If the semantics indicated by the execution information are complete and the execution information is used to indicate the target action, then a prompt content indicating the response time is generated;

[0024] In response to a confirmation operation of the user on the prompt content, it is determined that the voice scene corresponding to the first voice information is the long-wait scene.

[0025] Optionally, in some possible implementations, configuring a blacklist and whitelist corresponding to the target user based on the voice scenario includes:

[0026] If the voice scene is a truncated scene, obtaining semantically incomplete voice recognition content in the truncated scene, and configuring the semantically incomplete voice recognition content to a blacklist in the blacklist and whitelist corresponding to the target user;

[0027] If the voice scene is a long-wait scene, semantically complete voice recognition content in the long-wait scene is obtained, and the semantically complete voice recognition content is configured into a whitelist in the blacklist and whitelist corresponding to the target user.

[0028] Optionally, in some possible implementations, in response to the triggering of the second voice information, updating the list resource according to the black and white list includes:

[0029] In response to the triggering of the second voice information, the blacklist and whitelist are sent to the cloud, the second voice information is subjected to voiceprint recognition in the cloud to determine the target user, and the list resource corresponding to the target user is loaded;

[0030] The list resource corresponding to the target user is updated according to the black and white list.

[0031] Optionally, in some possible implementations, determining the voice end value corresponding to the second voice information through the updated list resource to identify the second voice information based on the voice end value includes:

[0032] Matching the voice content corresponding to the second voice information with the updated list resource to obtain list hit information;

[0033] Determining the hit scenario indicated by the list hit information;

[0034] The voice end value corresponding to the second voice information is determined based on the hit scenario, so as to identify the second voice information based on the voice end value.

[0035] Optionally, in some possible implementations, the method further includes:

[0036] If the list hit information indicates that there is no hit scenario matching the second voice information in the list resource, determining a user group associated with a target user corresponding to the second voice information;

[0037] Obtaining a semantic recognition model corresponding to the user group;

[0038] The voice end value corresponding to the second voice information is determined based on the list resource corresponding to the user group, so as to identify the second voice information based on the voice end value.

[0039] Optionally, in some possible implementations, the training process of the semantic recognition model corresponding to the user group includes:

[0040] Get training corpus;

[0041] Classifying the training corpus according to preset dimensions based on user group types to obtain training data corresponding to each user group, wherein the preset dimensions are set based on a language type dimension and an object description dimension;

[0042] The semantic recognition model corresponding to each user group is trained according to the training data.

[0043] According to a second aspect of an embodiment of the present application, there is provided a speech recognition device, including:

[0044] a recognition unit, configured to perform voice recognition on the first voice information to determine a target user and a voice scene corresponding to the first voice information;

[0045] A configuration unit, configured to configure a blacklist and a whitelist corresponding to the target user based on the voice scenario, wherein the blacklist and the whitelist are used to adjust the voice end value;

[0046] The configuration unit is further configured to update the list resource according to the black and white list in response to the triggering of the second voice information, wherein the second voice information is the voice information in the subsequent conversation of the first voice information;

[0047] The recognition unit is further used to determine the voice end value corresponding to the second voice information through the updated list resource, so as to recognize the second voice information based on the voice end value.

[0048] Optionally, in some possible implementations, the recognition unit is used to obtain voiceprint registration information when performing voice recognition on the first voice information to determine the target user and voice scene corresponding to the first voice information; compare the first voice information with the voiceprint registration information to determine the target user; perform voice recognition on the first voice information to determine the voice recognition result associated with the target user; and perform semantic judgment on the voice recognition result to determine the voice scene corresponding to the first voice information.

[0049] Optionally, in some possible implementations, the recognition unit is used to obtain first content and second content whose contents are adjacent in the speech recognition result when the speech scene is a truncated scene and semantic discrimination is performed on the speech recognition result to determine the speech scene corresponding to the first voice information; determine completeness information corresponding to the first content and the second content, wherein the completeness information is set based on content integrity and semantic integrity; if the completeness information indicates that the contents of the first content and the second content are complete, and the contents of the first content and the second content after semantic association are complete, then determine that the speech scene corresponding to the first voice information is the truncated scene.

[0050] Optionally, in some possible implementations, the recognition unit is used to determine the execution information in the voice recognition result for triggering the execution of an action when the voice scene is a long-waiting scene and semantic discrimination is performed on the voice recognition result to determine the voice scene corresponding to the first voice information; if the semantics indicated by the execution information are complete and the action corresponding to the execution information is not executed, it is determined that the voice scene corresponding to the first voice information is the long-waiting scene.

[0051] Optionally, in some possible implementations, the recognition unit is used to generate prompt content indicating the response time when, if the semantics indicated by the execution information are complete and the execution information is used to indicate the target action, the voice scene corresponding to the first voice information is the long waiting scene; and in response to the user's confirmation operation on the prompt content, determine that the voice scene corresponding to the first voice information is the long waiting scene.

[0052] Optionally, in some possible implementations, the configuration unit is used to, when configuring the blacklist and whitelist corresponding to the target user based on the voice scene, obtain semantically incomplete speech recognition content in the truncated scene if the voice scene is a truncated scene, and configure the semantically incomplete speech recognition content to the blacklist in the blacklist and whitelist corresponding to the target user; if the voice scene is a long-wait scene, obtain semantically complete speech recognition content in the long-wait scene, and configure the semantically complete speech recognition content to the whitelist in the blacklist and whitelist corresponding to the target user.

[0053] Optionally, in some possible implementations, the configuration unit is used to send the black and white list to the cloud in response to the triggering of the second voice information, perform voiceprint recognition on the second voice information in the cloud to determine the target user, and load the list resources corresponding to the target user; when the list resources are updated according to the black and white list in response to the triggering of the second voice information, the list resources corresponding to the target user are updated according to the black and white list.

