Voice recognition method and apparatus, electronic device, computer-readable storage medium, and computer program product

Abstract

The present application provides a voice recognition method and apparatus, an electronic device, a computer-readable storage medium, and a computer program product. The method comprises: performing integrity detection on received voice data until complete voice data is detected, and determining whether a silence duration reaches a silence duration threshold; when the silence duration reaches the silence duration threshold, determining first voice data; and performing voice recognition on the first voice data to obtain a recognition result corresponding to the first voice data. The present application can improve the accuracy of voice recognition.

Description

Speech recognition method and device, electronic equipment, computer readable storage medium and computer program product

[0001] Cross-reference to related applications

[0002] The present application is based on and claims priority to Chinese Patent Application No. 202411386472.5, filed on September 29, 2024, the entire contents of which are incorporated herein by reference. TECHNICAL FIELD

[0003] The present application relates to the technical field of speech processing, and in particular to a speech recognition method and device, electronic equipment, computer readable storage medium and computer program product. BACKGROUND

[0004] Currently, in the case that the duration of silence during the detection of user speech exceeds a preset duration, the speech activity detection (VAD) technology determines that the user has finished speaking, triggers a VAD end, and sends the received speech data to a speech recognition module for speech recognition. However, if the user hesitates for more than the preset duration of silence during the process of issuing a voice command, the speech will be truncated, which will further cause the speech recognition to be incorrect.

[0005] To solve the above problem, the related art usually triggers a VAD end to stop speech reception in the case that the duration of silence exceeds a preset duration, and then performs a completeness judgment on the truncated speech to determine whether the user's voice command has been input completely. If the judgment is that it is not complete, the speech data is received for a period of time, and finally the fragments of speech data obtained by several times of truncation are combined, analyzed and recognized. However, this method may lose some speech data between several times of truncation, so the speech data after combination is still an incomplete speech fragment, which results in low accuracy of speech recognition, and further reduces the accuracy of speech recognition. SUMMARY

[0006] The embodiments of the present application provide a speech recognition method, device, electronic equipment, computer readable storage medium and computer program product, which can improve the accuracy of speech recognition.

[0007] The technical solutions of the embodiments of the present application are as follows:

[0008] The embodiments of the present application provide a speech recognition method, which comprises:

[0009] Performing completeness detection on the speech data received from the start of speech activity, and determining whether the duration of silence reaches a silence duration threshold if complete speech data is detected.

[0010] determine first speech data in a case where the silence duration reaches the silence duration threshold;

[0011] perform speech recognition on the first speech data to obtain a recognition result corresponding to the first speech data.

[0012] Embodiments of the present application provide a speech recognition device, comprising:

[0013] a speech detection module, configured to perform integrity detection on speech data received since a speech activity starts, and determine whether a silence duration reaches a silence duration threshold if complete speech data is detected;

[0014] a speech determination module, configured to determine first speech data in a case where the silence duration reaches the silence duration threshold;

[0015] a speech recognition module, configured to perform speech recognition on the first speech data to obtain a recognition result corresponding to the first speech data.

[0016] Optionally, the speech detection module is further configured to, after performing integrity detection on the received speech data, continue receiving the speech data if complete speech data is not detected, and perform integrity detection on the speech data received since the speech activity starts after a first time interval.

[0017] Optionally, the speech detection module is further configured to, in a case where the silence duration reaches the silence duration threshold, stop receiving the speech data, and take the speech data received between the start of receiving and the stop of receiving as the first speech data.

[0018] Optionally, the speech determination module is further configured to, before performing speech recognition on the first speech data to obtain a recognition result corresponding to the first speech data, stop receiving if the silence duration in the first time interval is greater than the silence duration threshold, and take the speech data received between the start of receiving and the stop of receiving as the first speech data.

[0019] Optionally, the speech recognition module is further configured to determine whether a semantic intention of the first speech data is clear, and perform speech recognition on the first speech data to obtain a recognition result corresponding to the first speech data in a case where the semantic intention of the first speech data is clear.

[0020] Optionally, the voice recognition module is further configured to, after determining whether the semantic intention of the first voice data is clear, in a case where the semantic intention of the first voice data is not clear, detect whether voice activity occurs within a first time length; in a case where it is detected that voice activity occurs within the first time length, determine second voice data corresponding to the voice activity occurring within the first time length; splice the first voice data and the second voice data to obtain third voice data; and perform voice recognition on the third voice data to obtain a recognition result corresponding to the third voice data.

[0021] Optionally, the voice recognition module is further configured to, after detecting whether voice activity occurs within a first time length, in a case where no voice activity is detected within the first time length, perform voice recognition on the first voice data to obtain a recognition result corresponding to the first voice data.

[0022] Optionally, the voice detection module is further configured to, in a case where the silence time length does not reach the silence time length threshold, if voice data is received, determine that the voice activity has not ended.

[0023] Optionally, the voice detection module is further configured to, after determining whether the semantic intention of the first voice data is clear, in a case where the semantic intention of the first voice data is clear, detect whether voice activity occurs within a second time length; in a case where it is detected that voice activity occurs within the second time length, determine fourth voice data corresponding to the voice activity occurring within the second time length; determine a correlation degree of the first voice data and the fourth voice data; in a case where the correlation degree is greater than or equal to a correlation degree threshold, splice the first voice data and the fourth voice data to obtain fifth voice data; and perform voice recognition on the fifth voice data to obtain a recognition result corresponding to the fifth voice data.

[0024] Optionally, the voice determination module is further configured to perform integrity detection on voice data received in the voice activity occurring within the first time length, and if a silence time length after detecting complete voice data reaches the silence time length threshold, determine the voice data received between starting to receive and stopping to receive in the voice activity within the first time length as the second voice data.

[0025] Alternatively, the voice determination module is further configured to perform integrity detection on voice data received in the voice activity occurring within the second time length, and if a silence time length after detecting complete voice data reaches the silence time length threshold, determine the voice data received between starting to receive and stopping to receive in the voice activity within the second time length as the fourth voice data.

[0026] Optionally, the voice detection module is further configured to extract acoustic features and text features from the voice data received since the voice activity; and perform integrity detection on the voice data received since the voice activity according to the acoustic features and the text features.

[0027] An electronic device is provided in an embodiment of the present application, comprising:

[0028] a memory configured to store computer-executable instructions;

[0029] a processor configured to execute the computer-executable instructions stored in the memory to implement the voice recognition method provided in the embodiments of the present application.

[0030] A computer-readable storage medium is provided in an embodiment of the present application, which stores a computer program or computer-executable instructions, and is configured to implement the voice recognition method provided in the embodiments of the present application when executed by a processor.

