Voice instruction recognition method, device and equipment, medium and vehicle

Abstract

The invention relates to a voice instruction recognition method, device and equipment, a medium and a vehicle. The method comprises the following steps: receiving a first round of voice sent by a user; identifying and analyzing the first round of speech, and determining whether a first semantic expressed by the first round of speech is a truncated semantic; if yes, executing a clarification strategy to receive a secondary voice sent by the user; and after the secondary voice is received, identifying the secondary voice, and supplementing the first semantics by using the second semantics expressed by the secondary voice to obtain an instruction corresponding to the complete semantics. According to the method and the device, the truncated semantics expressed by the user voice can be recognized and clarified, so that the truncated semantics expressed by the user voice are supplemented, the instruction corresponding to the complete semantics is obtained, and the user experience is improved.

Description

Technical Field

[0001] The present disclosure relates to the field of speech recognition technology, and in particular to a speech command recognition method, device, equipment, medium and vehicle. Background Art

[0002] At present, with the rapid development of the vehicle industry, the voice interaction function of intelligent vehicles has been widely used. Vehicles are usually equipped with a voice input function to receive user voice commands. However, in the application scenario of voice interaction, users may pause or be interrupted when inputting voice commands. These phenomena may cause discontinuity of voice input and semantic truncation.

[0003] When a user pauses while inputting a voice command, and the interval between pauses exceeds the preset threshold, the machine will split the originally complete voice input into two rounds of voice input, causing the voice input to be truncated, resulting in incomplete semantics of the voice input. When the machine recognizes such two rounds of voice input, due to the lack of complete semantic information, the machine may not be able to recognize the user's command, affecting the user's experience. Summary of the invention

[0004] In order to solve the above technical problems, the present disclosure provides a voice command recognition method, device, equipment, medium and vehicle.

[0005] In a first aspect, the present disclosure provides a method for voice command recognition, comprising:

[0006] Receive the first round of voice from the user;

[0007] Identifying and analyzing the first round of speech, and determining whether the first semantics expressed by the first round of speech is truncation semantics;

[0008] If yes, a clarification strategy is executed to receive a second-round speech from the user;

[0009] After receiving the second-round voice, the second-round voice is recognized, and the first semantics is supplemented by the second semantics expressed by the second-round voice to obtain instructions corresponding to the complete semantics.

[0010] Optionally, the identifying and analyzing the first-round speech to determine whether the first semantics expressed by the first-round speech is a truncation semantics includes:

[0011] Recognize the first round of speech to obtain text information;

[0012] Perform dependency syntax analysis on the text information to determine whether the first semantics is truncation semantics.

[0013] Optionally, performing dependency syntactic analysis on the text information to determine whether the first semantics of the first-round speech expression is truncation semantics includes:

[0014] Performing dependency syntactic analysis on the text information to determine the syntactic structure of the text information;

[0015] If the syntactic structure does not include a subject-verb structure, determining that the first semantics is a truncation semantics;

[0016] If the syntactic structure includes a subject-verb structure, analyzing whether the intention expressed by the text information lacks relevant purpose parameters;

[0017] If the relevant purpose parameter is missing, it is determined that the first semantics is truncation semantics.

[0018] Optionally, executing the clarification strategy to receive a second-round speech from the user includes:

[0019] Obtain the user's historical behavior data and the current status of the in-vehicle equipment;

[0020] Predicting the user's intention based on the first semantics, the historical behavior data, and the current state of the in-vehicle device to obtain a prediction result;

[0021] A corresponding clarification strategy is selected and executed according to the prediction result to receive a second round of speech from the user.

[0022] Optionally, selecting and executing a corresponding clarification strategy according to the prediction result to receive a second-round speech from the user includes:

[0023] If the prediction result is an executable instruction, selecting and executing a first clarification strategy;

[0024] The first clarification strategy includes sending a confirmation voice of the executable instruction to the user according to the executable instruction, so as to guide the user to send a second round of voice in response to the confirmation voice;

[0025] The step of recognizing the second-round speech and supplementing the first semantics with the second semantics expressed by the second-round speech to obtain the instruction corresponding to the complete semantics includes:

[0026] Recognize the second-round speech, and determine whether the second semantics expressed by the second-round speech is characterized by executing the executable instruction;

[0027] If so, the executable instruction is determined to be an instruction corresponding to the complete semantics.