[0054] Optionally, in some possible implementations, the recognition unit is used to match the voice content corresponding to the second voice information with the updated list resources to obtain list hit information when determining the voice end value corresponding to the second voice information through the updated list resources to identify the second voice information based on the voice end value; determine the hit scenario indicated by the list hit information; and determine the voice end value corresponding to the second voice information based on the hit scenario to identify the second voice information based on the voice end value.

[0055] Optionally, in some possible implementations, the recognition unit is also used to determine the user group associated with the target user corresponding to the second voice information if the list hit information indicates that there is no hit scene in the list resource that matches the second voice information; obtain the semantic recognition model corresponding to the user group; and determine the voice end value corresponding to the second voice information based on the list resource corresponding to the user group, so as to recognize the second voice information based on the voice end value.

[0056] Optionally, in some possible implementations, the recognition unit is also used to obtain training corpus; classify the training corpus according to preset dimensions based on user group type to obtain training data corresponding to each user group, wherein the preset dimensions are set based on language type dimensions and object description dimensions; and train the semantic recognition model corresponding to each user group according to the training data.

[0057] According to a third aspect of an embodiment of the present application, there is provided a speech recognition device, including an input and output component and a processor;

[0058] The input-output component is used to obtain voice information;

[0059] The processor is used to perform voice content recognition on the voice information acquired by the input-output component by executing the voice recognition method as described in the first aspect or any implementation of the first aspect.

[0060] According to a fourth aspect of an embodiment of the present application, there is provided a speech recognition system, including an interactive client and a server;

[0061] The interactive client is used to obtain voice information and send the voice information to the server, and to display the voice recognition results output by the server;

[0062] The server is used to perform voice content recognition on the voice information obtained by the interactive client by executing the voice recognition method described in the first aspect or any implementation method of the first aspect.

[0063] According to a fifth aspect of an embodiment of the present application, a computer program product is provided, comprising: a computer program, which, when executed by a processor, implements the speech recognition method described in the first aspect or any implementation manner of the first aspect.

[0064] It can be seen from the above technical solutions that the embodiments of the present application have the following advantages:

[0065] By performing voice recognition on the first voice information, the target user and voice scenario corresponding to the first voice information are determined; then the black and white lists corresponding to the target users are configured based on the voice scenario, and the black and white lists are used to adjust the voice end value; and in response to the triggering of the second voice information, the list resources are updated according to the black and white lists, and the second voice information is the voice information in the subsequent conversation of the first voice information; and then the voice end value corresponding to the second voice information is determined through the updated list resources, so as to identify the second voice information based on the voice end value. Thus, a personalized voice end value configuration process is realized. Since the black and white lists are configured in combination with the voice characteristics of the user, and the list resources are maintained through the black and white lists, the voice end value can be adapted to different users, thereby improving the integrity of the content after the user's voice input and improving the accuracy of voice recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0067] Figure 1 Diagram of the network architecture running for the speech recognition system;

[0068] Figure 2 A flow chart of speech recognition provided in an embodiment of the present application;

[0069] Figure 3 A flow chart of a speech recognition method provided in an embodiment of the present application;

[0070] Figure 4 A schematic diagram of a blacklist and whitelist configuration process provided in an embodiment of the present application;

[0071] Figure 5 A schematic diagram of another process for taking effect of a list resource provided in an embodiment of the present application;

[0072] Figure 6 A schematic diagram of another configuration of a speech recognition model provided in an embodiment of the present application;

[0073] Figure 7 A schematic diagram of the structure of a speech recognition device provided in an embodiment of the present application;

[0074] Figure 8 A schematic diagram of the structure of a terminal device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0075] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0076] It should be understood that the speech recognition method provided in the present application can be applied to a system or program including a speech recognition function in a terminal device, such as an intelligent control application. Specifically, the speech recognition system can be run on Figure 1 In the network architecture shown in Figure 1 As shown in FIG. 1 , it is a network architecture diagram of the operation of the speech recognition system. As can be seen from the figure, the speech recognition system can provide a speech recognition process with multiple information sources, that is, the speech information is determined by the acquisition operation on the terminal side, and then sent to the server to determine the corresponding speech end value in the speech information, and then speech recognition is performed; it can be understood that Figure 1 A variety of terminal devices are shown in FIG. 1 . The terminal device may be a computer device. In actual scenarios, more or fewer types of terminal devices may participate in the speech recognition process. The specific number and type depend on the actual scenario and are not limited here. In addition, Figure 1 One server is shown in the figure, but in actual scenarios, multiple servers may be involved, especially in multi-disciplinary scenarios, and the specific number of servers depends on the actual scenario.

[0077] In this embodiment, the server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The terminal can be a smart phone, tablet computer, laptop computer, desktop computer, smart speaker, smart watch, etc., but is not limited to this. The terminal and the server can be directly or indirectly connected through wired or wireless communication, and the terminal and the server can be connected to form a blockchain network, which is not limited in this application.

[0078] It can be understood that the above-mentioned voice recognition system can run on a personal mobile terminal, for example: as an application such as a smart home, it can also run on a server, and it can also run on a third-party device to provide voice recognition to obtain the voice recognition processing results of the information source; the specific voice recognition system can be run in the above-mentioned device in the form of a program, or it can be run as a system component in the above-mentioned device, and it can also be used as a cloud service program. The specific operation mode depends on the actual scenario and is not limited here.

[0079] As smart devices become increasingly abundant, controlling devices through voice has become a common method. For example, voice activity detection technology can be used to identify voice activity segments and non-voice activity segments in audio signals for voice recognition.

[0080] Generally, in VAD, the speech segments are divided by judging whether the silence at the end of the effective speech reaches the End of Speech (EOS) duration. In phrase stream interaction, such as in car-mounted speech recognition, the EOS duration is usually fixed, which leads to early truncation and long waiting time, which can easily cause the user to lose content during the speech input process and affect the accuracy of speech recognition.