[0031] A computer program product is provided in an embodiment of the present application, which comprises a computer program or computer-executable instructions, and the computer program or computer-executable instructions are configured to implement the voice recognition method provided in the embodiments of the present application when executed by a processor.

[0032] The embodiments of the present application have the following beneficial effects:

[0033] During the voice data receiving process, the integrity of the voice data is detected until the complete voice data is detected, and then the judgment of the silence duration is performed. In the case where the silence duration reaches the silence duration threshold, the first voice data is determined. Thus, the voice data receiving is truncated by the judgment of the silence duration after the complete voice data is obtained based on the integrity detection. In this way, the determined first voice data meets the requirements of the integrity detection and the silence duration, thereby improving the integrity of the first voice data and further improving the accuracy of the voice recognition on the first voice data. BRIEF DESCRIPTION OF DRAWINGS

[0034] FIG. 1 is an optional flowchart of a voice recognition method provided in an embodiment of the present application;

[0035] FIG. 2 is an optional flowchart of a voice recognition method provided in an embodiment of the present application;

[0036] FIG. 3 is an optional flowchart of a voice recognition method provided in an embodiment of the present application;

[0037] FIG. 4 is an optional flowchart of a voice recognition method provided in an embodiment of the present application;

[0038] FIG. 5 is an optional flowchart of a voice recognition method applied to an actual scenario in an embodiment of the present application;

[0039] Fig. 6 is a structural schematic diagram of a speech recognition apparatus according to an embodiment of the present application;

[0040] Fig. 7 is a structural schematic diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings, and the described embodiments should not be regarded as limiting the present application. All other embodiments obtained by those of ordinary skill in the art without making any creative labor fall within the scope of protection of the present application.

[0042] In the following description, "some embodiments" are described, which describe a subset of all possible embodiments, but it can be understood that "some embodiments" can be the same subset or different subset of all possible embodiments, and can be combined with each other without conflict.

[0043] In the following description, the terms "first\second\third" are only to distinguish similar objects, and do not represent a specific order of the objects. It can be understood that "first\second\third" can be interchanged with a specific order or sequence as allowed, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein.

[0044] In the embodiments of the present application, the term "module" or "unit" refers to a computer program or a part of a computer program with a predetermined function, and works with other related parts to achieve a predetermined target, and can be implemented entirely or partially by using software, hardware (such as a processing circuit or a memory) or a combination thereof. Similarly, one processor (or multiple processors or memories) can be used to implement one or more modules or units. In addition, each module or unit can be a part of an integral module or unit that includes the functions of the module or unit.

[0045] Unless otherwise defined, all technical and scientific terms used in the embodiments of the present application have the same meanings as those commonly understood by those skilled in the art. The terms used in the embodiments of the present application are only for the purpose of describing the embodiments of the present application, and are not intended to limit the present application.

[0046] Currently, to ensure that the complete speech of a user is intercepted for speech recognition, the related technologies usually determine whether the user has finished speaking by detecting the speech activity of the user through VAD technology. For example, after detecting that the user starts speaking, it is determined that the speech activity starts (VAD-start), and the speech data of the user is continuously received. If it is detected that the user is silent (i.e., no longer speaking), a preset time period (silence duration, usually a few milliseconds) is waited for, and if the silence state still exists, it is considered that the user has finished speaking, and the currently received speech audio is sent to a speech recognition module for speech recognition. However, in the process of actual speech interaction of the user, hesitation or pause may occur in the process of issuing a speech instruction, so that the time of not speaking exceeds the silence duration, resulting in that the speech of the user is truncated, and the incomplete speech data is sent to the speech recognition module for speech recognition. For example, the user says "I want to listen to...Chopin's violin", and the duration of the pause exceeds the silence duration, so that only the "I want to listen" part of the speech data before the pause is truncated for speech recognition. Or, the user may speak a complete sentence, but the meaning of the sentence is not clear, which may also affect the recognition of the user's intention, resulting in inaccurate speech recognition results.

[0047] For the above problems, the related technologies usually first trigger VAD end to stop speech reception in the case that the silence duration times out, and then determine the completeness of the truncated speech to determine whether the speech instruction of the user is input completely. If it is determined that the speech is not complete, the speech data for a period of time is continuously received, and finally the fragments of the speech data truncated several times are combined, analyzed, and recognized. However, this way may lose some speech data between the several truncations, so that the speech data after the combination is still an incomplete speech fragment, thereby resulting in low accuracy of recognizing the speech data, and further reducing the accuracy of speech recognition. Moreover, the related technologies currently do not have a corresponding solution for the case that the user speaks a complete sentence, but the meaning of the sentence is not clear, thereby resulting in inaccurate speech recognition.

[0048] The embodiments of the present application provide a speech recognition method and device, electronic equipment, computer readable storage medium and computer program product, which can improve the accuracy of speech recognition. The method of the embodiments of the present application can be applied to electronic equipment, and the electronic equipment provided by the embodiments of the present application can be implemented as a notebook computer, a tablet computer, a desktop computer, a set-top box, a mobile device (for example, a mobile phone, a portable music player, a personal digital assistant, a dedicated message device, a portable game device) and various types of user terminals, and can also be implemented as a server. In some embodiments, the electronic equipment can include a vehicle control unit VCU, a car computer and other vehicle terminals on a vehicle, and the specific selection is made according to the actual situation, which is not limited in the embodiments of the present application.

[0049] Referring to FIG. 1, FIG. 1 is a first flowchart of a voice recognition method provided by an embodiment of the present application, as shown in FIG. 1, comprising S101-S103, as follows:

[0050] S101, performing integrity detection on voice data received since voice activity starts, and if complete voice data is detected, determining whether the silence duration reaches a silence duration threshold.

[0051] In S101, voice detection is performed based on a VAD technology, in a case where voice activity starts is detected, a voice activity start time is recorded, that is, VAD start, and real-time and continuous voice data reception is started, and integrity detection is performed on the real-time received voice data, to detect whether the voice data is complete. Exemplarily, the voice data comprises a voice instruction issued by a user.

[0052] Exemplarily, in a silent environment, voice activity of human voice is detected, and it can be determined that voice activity starts.

[0053] In some embodiments, in a case where voice activity starts is detected, whether the received voice data is complete is detected, until a case where the voice data is complete is detected, whether the silence duration reaches a silence duration threshold is determined.

[0054] In some embodiments, the data received by voice activity can be periodically detected according to a preset time interval, for example, a first time interval. It can be understood that, in a case where voice activity continuously receives new voice data, the voice data required for integrity detection by the first network model each time can comprise voice data from the voice activity start time to the time of the current detection.