[0081] In order to solve the above problems, this application proposes a speech recognition method, which is applied to Figure 2 In the process framework of speech recognition shown in Figure 2 As shown, a process architecture diagram of speech recognition provided in an embodiment of the present application is provided, which obtains speech information through interactive operations of the terminal, and determines the speech end value according to the black and white list resources of the server, and then performs the speech recognition process.

[0082] It can be understood that the speech recognition method provided in the present application can be written as a program to serve as a processing logic in a hardware system, or as a speech recognition device to implement the above processing logic in an integrated or external manner. As an implementation method, the speech recognition device performs speech recognition on the first voice information to determine the target user and voice scene corresponding to the first voice information; then configures the black and white list corresponding to the target user based on the voice scene, and the black and white list is used to adjust the voice end value; and in response to the triggering of the second voice information, updates the list resource according to the black and white list, and the second voice information is the voice information in the subsequent session of the first voice information; and then determines the voice end value corresponding to the second voice information through the updated list resource, so as to identify the second voice information based on the voice end value. Thereby, a personalized voice end value configuration process is realized. Since the black and white list configuration is combined with the user's voice characteristics, and the list resources are maintained through the black and white list, the voice end value can be adapted to different users, improving the integrity of the content after the user's voice input, and improving the accuracy of speech recognition.

[0083] Combined with the above process architecture, the speech recognition method in this application is introduced below. Figure 3 , Figure 3 A flowchart of a speech recognition method provided in an embodiment of the present application, the embodiment of the present application at least includes the following steps:

[0084] 301. Perform voice recognition on first voice information to determine a target user and a voice scene corresponding to the first voice information.

[0085] In this embodiment, speech recognition is performed based on voice activity detection technology (VAD), which is used to identify speech activity segments and non-speech activity segments in audio signals. Voice activity refers to audio segments that contain voice information, while non-speech activity refers to audio segments that do not contain voice information. Specifically, this embodiment uses semantic VAD to judge the semantic completeness of voice information, obtains the semantic completeness probability of the recognized content, and then dynamically adjusts the tail silence waiting time during the speech recognition process, wherein the tail silence waiting time is the end of speech value (End of Speech), which is used to segment the speech segments for recognition.

[0086] In one possible scenario, the speech recognition process of this embodiment is applied to smart device control scenarios such as in-vehicle voice control and smart home voice control. The voice information in this scenario is characterized by concise language and multi-phrase control. Compared with general long text voice content, it is more prone to voice truncation and long waiting problems. Therefore, targeted voice end values ​​are designed to address these problems in order to avoid the impact of the above scenarios on voice integrity.

[0087] Specifically, the dynamic configuration process of the above-mentioned voice end value is the process of configuring the blacklist and whitelist based on the voice content of different users in the voice truncation and long waiting scenarios. Figure 4 As shown, Figure 4 A flow chart of a black and white list configuration provided for an embodiment of the present application; the figure shows the process of configuring black and white lists for users in a targeted manner according to different scenarios, namely, first obtaining voiceprint registration information, the voiceprint registration process takes into account that users may have different accents, belong to different age groups, and their speaking speeds and speaking habits are different. For example, the voice habits of users in Shandong area often have inverted sentences, so personalized registration configuration is performed for each user; then the first voice information is compared with the voiceprint registration information to determine the target user; and voice recognition is performed on the first voice information to determine the voice recognition result associated with the target user; and semantic discrimination is performed on the voice recognition result to determine the voice scene corresponding to the first voice information.

[0088] Among them, voice scenarios include early cut-off scenarios, long waiting scenarios and other scenarios; in early cut-off scenarios, there are pauses in the user's speech, and the interaction ends prematurely before the end of the speech, resulting in incomplete user semantics. In long waiting scenarios, after the user has finished speaking, the voice recognition system is still waiting for a fixed EOS duration, resulting in a long recognition response time.

[0089] Therefore, for the process of determining the truncated scene, first obtain the first content and the second content whose content is adjacent in the speech recognition result; then determine the completeness information corresponding to the first content and the second content, and the completeness information is set based on the content integrity and semantic integrity; if the completeness information indicates that the content of the first content and the second content is complete, and the content after the semantic association of the first content and the second content is complete, then determine that the speech scene corresponding to the first speech information is a truncated scene.

[0090] Specifically, the semantic discrimination process of the truncation scene is to compare the semantic completeness of the user's previous and subsequent voice recognition contents (the first content and the second content) within a limited time to determine the truncation scene: if the user performs voice recognition twice in a row within a limited time, the semantics of the first recognition are incomplete, while the semantics of the second recognition are relatively complete, and the result of the second recognition is semantically complete after some words are added to the end of the result of the first recognition, then the scene is determined to be a truncation scene.

[0091] In addition, for the semantic discrimination process of the long waiting scene, the execution information used to trigger the execution of the action in the speech recognition result is first determined; if the semantics indicated by the execution information is complete and the action corresponding to the execution information is not executed, then the speech scene corresponding to the first voice information is determined to be a long waiting scene.

[0092] Specifically, the semantic identification process of the long waiting scene is that the user performs voice recognition only once within a limited time, and it is determined that the recognition content is semantically complete and triggers a certain type of execution action (execution information), then the scene is considered to be a suspected long waiting scene.

[0093] Furthermore, different users may have different perceptions of long waits, and therefore user confirmation is required, that is, if the semantics of the execution information indication are complete and the execution information is used to indicate the target action, a prompt content indicating the response time is generated; then, in response to the user's confirmation operation on the prompt content, it is determined that the voice scene corresponding to the first voice information is a long wait scene.

[0094] In one possible scenario, for the configuration of the prompt content, that is, when the user needs to confirm, the system pops up a window to ask the user whether the voice recognition response time they feel is too long and whether it can be optimized. After the user gives an affirmative reply (confirmation operation), the confirmation operation can be a voice reply or a click reply, and the scene can be determined as a long waiting scene.

[0095] In a possible scenario, the voice recognition process is the process of interaction between the terminal side and the cloud, and the terminal side is used by multiple users. At this time, user management is performed by registering voiceprints using the voice recognition service. That is, when using a smart car machine, there are more than one user of the smart home device. In this embodiment, the user who uses the voice recognition service has registered voiceprint information, and a set of black and white lists are established for each user on the terminal side, and the user list resource for newly registered voiceprints is empty. Then, when the user performs voice recognition, the cloud voice recognition system captures the truncation scene and long waiting scene for each user through the semantic discrimination module.

[0096] Through the configuration of the above blacklist and whitelist, the personalized blacklist and whitelist based on semantic VAD recognition are updated and stored in the end-side device. The end-side device will upload it to the cloud-based speech recognition service the next time speech recognition is performed.

[0097] 302. Configure a blacklist and a whitelist corresponding to the target user based on the voice scenario, where the blacklist and the whitelist are used to adjust the voice end value.

[0098] In this embodiment, the blacklist in the black and white list is used to adjust the voice end value to the upper limit, with the purpose of collecting complete voice content as much as possible; and the whitelist in the black and white list is used to adjust the voice end value to the lower limit, with the purpose of saving time as much as possible while collecting complete voice content.

[0099] Therefore, for the configuration process of the blacklist and whitelist, if the voice scene is a truncated scene, the semantically incomplete speech recognition content in the truncated scene is obtained, and the semantically incomplete speech recognition content is configured to the blacklist in the blacklist and whitelist corresponding to the target user, that is, for the truncated scene, the system captures the first semantically incomplete speech recognition content and adds it to the user's blacklist.

[0100] For voice scenarios that are long-wait scenarios, the semantically complete speech recognition content in the long-wait scenario is obtained, and the semantically complete speech recognition content is configured to the whitelist in the blacklist and whitelist corresponding to the target user, that is, for long-wait scenarios, the system adds the semantically complete speech recognition content to the user's whitelist.

[0101] 303. In response to the triggering of the second voice information, the list resource is updated according to the black and white lists, and the second voice information is the voice information in the subsequent conversation of the first voice information.

[0102] In this embodiment, the list resource is a collection of black and white lists, which can be configured locally, but considering the huge amount of data, it can also be configured in the cloud to improve the efficiency of speech recognition; this embodiment takes the cloud configuration as an example for explanation.

[0103] Specifically, the second voice information is the voice information in the subsequent conversation, which can be the next round of conversation or the specified subsequent conversation (for example, a user-specified conversation to start voice recognition optimization); therefore, in the process of updating the list resources in the cloud, the black and white lists can be sent to the cloud in response to the trigger of the second voice information, and the second voice information can be subjected to voiceprint recognition in the cloud to determine the target user, and the list resources corresponding to the target user are loaded; then, the list resources corresponding to the target user are updated according to the black and white lists.

[0104] The process of updating the list resources is a hot update process, that is, after the next round of conversation starts, the personalized black, white and gray list resources of all users of the end-side device are uploaded to the cloud-based voice recognition service. The recognition service loads the black and white list resources of the corresponding user according to the voiceprint information, completes the hot update and takes effect in the current conversation.

[0105] It can be seen that the semantic VAD black, white and gray list hot update mechanism can realize the online upgrade of the semantic VAD service and avoid the restart of the speech recognition service. It can speed up the update speed and greatly reduce the development and operation and maintenance costs of the speech recognition service.

[0106] The cloud-based voice identifier is combined with voiceprint identification to obtain the user's feedback on the voice recognition system and the corresponding voice recognition results. The semantic identification module extracts the user's personalized blacklist and whitelist resources based on the user's feedback on the voice recognition system and the user's voice recognition results and stores them on the client side. The user-level blacklist and whitelist resources on the client side are uploaded at the beginning of cloud-based voice recognition and take effect in the current recognition session, realizing the hot update of semantic VAD personalized list resources.

[0107] 304. Determine the voice end value corresponding to the second voice information through the updated list resource, so as to recognize the second voice information based on the voice end value.

[0108] In this embodiment, the process of determining the voice end value corresponding to the second voice information, that is, determining the hit status of the second voice information for the scene, is firstly based on the voice content corresponding to the second voice information, matching with the updated list resource to obtain the list hit information; then determining the hit scene indicated by the list hit information; and determining the voice end value corresponding to the second voice information based on the hit scene, so as to identify the second voice information based on the voice end value, that is, segmenting the second voice information according to the duration corresponding to the voice end value, and then performing voice recognition on the segmented content.

[0109] It is understandable that considering that users may have different accents and belong to different age groups, their speaking speed and speaking habits are different. For example, users in Shandong often have inverted sentences in their speech habits. That is, in the scenario of a large number of users, the amount of data corresponding to the maintenance of the list resources is huge, and the computing power consumption of configuring the list resources one by one is too much. At this time, the EOS value can be determined based on the user group, that is, for the list hit information indicating that there is no hit scenario matching the second voice information in the list resource, the user group associated with the target user corresponding to the second voice information is determined; then the semantic recognition model corresponding to the user group is obtained; and the voice end value corresponding to the second voice information is determined based on the list resource corresponding to the user group, so as to recognize the second voice information based on the voice end value.