[0055] In S101, in a case where the voice data is complete is detected, it indicates that the user may have finished a complete sentence. The electronic device can start recording the silence duration, to determine whether the silence duration reaches a preset silence duration threshold. It should be noted that the silence duration is accumulated from the silence state when the voice activity is detected.

[0056] In some embodiments, if complete voice data is not detected, that is, in a case where the voice data received since voice activity starts is detected to be incomplete, it is determined that the voice activity has not ended, and the voice data of the voice activity is continuously received, and after a first time interval, the voice data received since the voice activity starts is detected, and such periodic detection is performed until complete voice data is detected. Wherein, after the first time interval, the integrity detection is performed on the voice data, which comprises all the voice data received since the voice activity starts and in the process of the first time interval.

[0057] It should be noted that the first time interval can be a preset fixed time interval, or a variable time interval according to actual conditions. For example, the specific duration of the first time interval is determined according to the text content or acoustic characteristics of the received voice data, and the like. The specific selection is based on actual conditions, and the embodiments of the present application are not limited.

[0058] S102, in the case where the silence duration reaches the silence duration threshold, determining the first voice data.

[0059] Here, since the integrity of the voice data has been determined in S101, in the case where the silence duration reaches the silence duration threshold, it means that the user pauses for a silence duration threshold time after inputting a complete sentence. This situation usually represents that the user has completed all voice data input. Therefore, the electronic device determines that the voice activity ends, triggers the VAD end in the VAD detection, stops receiving the voice data, records the voice activity end time, and records the voice data received between the voice activity start and the voice activity end as the first voice data.

[0060] It can be seen that, compared with the technical solution in the related art that detects voice activity, starts VAD start, and triggers VAD end after a fixed silence duration, the VAD end in the embodiments of the present application can be dynamically postponed according to the integrity of the currently received voice data of the user. That is, in the case where the voice data of the user is incomplete, the silence duration is not judged, which is equivalent to postponing the VAD end, and in the case where the voice data is complete, the silence duration is judged, and then the VAD end is determined, so as to ensure that the first voice data between the VAD start and the VAD end is complete voice data, thereby avoiding cutting off the voice in the case where the voice data is incomplete, reducing data loss, and improving the accuracy of voice recognition on the first voice data.

[0061] In some embodiments, in the case where the silence duration does not reach the silence duration threshold, if voice data is received, it is determined that the voice activity has not ended, and the silence duration is reset until the silence duration reaches the silence duration threshold and it is determined that the voice activity has ended.

[0062] For example, in the case where the voice activity enters the silence state, but the silence duration does not reach the silence duration threshold, if new voice data is received, it means that the user has started voice input again. In order to ensure the integrity of the voice data, it is determined that the voice activity has not ended, VAD end is not triggered, and the silence duration is reset, such as being cleared, until the voice activity enters the silence state again, the silence duration is accumulated from zero, and the voice activity is determined to end when the silence duration reaches the silence duration threshold.

[0063] S103, performing speech recognition on the first voice data to obtain a recognition result corresponding to the first voice data.

[0064] In S103, in a case where the first voice data is determined, speech recognition is performed on the first voice data that passes the integrity detection to obtain a recognition result corresponding to the first voice data.

[0065] Exemplarily, the recognition result corresponding to the first voice data can include a recognized user operation instruction, a query request, or a device interaction request, etc. The specific selection is based on actual conditions, and the embodiments of the present application are not limited.

[0066] It can be understood that in the voice data receiving process, the integrity of the voice data is detected until the complete voice data is detected, and then the judgment of the silence duration is performed. In a case where the silence duration reaches the silence duration threshold, the first voice data is determined. Thus, the receiving of the voice data is truncated by the judgment of the silence duration based on the integrity detection to obtain the complete voice data. In this way, the determined first voice data meets the requirements of the integrity detection and the silence duration, thereby improving the integrity of the first voice data and further improving the accuracy of the speech recognition on the first voice data.

[0067] In some embodiments, in the process of detecting the currently continuously received voice data, in a case where the voice data is incomplete, it is determined that the voice activity has not ended, and the voice data continues to be received. After a first time interval, the next integrity detection is initiated, and so on. If the voice activity appears in a silence state within the first time interval, and the silence duration within the first time interval is greater than the silence duration threshold, it can be determined that the voice activity has ended, triggering the VAD end and stopping the reception of the voice data. It can be seen that in this case, the time point of the VAD end represents the time point of the first time interval counting moment extended by the silence duration threshold. The voice data received between the start of the reception (the VAD start time point) and the stop of the reception (the VAD end time point, i.e. the time point of the first time interval counting moment extended by the silence duration threshold) is taken as the first voice data, and the first voice data is further subjected to speech recognition to obtain a recognition result corresponding to the first voice data.

[0068] That is, in the process of dynamically postponing the VAD end according to the voice integrity detection, if the silence duration greater than the silence duration threshold appears, it indicates that the user may want to end the current voice input, at this time, it is determined that the voice activity has ended, and the voice data is intercepted for detection, which can avoid long waiting for user input and improve the response speed of voice interaction.

[0069] In some embodiments, acoustic features and text features can be extracted from the voice data received since the start of the voice activity; and integrity detection is performed on the voice data received since the start of the voice activity according to the acoustic features and the text features.

[0070] Exemplarily, the integrity detection on the voice data can be performed by a pre-trained first network model. The first network model can extract acoustic features and / or text features from the currently received voice data, and determine whether the currently received voice data is complete voice data according to the extracted acoustic features and / or text features.

[0071] Exemplarily, the acoustic features and the text features of the voice data received since the start of the voice activity are extracted by the first network model; and it is determined whether the received voice data is complete according to the acoustic features and the text features. For example, according to the acoustic features, it can be determined whether the sound features of the voice data are complete. For example, the sound features being complete can represent whether the voice data conforms to the sound characteristics of normal voice data start and end, or does not conform to the sound characteristics of sudden interruption or truncation. According to the text features, it can be determined whether the text content of the voice data is complete. In the case that the sound features of the voice data are complete and the text content is complete, it is determined that the voice data is complete, and the integrity detection is performed.

[0072] Exemplarily, the first network model can include, but is not limited to, any one of a Long Short-Term Memory (LSTM), a Convolutional Neural Networks (CNN), or a text recognition neural network, or a combination of two or more thereof, which is specifically selected according to actual conditions, and the embodiments of the present application are not limited.

[0073] It can be understood that, by combining the acoustic features and the text features of the voice data for integrity detection, the accuracy of the integrity detection is improved, and thus the accuracy of the voice recognition on the first voice data obtained based on the integrity detection is improved.

[0074] In some embodiments, based on FIG. 1, as shown in FIG. 2, S103 can be implemented by performing the processes of S1031-S1032, as follows:

[0075] S1031, determine whether the semantic intention of the first voice data is clear.