[0110] Therefore, combining the above-mentioned list resources with the speech recognition process of the user group, we can use Figure 5 The process shown, Figure 5A schematic diagram of another process for taking effect of a list resource provided in an embodiment of the present application; the diagram shows that when the next round of voice recognition session (second voice information) is started, the end side uploads the black and white list of all users who have registered voiceprints on the end side to the cloud voice recognition service, and the cloud recognition service dynamically uses the list resource of the corresponding user according to the voiceprint of this voice recognition, and adds it to the public black and white list resource of the recognition service in the form of hot update, which takes effect in the current voice recognition session. At the same time, the voice recognition service performs audio attribute analysis. For unregistered users (or those who hit the content in the list resource), the user group to which the user belongs is determined based on the user's accent, age, and gender, and the model resource used by the semantic VAD service is the model resource of the user group.

[0111] Furthermore, for the semantic recognition model configured for different user groups, the training process is as follows: Figure 6 As shown, Figure 6 A schematic diagram of the configuration of another speech recognition model provided for an embodiment of the present application; the figure shows that training corpus is first obtained; then the training corpus is classified according to preset dimensions based on user group type to obtain training data corresponding to each user group, and the preset dimensions are based on language type dimensions (such as dialect type, language system, etc.) and object description dimension settings (such as age, gender, etc.); and then the semantic recognition model corresponding to each user group is trained according to the training data.

[0112] In one possible scenario, the cloud recognition service divides users into different user groups (such as Shandong dialect adult female group, Minnan dialect child male group, etc.) based on desensitized information such as user accent, age, and gender. The cloud recognition service regularly aggregates blacklist and whitelist resources of different user groups, cleans them, and uses them as fine-tune training corpus for the personalized semantic VAD model of the user group, thereby training semantic VAD models for different user groups.

[0113] It is understandable that the blacklist and whitelist of the first voice information (current session) will also update the semantic recognition model of each user group mentioned above, that is, the list resources in step 303 can be configured by different user groups; that is, the semantic recognition model is updated and launched with the cloud recognition service. In addition to the personal personalized blacklist and whitelist resources, users can also use the semantic VAD model resources of their user groups, and the generalization ability of user-level personalized semantic VAD services is enhanced. For new users who do not have personalized blacklist and whitelist resources, the user group semantic VAD model also improves the user's voice recognition experience.

[0114] In addition, for the acquisition of training corpus, we can regularly collect list resources, retain entries with high repetition rate, and then update the semantic recognition model. Through cloud recognition services, we can divide different user groups according to the results of audio attribute analysis, regularly collect blacklist and whitelist resources of different user groups, train semantic VAD models that adapt to different accents, different ages, and genders, enhance generalization capabilities, and further optimize the user semantic VAD experience.

[0115] It is understandable that the trained semantic recognition model can be applied to smart device control scenarios such as in-vehicle and smart home. Through the personalized semantic VAD system, the problems of early truncation and long waiting are alleviated, which significantly improves the convenience and comfort of users when using in-vehicle and smart home voice interaction systems.

[0116] In combination with the above embodiments, it can be known that the target user and voice scene corresponding to the first voice information are determined by performing voice recognition on the first voice information; then the black and white lists corresponding to the target users are configured based on the voice scene, and the black and white lists are used to adjust the voice end value; and in response to the triggering of the second voice information, the list resources are updated according to the black and white lists, and the second voice information is the voice information in the subsequent conversation of the first voice information; and then the voice end value corresponding to the second voice information is determined through the updated list resources, so as to identify the second voice information based on the voice end value. Thus, a personalized voice end value configuration process is realized. Since the black and white lists are configured in combination with the voice characteristics of the user, and the list resources are maintained through the black and white lists, the voice end value can be adapted to different users, thereby improving the integrity of the content after the user's voice input and improving the accuracy of voice recognition.

[0117] In order to better implement the above solution of the embodiment of the present application, the following also provides related devices for implementing the above solution. Figure 7 , Figure 7 The present invention provides a schematic diagram of the structure of a speech recognition device according to an embodiment of the present invention. The speech recognition device 700 includes:

[0118] The recognition unit 701 is used to perform voice recognition on the first voice information to determine the target user and voice scene corresponding to the first voice information;

[0119] A configuration unit 702, configured to configure a blacklist and a whitelist corresponding to the target user based on the voice scenario, wherein the blacklist and the whitelist are used to adjust the voice end value;

[0120] The configuration unit 702 is further configured to update the list resource according to the black and white list in response to the triggering of the second voice information, wherein the second voice information is the voice information in the subsequent session of the first voice information;

[0121] The recognition unit 701 is further configured to determine a voice end value corresponding to the second voice information through the updated list resource, so as to recognize the second voice information based on the voice end value.

[0122] Optionally, in some possible implementations, the recognition unit 701 is used to obtain voiceprint registration information when performing voice recognition on the first voice information to determine the target user and voice scene corresponding to the first voice information; compare the first voice information with the voiceprint registration information to determine the target user; perform voice recognition on the first voice information to determine the voice recognition result associated with the target user; and perform semantic judgment on the voice recognition result to determine the voice scene corresponding to the first voice information.

[0123] Optionally, in some possible implementations, the recognition unit 701 is used to obtain first and second contents whose contents are adjacent in the speech recognition result when the speech scene is a truncated scene and semantic discrimination is performed on the speech recognition result to determine the speech scene corresponding to the first voice information; determine completeness information corresponding to the first content and the second content, wherein the completeness information is set based on content integrity and semantic integrity; if the completeness information indicates that the contents of the first content and the second content are complete, and the contents of the first content and the second content after semantic association are complete, then determine that the speech scene corresponding to the first voice information is the truncated scene.

[0124] Optionally, in some possible implementations, the recognition unit 701 is used to determine the execution information in the voice recognition result for triggering the execution of an action when the voice scene is a long-wait scene and semantic discrimination is performed on the voice recognition result to determine the voice scene corresponding to the first voice information; if the semantics indicated by the execution information are complete and the action corresponding to the execution information is not executed, it is determined that the voice scene corresponding to the first voice information is the long-wait scene.

[0125] Optionally, in some possible implementations, the recognition unit 701 is used to generate prompt content indicating the response time when, if the semantics indicated by the execution information are complete and the execution information is used to indicate the target action, determining that the voice scene corresponding to the first voice information is the long waiting scene; and in response to the user's confirmation operation on the prompt content, determining that the voice scene corresponding to the first voice information is the long waiting scene.

[0126] Optionally, in some possible implementations, the configuration unit 702 is used to, when configuring the blacklist and whitelist corresponding to the target user based on the voice scene, obtain semantically incomplete speech recognition content in the truncated scene, and configure the semantically incomplete speech recognition content to the blacklist in the blacklist corresponding to the target user; if the voice scene is a long-wait scene, obtain semantically complete speech recognition content in the long-wait scene, and configure the semantically complete speech recognition content to the whitelist in the blacklist and whitelist corresponding to the target user.

[0127] Optionally, in some possible implementations, the configuration unit 702 is used to send the black and white list to the cloud in response to the triggering of the second voice information, perform voiceprint recognition on the second voice information in the cloud to determine the target user, and load the list resources corresponding to the target user; when the list resources are updated according to the black and white list in response to the triggering of the second voice information, the list resources corresponding to the target user are updated according to the black and white list.

[0128] Optionally, in some possible implementations, the recognition unit 701 is used to match the voice content corresponding to the second voice information with the updated list resources to obtain list hit information when determining the voice end value corresponding to the second voice information through the updated list resources to identify the second voice information based on the voice end value; determine the hit scenario indicated by the list hit information; and determine the voice end value corresponding to the second voice information based on the hit scenario to identify the second voice information based on the voice end value.

[0129] Optionally, in some possible implementations, the recognition unit 701 is also used to determine the user group associated with the target user corresponding to the second voice information if the list hit information indicates that there is no hit scene in the list resource that matches the second voice information; obtain the semantic recognition model corresponding to the user group; and determine the voice end value corresponding to the second voice information based on the list resource corresponding to the user group, so as to recognize the second voice information based on the voice end value.

[0130] Optionally, in some possible implementations, the recognition unit 701 is also used to obtain training corpus; classify the training corpus according to preset dimensions based on user group type to obtain training data corresponding to each user group, wherein the preset dimensions are set based on language type dimensions and object description dimensions; and train the semantic recognition model corresponding to each user group according to the training data.

[0131] By performing voice recognition on the first voice information, the target user and voice scenario corresponding to the first voice information are determined; then the black and white lists corresponding to the target users are configured based on the voice scenario, and the black and white lists are used to adjust the voice end value; and in response to the triggering of the second voice information, the list resources are updated according to the black and white lists, and the second voice information is the voice information in the subsequent conversation of the first voice information; and then the voice end value corresponding to the second voice information is determined through the updated list resources, so as to identify the second voice information based on the voice end value. Thus, a personalized voice end value configuration process is realized. Since the black and white lists are configured in combination with the voice characteristics of the user, and the list resources are maintained through the black and white lists, the voice end value can be adapted to different users, thereby improving the integrity of the content after the user's voice input and improving the accuracy of voice recognition.

[0132] The speech recognition device provided in this embodiment belongs to the same application concept as the method provided in the above embodiments of this application, can execute the speech recognition method provided in any of the above embodiments of this application, and has the corresponding functional modules and beneficial effects of executing the speech recognition method. For technical details not fully described in this embodiment, please refer to the specific processing content of the speech recognition method provided in the above embodiments of this application, which will not be repeated here.

[0133] The functions implemented by the above identification unit 701 and configuration unit 702 may be implemented by the same or different processors, respectively, and are not limited in the embodiments of the present application.

[0134] It should be understood that each unit in the above device can be implemented in the form of a processor calling software. For example, the device includes a processor, the processor is connected to a memory, and instructions are stored in the memory. The processor calls the instructions stored in the memory to implement any of the above methods or realize the functions of each unit of the device, wherein the processor can be a general-purpose processor, such as a CPU or a microprocessor, etc., and the memory can be a memory in the device or a memory outside the device. Alternatively, the unit in the device can be implemented in the form of a hardware circuit, and the functions of some or all units can be realized by designing the hardware circuit. The hardware circuit can be understood as one or more processors; for example, in one implementation, the hardware circuit is an ASIC, and the functions of some or all units above are realized by designing the logical relationship of the components in the circuit; for another example, in another implementation, the hardware circuit can be implemented by PLD, taking FPGA as an example, which can include a large number of logic gate circuits, and the connection relationship between the logic gate circuits is configured by a configuration file, so as to realize the functions of some or all units above. All units of the above device can be implemented in the form of a processor calling software, or in the form of a hardware circuit, or in the form of a processor calling software, and the remaining part can be implemented in the form of a hardware circuit.

[0135] In an embodiment of the present application, a processor is a circuit with the ability to process signals. In one implementation, the processor may be a circuit with the ability to read and run instructions, such as a CPU, a microprocessor, a GPU, or a DSP; in another implementation, the processor may implement certain functions through the logical relationship of a hardware circuit, and the logical relationship of the hardware circuit is fixed or reconfigurable, such as a hardware circuit implemented by an ASIC or PLD, such as an FPGA. In a reconfigurable hardware circuit, the process of the processor loading a configuration document to implement the hardware circuit configuration can be understood as the process of the processor loading instructions to implement the functions of some or all of the above units. In addition, it can also be a hardware circuit designed for artificial intelligence, which can be understood as an ASIC, such as an NPU, TPU, DPU, etc.

[0136] It can be seen that each unit in the above device can be one or more processors (or processing circuits) configured to implement the above method, such as: CPU, GPU, NPU, TPU, DPU, microprocessor, DSP, ASIC, FPGA, or a combination of at least two of these processor forms.