[0076] In the embodiments of the present application, based on the first voice data determined through the integrity detection, it can be further determined whether the semantic intention of the first voice data is clear before performing voice recognition on the first voice data. In some embodiments, the first voice data is input into a second network model, and the semantic intention of the first voice data is determined through the second network model to identify the semantic intention of the first voice data, to determine whether the semantic intention of the first voice data is clear.

[0077] Exemplarily, the second network model can determine whether the semantic intention of the first voice data is clear by converting the first voice data into text, and determining the semantic intention of the first voice data according to the converted text.

[0078] Exemplarily, the second network model can include a natural language processing (NLP) model, and can also include other types of neural network models for analyzing semantic intention, which is selected according to actual conditions, and the embodiments of the present application are not limited.

[0079] S1032, in the case that the semantic intention of the first voice data is clear, performing voice recognition on the first voice data to obtain a recognition result corresponding to the first voice data.

[0080] In the embodiments of the present application, in the case that the semantic intention of the first voice data is clear, it means that the data integrity and intention clarity of the first voice data have passed the check, and the first voice data can be input into a voice recognition model to perform voice recognition on the first voice data to obtain a recognition result corresponding to the first voice data as the recognition result of the voice activity.

[0081] It can be understood that by identifying whether the semantic intention of the first voice data is clear, the final voice recognition is triggered in the case that the semantic intention is clear, and the recognition result is obtained. In this way, voice recognition based on complete and clear-intention voice data is realized, the influence of incomplete voice data or unclear-intention voice data on the final recognition result is reduced, and the accuracy of voice recognition is improved.

[0082] In some embodiments, as shown in FIG. 3, after determining whether the semantic intention of the first voice data is clear through S1031, S201-S204 can also be performed, as follows:

[0083] S201, in the case that the semantic intention of the first voice data is not clear, detecting whether voice activity occurs within the first time length.

[0084] In S201, in the case that the semantic intention of the first voice data is not clear, it is explained that the user may continue to supplement the previous voice. The electronic device waits for a preset first time length, and detects whether voice activity occurs within the first time length. Here, the first time length can be set according to actual conditions, and the embodiments of the present application are not limited.

[0085] In S202, in the case that voice activity occurs within the first time length, the second voice data corresponding to the voice activity occurring within the first time length is determined.

[0086] In S202, in the case that voice activity occurs within the first time length, it is explained that the user starts voice input again in the case that the semantic intention of the previously input first voice data is not clear, the correlation between the two voice inputs is high, and the electronic device receives the voice data received through the voice activity occurring within the first time length as the corresponding second voice data.

[0087] In some embodiments, the voice data received through the voice activity occurring within the first time length can be subjected to integrity detection. If the silence time length after detecting the complete voice data reaches a silence time length threshold, the voice data received between the start of receiving and the stop of receiving the voice activity within the first time length is determined as the second voice data.

[0088] Alternatively, the reception of voice data can also be started when it is detected that voice activity occurs within the first time length, and the reception of voice data is stopped when the silence time length of the voice activity occurring within the first time length reaches a silence time length threshold, and the voice data received between the start of receiving and the stop of receiving the voice activity within the first time length is determined as the second voice data.

[0089] That is, the voice activity occurring within the first time length, the VAD start time when the voice activity starts within the first time length, and the VAD end time when the voice activity ends can be determined according to whether the silence time length is greater than a preset silence time length threshold, or the voice data of the voice activity occurring within the first time length can be subjected to integrity detection in the same way as S101-S102. In the case that the voice data of the voice activity occurring within the first time length is complete, it is determined whether the silence time length reaches the silence time length threshold, and in the case that the silence time length reaches the silence time length threshold, it is determined that the voice activity occurring within the first time length ends. In this way, the voice data received between the start of the voice activity and the end of the voice activity of the voice activity occurring within the first time length is taken as the second voice data.

[0090] In S203, the first voice data and the second voice data are spliced to obtain third voice data.

[0091] S204, perform speech recognition on the third speech data to obtain a recognition result corresponding to the third speech data.

[0092] In the embodiments of the present application, since the second speech data has high correlation with the first speech data, the second speech data is spliced after the first speech data to obtain the third speech data. In this way, the third speech data contains complete speech input by the user, and the speech recognition is performed on the third speech data to obtain a recognition result corresponding to the third speech data.

[0093] It can be understood that, by receiving the second speech data corresponding to the speech activity occurring in the first time length in the case that the semantic intention of the first speech data is not clear, splicing the first speech data and the second speech data to obtain the third speech data, and identifying the recognition result corresponding to the third speech data, the completeness of the speech data is further improved, thereby improving the accuracy of the speech recognition based on the speech data.

[0094] In some embodiments, after detecting whether speech activity occurs in the first time length, if no speech activity is detected in the first time length, it indicates that the user no longer inputs speech, and the speech recognition is no longer continued on the first speech data to obtain a recognition result corresponding to the first speech data. In this way, the response speed of the user's speech instruction can be improved.

[0095] In some embodiments, after determining whether the semantic intention of the first speech data is clear in S1031, S301-S305 can also be performed as shown in FIG. 4, as follows:

[0096] S301, in the case that the semantic intention of the first speech data is clear, detecting whether speech activity occurs in a second time length.

[0097] S302, in the case that speech activity occurs in the second time length, determining fourth speech data corresponding to the speech activity occurring in the second time length.

[0098] In some embodiments, the speech data received by the speech activity occurring in the second time length can be subjected to completeness detection. If the silence time after detecting complete speech data reaches a silence time threshold, the speech data received between the start of receiving and the stop of receiving the speech activity in the second time length is determined as the fourth speech data.

[0099] Or, the receiving of the voice data can also be started when voice activity in a second time length is detected, the voice data receiving is stopped when a silence time length in the second time length reaches a silence time length threshold, the voice data received between the start of receiving and the stop of receiving of the voice activity in the second time length is determined as fourth voice data.

[0100] That is, the voice activity in the second time length, the VAD start time when the voice activity starts in the second time length, and the VAD end time when the voice activity ends can be determined according to whether the silence time length is greater than the preset silence time length threshold, or the same method as S101-S102 can be used to detect the integrity of the voice data of the voice activity in the second time length, and in the case that the voice data of the voice activity in the second time length is complete, it is determined whether the silence time length reaches the silence time length threshold, and in the case that the silence time length reaches the silence time length threshold, it is determined that the voice activity in the second time length ends. In this way, the voice data received between the start of the voice activity and the end of the voice activity of the voice activity in the second time length is taken as the fourth voice data.

[0101] S303, determine the relevance of the first voice data and the fourth voice data.

[0102] S304, in the case that the relevance is greater than or equal to a relevance threshold, splice the first voice data and the fourth voice data to obtain fifth voice data.

[0103] S305, perform voice recognition on the fifth voice data to obtain a recognition result corresponding to the fifth voice data.

[0104] Here, in the case that the semantic intent of the first voice data is clear, a preset second time length is still waited to determine whether voice activity occurs in the second time length. The second time length can be the same as or different from the first time length. In the case that voice activity in the second time length is detected, the voice data corresponding to the voice activity in the second time length is determined as the fourth voice data through a similar process to the second voice data corresponding to the determination of the voice activity in the first time length. Since the fourth voice data is obtained in the case that the semantic intent of the first voice data is clear, it is necessary to further verify the relevance of the first voice data and the fourth voice data. In the case that the relevance is greater than or equal to a relevance threshold, the first voice data and the fourth voice data are spliced, and the fourth voice data is spliced after the first voice data, for example, to obtain fifth voice data. Further, the fifth voice data is subjected to voice recognition to obtain a recognition result corresponding to the fifth voice data.

[0105] In some embodiments, the relevance of the first voice data and the fourth voice data can be determined by a third network model.

[0106] In this way, for some users, after inputting complete and intended voice data, the user then makes a supplement, such as the user first says "please draw a flower pattern", and after a pause exceeding a preset silence duration threshold, the user then opens his mouth to make a supplement in the second duration, saying "the flower is red". In this case, although the previous voice input "please draw a flower pattern" is complete and intended, it needs to be combined with the subsequent input "the flower is red" with high relevance to perform voice recognition to completely recognize the user's intention.

[0107] It can be understood that, by the method in the embodiment of the present application, two pieces of voice data with high relevance input by the user in sequence can be combined and then subjected to voice recognition, greatly improving the accuracy of voice recognition.

[0108] It should be noted that, if the relevance is less than the relevance threshold, it means that the fourth voice data input later is irrelevant to the first voice data, and therefore the first voice data is not spliced, and the first voice data is input into the voice recognition model to perform voice recognition on the first voice data to obtain the recognition result corresponding to the first voice data.

[0109] Next, the voice recognition method of the embodiment of the present application will be described in combination with an actual scenario, as shown in FIG. 5, as follows:

[0110] In actual application, when the user speaks, VAD start (voice activity start) is triggered, and every T-max milliseconds (first time interval), detection is performed on whether the user has finished speaking (whether the received voice data is complete), and the judgment is provided by model A (first network model). Model A will determine whether the user has finished speaking according to the text converted from the current received voice data and the acoustic features of the current received voice data.

[0111] If model A determines that the user has finished speaking, that is, in the case that the received voice data from the start of voice activity is complete voice data, VAD end (determination of voice activity end) is triggered when the silence duration reaches the silence duration threshold (such as 400 ms of silence), and a judgment on whether the semantic intention of the current complete received voice data (first voice data) is clear is performed at the same time. The judgment capability is provided by model B (second network model), and if the judgment result is that the semantic intention of the first voice data is clear, the first voice data is subjected to voice recognition to obtain the recognition result corresponding to the first voice data.

[0112] If model A judges that the user has not finished speaking, VAD end forces to extend the waiting T-max milliseconds, if the user continues to speak in T-max milliseconds, the corresponding speech data is received and the next round of integrity detection is continued. If the user does not continue to speak in T-max milliseconds, VAD end is triggered, and speech recognition is performed on the speech data received from the start of speech activity.

[0113] If model A judges that the user has finished speaking, and model B judges that the semantic intention of the first speech data is not clear, if speech activity (second speech activity) occurs within X ms (first time length) thereafter, the first speech data and the second speech data received within the subsequent X ms are spliced to obtain third speech data, and then speech recognition is performed on the third speech data to obtain the recognition result corresponding to the third speech data.

[0114] It can be understood that the embodiments of the present application can intelligently judge whether the user has finished speaking or not, instead of waiting for a fixed silence duration as in the related art, thereby reducing the influence of the user's hesitation to speak and thinking on the speech recognition, and improving the accuracy of speech recognition.

[0115] The embodiments of the present application provide a speech recognition method, including the following: receiving initial speech data, in the case that the initial speech data is incomplete, obtaining a target silence duration threshold, the target silence duration threshold including an initial silence duration threshold and a forced silence duration threshold, in the case that the silence duration of the initial speech data does not reach the target silence duration threshold, continuing to receive supplementary speech data, and obtaining target speech data, the target speech data including the initial speech data and the supplementary speech data.

[0116] In some embodiments, speech detection is performed based on VAD technology, in the case that speech activity is detected to start, based on the time when the speech activity starts, i.e. VAD start, real-time and continuous reception of initial speech data is started, and the received initial speech data can be detected in real time for integrity. In one example, a polling mechanism with a period of 200 milliseconds can be used to detect the integrity of the initial speech data received from the start of speech activity. For example, a complete speech data sample can be used to train a first network model, and the trained first network model polls at a period of 200 milliseconds to detect whether the received initial speech data is complete. It can be understood that the above-mentioned "200 milliseconds" is only an example, and the present application can also use other periods, such as 100 milliseconds, 150 milliseconds, 250 milliseconds, 300 milliseconds, etc.

[0117] In the voice recognition system, an initial silence duration threshold and a forced silence duration threshold are pre-stored, and a target silence duration threshold can be the sum of the initial silence duration threshold and the forced silence duration threshold. When it is detected that the initial voice data is incomplete, the initial silence duration threshold and the target silence duration threshold are obtained, the supplementary voice data is continuously received, and the completeness of the supplementary voice data is detected. The initial voice data and the supplementary voice data are determined as the target voice data. When the silence duration of the received supplementary voice data reaches the initial silence duration threshold but does not reach the target silence duration threshold, the VAD end is not triggered, but the supplementary voice data is continuously received and the completeness of the continuously received supplementary voice data is detected. The initial voice data and the continuously received supplementary voice data are determined as the target voice data. When the target voice data is incomplete, and the silence duration of the received supplementary voice data reaches the target silence duration threshold, the VAD end is triggered, and the reception of the supplementary voice data is stopped. Or when the target voice data is complete, and the silence duration of the received supplementary voice data reaches the initial silence duration threshold, the VAD end is triggered, and the reception of the supplementary voice data is stopped.

[0118] In some embodiments, when the target voice data is complete, the initial silence duration threshold is obtained, and when the silence duration of the target voice data reaches the initial silence duration threshold, the target voice data is determined as the first voice data.