[0137] In addition, all or part of the units in the above device can be integrated together, or can be implemented independently. In one implementation, these units are integrated together and implemented in the form of a SOC. The SOC may include at least one processor for implementing any of the above methods or implementing the functions of each unit of the device. The type of the at least one processor may be different, for example, including a CPU and an FPGA, a CPU and an artificial intelligence processor, a CPU and a GPU, etc.

[0138] The present application also provides a speech recognition device, which includes a processor and an input / output component. The input / output component is used to obtain speech information.

[0139] The processor is used to perform voice content recognition on the voice information obtained by the input-output component by executing any one of the voice recognition methods in the above embodiments.

[0140] The above-mentioned interface circuit can be any interface circuit that can realize the data communication function, for example, it can be a USB interface circuit, a Type-C interface circuit, a serial port circuit, a PCIE circuit, etc.

[0141] Optionally, the embodiment of the present application further provides a system, wherein the interactive client is used to obtain voice information and send the voice information to the server, and to display the voice recognition result output by the server;

[0142] The server is used to perform voice content recognition on the voice information obtained by the interactive client by executing the voice recognition method described in any one of the above embodiments.

[0143] Another embodiment of the present application also provides a voice recognition device, see Figure 8 As shown, the device includes:

[0144] Memory 800 and processor 810;

[0145] The memory 800 is connected to the processor 810 and is used to store programs;

[0146] The processor 810 is used to implement the speech recognition method disclosed in any of the above embodiments by running the program stored in the memory 800.

[0147] Specifically, the above-mentioned speech recognition device may further include: a bus, a communication interface 820 , an input device 830 and an output device 840 .

[0148] The processor 810, the memory 800, the communication interface 820, the input device 830 and the output device 840 are connected to each other via a bus.

[0149] A bus may include a pathway that transfers information between components of a computer system.

[0150] The processor 810 may be a general-purpose processor, such as a general-purpose central processing unit (CPU), a microprocessor, etc., or an application-specific integrated circuit (ASIC), or one or more integrated circuits for controlling the execution of the program of the scheme of the present invention. It may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0151] The processor 810 may include a main processor, and may also include a baseband chip, a modem, and the like.

[0152] The memory 800 stores a program for executing the technical solution of the present invention, and may also store an operating system and other key services. Specifically, the program may include a program code, and the program code includes a computer operation instruction. More specifically, the memory 800 may include a read-only memory (ROM), other types of static storage devices that can store static information and instructions, a random access memory (RAM), other types of dynamic storage devices that can store information and instructions, a disk storage, a flash, and the like.

[0153] The input device 830 may include a device for receiving data and information input by a user, such as a keyboard, a mouse, a camera, a scanner, a light pen, a voice input device, a touch screen, a pedometer, or a gravity sensor.

[0154] Output device 840 may include devices that allow information to be output to a user, such as a display screen, printer, speaker, etc.

[0155] The communication interface 820 may include any transceiver or the like to communicate with other devices or communication networks, such as Ethernet, a radio access network (RAN), a wireless local area network (WLAN), etc.

[0156] The processor 810 executes the program stored in the memory 800 and calls other devices, which can be used to implement each step of any speech recognition method provided in the above embodiments of the present application.

[0157] In addition to the above-mentioned methods and devices, an embodiment of the present application may also be a computer program product, which includes computer program instructions, which, when executed by a processor, enable the processor to execute the steps of the speech recognition method according to various embodiments of the present application described in any of the above embodiments of this specification.

[0158] The computer program product may be written in any combination of one or more programming languages ​​to write program codes for performing the operations of the embodiments of the present application, including object-oriented programming languages, such as Java, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as an independent software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0159] In addition, the embodiment of the present application may also be a storage medium on which a computer program is stored. The computer program is executed by a processor to execute the steps of the speech recognition method according to various embodiments of the present application described in any of the above embodiments of this specification, and specifically the following steps may be implemented:

[0160] Performing voice recognition on the first voice information to determine a target user and a voice scene corresponding to the first voice information;

[0161] Based on the voice scenario, a blacklist and a whitelist corresponding to the target user are configured, wherein the blacklist and the whitelist are used to adjust the voice end value;

[0162] In response to the triggering of the second voice information, the list resource is updated according to the black and white list, the second voice information being the voice information in the subsequent conversation of the first voice information;

[0163] The voice end value corresponding to the second voice information is determined through the updated list resource, so as to identify the second voice information based on the voice end value.

[0164] For the aforementioned method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the order of the actions described, because according to the present application, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application.

[0165] It should be noted that each embodiment in this specification is described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same or similar parts between the embodiments can be referred to each other. For the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0166] The steps in the methods of each embodiment of the present application can be adjusted in order, combined and deleted according to actual needs, and the technical features recorded in each embodiment can be replaced or combined.

[0167] The modules and sub-modules in the devices and terminals of the various embodiments of the present application can be combined, divided and deleted according to actual needs.

[0168] In the several embodiments provided in the present application, it should be understood that the disclosed terminals, devices and methods can be implemented in other ways. For example, the terminal embodiments described above are only schematic, for example, the division of modules or submodules is only a logical function division, and there may be other division methods in actual implementation, for example, multiple submodules or modules can be combined or integrated into another module, 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 an indirect coupling or communication connection through some interfaces, devices or modules, which can be electrical, mechanical or other forms.

[0169] The modules or submodules described as separate components may or may not be physically separated, and the components of the modules or submodules may or may not be physical modules or submodules, that is, they may be located in one place, or they may be distributed on multiple network modules or submodules. Some or all of the modules or submodules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0170] In addition, each functional module or submodule in each embodiment of the present application may be integrated into one processing module, or each module or submodule may exist physically separately, or two or more modules or submodules may be integrated into one module. The above-mentioned integrated modules or submodules may be implemented in the form of hardware or in the form of software functional modules or submodules.