[0119] In some embodiments, when it is detected that the target voice data is incomplete, and the silence duration of the target voice data does not reach the target silence duration threshold, it is determined that the voice activity is not ended. At this time, the supplementary voice data can be continuously received, and the initial voice data and the continuously received supplementary voice data are determined as the target voice data. When the target voice data is complete and the silence duration of the target voice data reaches the initial silence duration threshold, the target voice data is determined as the first voice data.

[0120] It can be understood that the application can receive multiple pieces of supplementary voice data, splice the initial voice data and the multiple pieces of supplementary voice data to obtain target voice data, and perform integrity detection on the target voice data until the target voice data is complete. That is, in the case where the initial voice data is incomplete, and the silence duration of the initial voice data does not reach the target silence duration threshold, the first supplementary voice data can be continuously received, and the initial voice data and the first voice data are determined as the target voice data. When it is detected that the target voice data is incomplete, and the silence duration of the first supplementary voice data does not reach the target silence duration threshold, the second supplementary voice data can be continuously received, and the initial voice data, the first voice data and the second voice data are taken as the target voice data. When it is detected that the target voice data is incomplete, and the silence duration of the second supplementary voice data does not reach the target silence duration threshold, the third supplementary voice data can be continuously received, and the initial voice data, the first voice data, the second voice data and the third voice data are taken as the target voice data. In this way, the target voice data is complete, and the silence duration of the target voice data reaches the initial silence duration threshold, triggering the VAD end, and stopping the reception of the voice data.

[0121] In the case where it is detected that the target voice data is complete, and in the case where it is determined that the voice activity (i.e., the target voice data) has a silence state and the duration of the silence state is greater than the initial silence duration threshold, the VAD end can be triggered, and the reception of the target voice data is stopped. In this case, the target voice data is determined as the first voice data. Wherein, the target voice data refers to the voice data received between the start of the voice activity VAD start and the end of the voice activity VAD end, that is, the target voice data includes the initial voice data and the supplementary voice data. After obtaining the first voice data, speech recognition is performed on the first voice data to obtain the recognition result corresponding to the first voice data.

[0122] In some examples, the initial silence duration threshold, the forced silence duration threshold, and the target silence duration threshold can be empirical values.

[0123] In some embodiments, the initial silence duration threshold is adjusted based on the type of the target voice data.

[0124] The initial silence duration threshold can be adjusted according to the type of the received target voice data. The type of the received target voice data can be determined by the first network model. For example, when the first network model determines that the received target voice data is first type voice data, the initial silence duration threshold is maintained or reduced; and when the first network model determines that the received target voice data is second type voice data, the initial silence duration threshold is maintained or increased.

[0125] In one example, the first type of voice data can be task type instructions, and thus the initial silence duration threshold is shortened or kept when the first network model determines that the task type instructions are received. For example, the initial silence duration threshold can be shortened from 500 milliseconds to 300 milliseconds. It can be understood that the initial silence duration threshold can also be shortened to other values by the present application. The task type instructions can refer to instructions for issuing tasks, such as "open the sunroof", "turn off the air conditioner", and the like.

[0126] In one example, the second type of voice data can be interactive type instructions, and thus the initial silence duration threshold is extended or kept when the first network model determines that the interactive type instructions are received. For example, the initial silence duration threshold can be extended from 500 milliseconds to 550 milliseconds, or kept unchanged. It can be understood that the initial silence duration threshold can also be extended to other values by the present application. The interactive type instructions can refer to instructions in a "chat" mode, such as "what's the weather today", and the like.

[0127] It is worth noting that the present application does not limit the type of target voice data. In the present application, the initial silence duration threshold can also be adjusted according to other voice types, such as different adjustments to the initial silence duration threshold when different keywords in the voice are detected. In addition, the present application can also decrease or keep the initial silence duration threshold when the first network model determines that the interactive type instructions are received, and increase or keep the initial silence duration threshold when the first network model determines that the task type instructions are received.

[0128] In some embodiments, the method further comprises: determining a semantic intention of the first voice data; in a case where the semantic intention is clear, recognizing the first voice data to obtain a recognition result of the first voice data; in a case where the semantic intention is unclear, detecting a first voice activity occurring within a first duration, determining second voice data corresponding to the first voice activity, splicing the first voice data and the second voice data to obtain third voice data, and performing voice recognition on the third voice data to obtain a recognition result of the third voice data. The implementation manner of the above-mentioned embodiments can refer to the implementation manners of steps S1031, S201-S204, which will not be described here. The first duration is after a voice activity corresponding to the first voice data ends VAD end.

[0129] In some embodiments, the method further comprises: in the case that the semantic intention is clear, detecting a second voice activity occurring within a second time duration, determining fourth voice data corresponding to the second voice activity; determining a correlation degree of the first voice data and the fourth voice data; in the case that the correlation degree is greater than a correlation degree threshold, splicing the first voice data and the fourth voice data to obtain fifth voice data, and performing voice recognition on the fifth voice data to obtain a recognition result corresponding to the fifth voice data. The implementation manner of the above embodiments can refer to the implementation manners of steps S301-S305, which will not be described herein. The second time duration is after the voice activity end VAD end corresponding to the first voice data.

[0130] In some embodiments, the target silence duration threshold can also be adjusted based on the type of the initial voice data.

[0131] The target silence duration threshold can be adjusted according to the type of the received initial voice data. In some examples, the adjustment of the target silence duration threshold can be realized by adjusting the forced silence duration threshold. The adjustment method of the forced silence duration threshold is similar to the adjustment method of the initial silence duration threshold. For example, the type of the received initial voice data can be determined by the first network model. For example, when the first network model determines that the received initial voice data is first type voice data, the forced silence duration threshold is kept or reduced, thereby keeping or reducing the target silence duration threshold; and when the first network model determines that the received initial voice data is second type voice data, the forced silence duration threshold is kept or increased, thereby keeping or increasing the target silence duration threshold. It can be understood that the application can also reduce or keep the forced silence duration threshold when the first network model determines that an interactive instruction is received, thereby keeping or reducing the target silence duration threshold; and increase or keep the forced silence duration threshold when the first network model determines that a task type instruction is received, thereby keeping or increasing the target silence duration threshold. It is worth noting that the application does not limit the type of the initial voice data. In the application, the forced silence duration threshold can also be adjusted according to other voice types, such as different adjustments of the forced silence duration threshold when different keywords in the voice are detected. In addition, in the application, the forced silence duration threshold can also not be adjusted, such as using a fixed forced silence duration threshold, thereby using a fixed target silence duration threshold. In some examples, the fixed target silence duration threshold can be 2000 milliseconds.

[0132] The embodiment of the application provides a voice recognition device 1, as shown in FIG. 6, comprising:

[0133] The voice detection module 11 is configured to perform integrity detection on the voice data received from the voice activity start, and determine whether the silence duration reaches the silence duration threshold if the complete voice data is detected.

[0134] The voice determination module 12 is configured to determine the first voice data when the silence duration reaches the silence duration threshold.

[0135] The voice recognition module 13 is configured to perform voice recognition on the first voice data to obtain a recognition result corresponding to the first voice data.

[0136] In some embodiments, the voice detection module 11 is further configured to, after performing integrity detection on the received voice data, continue receiving voice data if no complete voice data is detected, and perform integrity detection on the voice data received from the start of the voice activity after the first time interval.

[0137] In some embodiments, the voice detection module 11 is further configured to, when the silence duration reaches the silence duration threshold, stop receiving voice data, and take the voice data received between the start of receiving and the stop of receiving as the first voice data.

[0138] In some embodiments, the voice determination module 12 is further configured to, before performing voice recognition on the first voice data to obtain a recognition result corresponding to the first voice data, stop receiving if the silence duration in the first time interval is greater than the silence duration threshold, and take the voice data received between the start of receiving and the stop of receiving as the first voice data.

[0139] In some embodiments, the voice recognition module 13 is further configured to determine whether the semantic intent of the first voice data is clear, and perform voice recognition on the first voice data to obtain a recognition result corresponding to the first voice data when the semantic intent of the first voice data is clear.

[0140] In some embodiments, the voice recognition module 13 is further configured to, after determining whether the semantic intent of the first voice data is clear, detect whether voice activity occurs in a first duration when the semantic intent of the first voice data is not clear, determine second voice data corresponding to the voice activity occurring in the first duration when voice activity is detected in the first duration, splice the first voice data and the second voice data to obtain third voice data, and perform voice recognition on the third voice data to obtain a recognition result corresponding to the third voice data.

[0141] In some embodiments, the voice recognition module 13 is further configured to, after detecting whether voice activity occurs in a first duration, perform voice recognition on the first voice data to obtain a recognition result corresponding to the first voice data when no voice activity is detected in the first duration.

[0142] In some embodiments, the voice detection module 11 is further configured to, if the voice data is received when the silence duration does not reach the silence duration threshold, determine that the voice activity is not ended.

[0143] In some embodiments, the voice detection module 11 is further configured to, after determining whether the semantic intention of the first voice data is clear, if the semantic intention of the first voice data is clear, detect whether voice activity occurs in a second duration; if it is detected that voice activity occurs in the second duration, determine fourth voice data corresponding to the voice activity occurring in the second duration; determine a relevance degree of the first voice data and the fourth voice data; if the relevance degree is greater than or equal to a relevance degree threshold, splice the first voice data and the fourth voice data to obtain fifth voice data; and perform voice recognition on the fifth voice data to obtain a recognition result corresponding to the fifth voice data.

[0144] In some embodiments, the voice determination module 12 is further configured to perform integrity detection on voice data received in the voice activity occurring in the first duration, and if a silence duration after detecting complete voice data reaches the silence duration threshold, determine the voice data received between the start of receiving and the stop of receiving in the voice activity occurring in the first duration as the second voice data.

[0145] Alternatively, the voice determination module 12 is further configured to perform integrity detection on voice data received in the voice activity occurring in the second duration, and if a silence duration after detecting complete voice data reaches the silence duration threshold, determine the voice data received between the start of receiving and the stop of receiving in the voice activity occurring in the second duration as the fourth voice data.

[0146] In some embodiments, the voice detection module 11 is further configured to extract acoustic features and text features from the voice data received from the start of the voice activity, and perform integrity detection on the voice data received from the start of the voice activity according to the acoustic features and the text features.

[0147] It should be noted that the above description of the device embodiments is similar to the description of the above method embodiments, and has similar beneficial effects as the method embodiments. For technical details not disclosed in the device embodiments of the present application, please refer to the description of the method embodiments of the present application for understanding.

[0148] The embodiment of the present application further provides an electronic device, and Fig. 7 is a schematic structural diagram of an optional electronic device 3 according to the embodiment of the present application. As shown in Fig. 7, the electronic device 3 comprises a memory 32 and a processor 33. The memory 32 and the processor 33 are connected through a communication bus 34. The memory 32 is configured to store executable instructions. The processor 33 is configured to execute the executable instructions stored in the memory 32, so as to implement the voice recognition method provided by the embodiment of the present application.

[0149] The embodiment of the present application provides a vehicle, which comprises the voice recognition device as shown in Fig. 6, or comprises the electronic device as shown in Fig. 7.

[0150] The embodiment of the present application provides a computer readable storage medium storing executable instructions, wherein the executable instructions, when executed by the processor, cause the processor to execute the voice recognition method provided by the embodiment of the present application.

[0151] In some embodiments, the computer readable storage medium can be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, flash memory, magnetic surface memory, optical disc, or CD-ROM, etc.; or can be various devices comprising one or any combination of the above memories.

[0152] In some embodiments, the executable instructions can be in the form of programs, software, software modules, scripts or codes, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and can be deployed in any form, including being deployed as independent programs or being deployed as modules, components, subroutines or other units suitable for use in a computing environment.

[0153] As an example, the executable instructions can but not necessarily correspond to files in a file system, can be stored in a part of a file storing other programs or data, for example, stored in one or more scripts in a Hyper Text Markup Language (HTML) document, stored in a single file dedicated to the program in question, or stored in multiple cooperative files (for example, files storing one or more modules, subroutines or code portions). As an example, the executable instructions can be deployed to execute on one computing device, or on multiple computing devices located at one site, or on multiple computing devices distributed at multiple sites and interconnected through a communication network.

[0154] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0155] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in one or more flowchart illustrations and / or one or more block diagrams.

[0156] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means that implement the functions specified in one or more flowcharts and / or one or more block diagrams.

[0157] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, such that the instructions, which execute on the computer or other programmable apparatus, provide steps for implementing the functions specified in one or more flowcharts and / or one or more block diagrams.

[0158] The above description is merely a preferred embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, and improvements made within the spirit and scope of this application are included within the scope of protection of this application.