[0171] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0172] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented directly by hardware, software units executed by a processor, or a combination of the two. The software units may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0173] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0174] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A speech recognition method, characterized in that: include: Performing voice recognition on the first voice information to determine a target user and a voice scene corresponding to the first voice information; Based on the voice scenario, a blacklist and a whitelist corresponding to the target user are configured, wherein the blacklist and the whitelist are used to adjust the voice end value; In response to the triggering of the second voice information, the list resource is updated according to the black and white list, the second voice information being the voice information in the subsequent conversation of the first voice information; The voice end value corresponding to the second voice information is determined through the updated list resource, so as to identify the second voice information based on the voice end value.

2. The method according to claim 1, characterized in that Performing voice recognition on the first voice information to determine a target user and a voice scenario corresponding to the first voice information includes: Get voiceprint registration information; Comparing the first voice information with the voiceprint registration information to determine the target user; Performing voice recognition on the first voice information to determine a voice recognition result associated with the target user; The speech recognition result is semantically determined to determine the speech scene corresponding to the first speech information.

3. The method according to claim 2, characterized in that The voice scene is a truncated scene, and the semantic discrimination of the voice recognition result to determine the voice scene corresponding to the first voice information includes: Acquire first content and second content whose contents are adjacent to each other in the speech recognition result; Determining integrity information corresponding to the first content and the second content, wherein the integrity information is set based on content integrity and semantic integrity; If the completeness information indicates that the first content and the second content are complete, and the semantically associated content of the first content and the second content is complete, it is determined that the voice scene corresponding to the first voice information is the truncated scene.

4. The method according to claim 2, characterized in that: The voice scene is a long-wait scene, and the semantic discrimination of the voice recognition result to determine the voice scene corresponding to the first voice information includes: Determining execution information in the speech recognition result for triggering execution of an action; If the semantics indicated by the execution information are complete and the action corresponding to the execution information has not been executed, it is determined that the voice scene corresponding to the first voice information is the long waiting scene.

5. The method according to claim 4, characterized in that If the semantics indicated by the execution information is complete and the execution information is used to indicate a target action, determining that the voice scene corresponding to the first voice information is the long waiting scene includes: If the semantics indicated by the execution information are complete and the execution information is used to indicate the target action, then a prompt content indicating the response time is generated; In response to a confirmation operation of the user on the prompt content, it is determined that the voice scene corresponding to the first voice information is the long-wait scene.

6. The method according to claim 1, characterized in that The configuring the blacklist and whitelist corresponding to the target user based on the voice scenario includes: If the voice scene is a truncated scene, obtaining semantically incomplete voice recognition content in the truncated scene, and configuring the semantically incomplete voice recognition content to a blacklist in the blacklist and whitelist corresponding to the target user; If the voice scene is a long-wait scene, semantically complete voice recognition content in the long-wait scene is obtained, and the semantically complete voice recognition content is configured into a whitelist in the blacklist and whitelist corresponding to the target user.

7. The method according to claim 1, characterized in that The updating of the list resource according to the black and white list in response to the triggering of the second voice information includes: In response to the triggering of the second voice information, the blacklist and whitelist are sent to the cloud, the second voice information is subjected to voiceprint recognition in the cloud to determine the target user, and the list resource corresponding to the target user is loaded; The list resource corresponding to the target user is updated according to the black and white list.

8. The method according to claim 1, characterized in that The step of determining the voice end value corresponding to the second voice information by using the updated list resource, so as to identify the second voice information based on the voice end value, includes: Matching the voice content corresponding to the second voice information with the updated list resource to obtain list hit information; Determining the hit scenario indicated by the list hit information; The voice end value corresponding to the second voice information is determined based on the hit scenario, so as to identify the second voice information based on the voice end value.

9. The method according to claim 8, characterized in that The method further comprises: If the list hit information indicates that there is no hit scenario matching the second voice information in the list resource, determining a user group associated with a target user corresponding to the second voice information; Obtaining a semantic recognition model corresponding to the user group; The voice end value corresponding to the second voice information is determined based on the list resource corresponding to the user group, so as to identify the second voice information based on the voice end value.

10. The method according to claim 9, characterized in that The training process of the semantic recognition model corresponding to the user group includes: Get training corpus; Classifying the training corpus according to preset dimensions based on user group types to obtain training data corresponding to each user group, wherein the preset dimensions are set based on a language type dimension and an object description dimension; The semantic recognition model corresponding to each user group is trained according to the training data.

11. A speech recognition device, characterized in that: include: a recognition unit, configured to perform voice recognition on the first voice information to determine a target user and a voice scene corresponding to the first voice information; A configuration unit, configured to configure a blacklist and a whitelist corresponding to the target user based on the voice scenario, wherein the blacklist and the whitelist are used to adjust the voice end value; The configuration unit is further configured to update the list resource according to the black and white list in response to the triggering of the second voice information, wherein the second voice information is the voice information in the subsequent conversation of the first voice information; The recognition unit is further used to determine the voice end value corresponding to the second voice information through the updated list resource, so as to recognize the second voice information based on the voice end value.

12. A speech recognition device comprising an input and output component and a processor; The input and output components are used to obtain voice information; The processor is used to perform voice content recognition on the voice information obtained by the input-output component by executing the voice recognition method as described in any one of claims 1 to 10.

13. A speech recognition system, comprising an interactive client and a server; The interactive client is used to obtain voice information and send the voice information to the server, and to display the voice recognition results output by the server; The server is used to perform voice content recognition on the voice information obtained by the interactive client by executing the voice recognition method according to any one of claims 1 to 10.

14. A computer program product, characterized in that include: A computer program, wherein when the computer program is executed by a processor, the speech recognition method according to any one of claims 1 to 10 is implemented.

Images