Claims

A voice recognition method characterized by comprising: The method comprises: performing integrity detection on voice data received from the start of voice activity, and determining whether the length of silence reaches a silence length threshold if complete voice data is detected; determining first voice data if the length of silence reaches the silence length threshold. performing voice recognition on the first voice data to obtain a recognition result corresponding to the first voice data. The method of claim 1, wherein After the integrity detection on the voice data received from the start of voice activity, the method further comprises: if complete voice data is not detected, continuing to receive voice data, and performing integrity detection on the voice data received from the start of voice activity after a first time interval. The method according to claim 1 or 2, characterized in that If the length of silence reaches the silence length threshold, the method further comprises: stopping the reception of voice data, and taking the voice data received from the start of reception to the stop of reception as the first voice data. The method according to claim 2, characterized in that Before the voice recognition on the first voice data to obtain a recognition result corresponding to the first voice data, the method further comprises: if the length of silence in the first time interval is greater than the silence length threshold, stopping the reception of voice data, and taking the voice data received from the start of reception to the stop of reception as the first voice data. The method according to any one of claims 1 to 4, characterized in that The voice recognition on the first voice data to obtain a recognition result corresponding to the first voice data comprises: determining whether the semantic intent of the first voice data is clear; if the semantic intent of the first voice data is clear, performing voice recognition on the first voice data to obtain a recognition result corresponding to the first voice data. The method according to claim 5, characterized in that After the determination of whether the semantic intent of the first voice data is clear, the method further comprises: if the semantic intent of the first voice data is not clear, detecting whether voice activity occurs within a first time length; if it is detected that voice activity occurs within the first time length, determining second voice data corresponding to the voice activity occurring within the first time length; splicing the first voice data and the second voice data to obtain third voice data; performing voice recognition on the third voice data to obtain a recognition result corresponding to the third voice data. The method according to claim 6, characterized in that After the detection of whether voice activity occurs within the first time length, the method further comprises: if voice activity is not detected within the first time length, performing voice recognition on the first voice data to obtain a recognition result corresponding to the first voice data. The method according to any one of claims 5 to 7, characterized in that After the determination of whether the semantic intent of the first voice data is clear, the method further comprises: if the semantic intent of the first voice data is clear, detecting whether voice activity occurs within a second time length; if it is detected that voice activity occurs within the second time length, determining fourth voice data corresponding to the voice activity occurring within the second time length; determining the relevance of the first voice data and the fourth voice data; if the relevance is greater than or equal to a relevance threshold, splicing the first voice data and the fourth voice data to obtain fifth voice data; perform speech recognition on the fifth voice data to obtain a recognition result corresponding to the fifth voice data. The method according to any one of claims 6 to 8, characterized in that determine that the second voice data includes: perform integrity detection on voice data received during voice activity occurring in the first time length, and if a silence time length after detecting complete voice data reaches the silence time length threshold, determine voice data received between starting receiving and stopping receiving voice activity in the first time length as the second voice data. determine that the fourth voice data includes: perform integrity detection on voice data received during voice activity occurring in the second time length, and if a silence time length after detecting complete voice data reaches the silence time length threshold, determine voice data received between starting receiving and stopping receiving voice activity in the second time length as the fourth voice data. The method according to any one of claims 1 to 9, characterized in that the integrity detection on voice data received from starting receiving voice activity includes: extract acoustic features and text features from the voice data received from starting receiving voice activity; perform integrity detection on the voice data received from starting receiving voice activity according to the acoustic features and the text features. The method according to any one of claims 5 to 10, characterized in that wherein the integrity detection is performed by a first model, and the determination of the semantic intent of the first voice data is performed by a second model. A voice recognition method, characterized by, include: receive initial voice data; in the case that the initial voice data is incomplete, obtain a target silence time length threshold, wherein the target silence time length threshold includes an initial silence time length threshold and a forced silence time length threshold; in the case that a silence time length of the initial voice data does not reach the target silence time length threshold, continue to receive supplementary voice data, and obtain target voice data, wherein the target voice data includes the initial voice data and the supplementary voice data. The method of claim 12, wherein further include: in the case that the target voice data is complete, obtain the initial silence time length threshold; in the case that a silence time length of the target voice data reaches the initial silence time length threshold, stop receiving the supplementary voice data, and determine the target voice data as first voice data. The method according to claim 12 or 13, characterized in that further include: adjust the initial silence time length threshold based on a target type of the target voice data. The method of claim 14, wherein the adjustment of the initial silence time length threshold based on the target type of the target voice data includes: in the case that the target type of the target voice data is instruction type voice data, decrease the initial silence time length threshold. The method according to claim 14 or 15, characterized in that the adjustment of the initial silence time length threshold based on the target type of the target voice data includes: in the case that the target type of the target voice data is interactive type voice data, increase the initial silence time length threshold. The method according to any one of claims 13 to 16, characterized in that further include: determine a semantic intent of the first voice data; in the case that the semantic intent is clear, perform recognition on the first voice data to obtain a recognition result of the first voice data; In a case where the semantic intention is not clear, first voice activity occurring within a first time length is detected, second voice data corresponding to the first voice activity is determined, the first voice data and the second voice data are spliced to obtain third voice data, speech recognition is performed on the third voice data, and a recognition result of the third voice data is obtained. The method of claim 17, wherein Further comprising: In a case where the semantic intention is clear, second voice activity occurring within a second time length is detected, fourth voice data corresponding to the second voice activity is determined; A relevance of the first voice data and the fourth voice data is determined; In a case where the relevance is greater than a relevance threshold, the first voice data and the fourth voice data are spliced to obtain fifth voice data, speech recognition is performed on the fifth voice data, and a recognition result corresponding to the fifth voice data is obtained. The method according to any one of claims 12 to 18, characterized in that Further comprising: The target silence time length threshold is adjusted based on a type of the initial voice data. The method of claim 19, wherein The target silence time length threshold is adjusted based on the type of the initial voice data, and the adjusting comprises: In a case where the type of the initial voice data is instruction type voice data, the target silence time length threshold is decreased. The method according to claim 19 or 20, characterized in that The target silence time length threshold is adjusted based on the type of the initial voice data, and the adjusting comprises: In a case where the type of the initial voice data is interactive type voice data, the target silence time length threshold is increased. A speech recognition apparatus characterized by comprising: The apparatus comprises: a voice detection module configured to perform integrity detection on voice data received from a start of voice activity, and determine whether a silence time length reaches a silence time length threshold if complete voice data is detected; a voice determination module configured to determine first voice data in a case where the silence time length reaches the silence time length threshold; a voice recognition module configured to perform speech recognition on the first voice data, and obtain a recognition result corresponding to the first voice data. An electronic device, characterized by comprising: The electronic device comprises: a memory configured to store computer executable instructions; a processor configured to execute the computer executable instructions stored in the memory, and implement the method in any one of claims 1 to 18. A computer readable storage medium storing computer executable instructions, characterized in that, The computer executable instructions or computer program are executed by the processor to implement the method in any one of claims 1 to 21. A computer program product comprising computer-executable instructions, characterized in that, The computer executable instructions or computer program are executed by the processor to implement the method in any one of claims 1 to 21. A vehicle characterized by comprising: The vehicle comprises the voice recognition apparatus in claim 22, or the electronic device in claim 23.

Images