[0028] Optionally, selecting and executing a corresponding clarification strategy according to the prediction result to receive a second-round speech from the user includes:

[0029] If the prediction result is an unexecutable instruction lacking a slot, selecting and executing a second clarification strategy;

[0030] The second clarification strategy includes sending a slot inquiry voice to the user according to the slot missing from the unexecutable instruction, so as to guide the user to send a second round of voice in response to the slot inquiry voice;

[0031] The step of recognizing the second-round speech and supplementing the first semantics with the second semantics expressed by the second-round speech to obtain the instruction corresponding to the complete semantics includes:

[0032] Recognize the second-round speech, and determine whether the second semantics expressed by the second-round speech includes target information to supplement the slot;

[0033] If so, the target information is used to supplement the missing slots of the non-executable instructions to form target executable instructions, and the target executable instructions are determined as instructions corresponding to complete semantics.

[0034] Optionally, selecting and executing a corresponding clarification strategy according to the prediction result to receive a second-round speech from the user includes:

[0035] If the prediction result is that the user's intention is not predicted, selecting and executing a third clarification strategy;

[0036] The third clarification strategy includes sending a follow-up voice to the user to guide the user to send a second round of voice to clarify the intention, or silently waiting for the user to send a second round of voice to clarify the intention;

[0037] The step of recognizing the second-round speech and supplementing the first semantics with the second semantics expressed by the second-round speech to obtain the instruction corresponding to the complete semantics includes:

[0038] Recognizing the second-round speech, and concatenating the second semantics expressed by the second-round speech with the first semantics to obtain concatenated semantics;

[0039] The spliced ​​semantically corresponding instructions are determined as complete semantically corresponding instructions.

[0040] In a second aspect, the present disclosure provides a voice command recognition device, comprising:

[0041] A receiving module, used to receive the first round of voice sent by the user;

[0042] An analysis module, used for identifying and analyzing the first-round speech, and determining whether the first semantics expressed by the first-round speech is a truncation semantics;

[0043] A clarification module, used for executing a clarification strategy to receive a second-round speech from a user;

[0044] The processing module is used to recognize the second-round voice after receiving the second-round voice, and use the second semantics expressed by the second-round voice to supplement the first semantics to obtain instructions corresponding to the complete semantics.

[0045] Optionally, when the analysis module identifies and analyzes the first round of speech and determines whether the first semantics expressed by the first round of speech is truncated semantics, it is specifically used to identify the first round of speech to obtain text information; perform dependency syntactic analysis on the text information to determine whether the first semantics is truncated semantics.

[0046] Optionally, when the analysis module performs dependency syntactic analysis on the text information to determine whether the first semantics of the first round of speech expression is truncation semantics, the analysis module is specifically used to perform dependency syntactic analysis on the text information to determine the syntactic structure of the text information; if the syntactic structure does not contain a subject-verb structure, the first semantics is determined to be truncated semantics; if the syntactic structure contains a subject-verb structure, whether the intention expressed by the text information lacks relevant purpose parameters is analyzed; if the relevant purpose parameters are missing, the first semantics is determined to be truncated semantics.

[0047] Optionally, when the clarification module executes the clarification strategy to receive the second round of voice issued by the user, it is specifically used to obtain the user's historical behavior data and the current status of the in-vehicle device; predict the user's intention based on the first semantics, the historical behavior data and the current status of the in-vehicle device to obtain a prediction result; select and execute a corresponding clarification strategy according to the prediction result to receive the second round of voice issued by the user.

[0048] Optionally, when the clarification module selects and executes a corresponding clarification strategy according to the prediction result to receive a second-round voice from the user, if the prediction result is an executable instruction, then select and execute a first clarification strategy; the first clarification strategy includes sending a confirmation voice of the executable instruction to the user according to the executable instruction to guide the user to send a second-round voice in response to the confirmation voice;

[0049] When the processing module recognizes the secondary round of speech and uses the second semantics expressed by the secondary round of speech to supplement the first semantics to obtain the instruction corresponding to the complete semantics, it is specifically used to recognize the secondary round of speech and determine whether the second semantics expressed by the secondary round of speech is represented as executing the executable instruction; if so, the executable instruction is determined as the instruction corresponding to the complete semantics.

[0050] Optionally, when the clarification module selects and executes a corresponding clarification strategy according to the prediction result to receive a second round of voice from the user, if the prediction result is an unexecutable instruction lacking a slot, then select and execute a second clarification strategy; the second clarification strategy includes sending a slot inquiry voice to the user according to the slot lacking the unexecutable instruction, so as to guide the user to send a second round of voice in response to the slot inquiry voice;

[0051] When the processing module recognizes the secondary round of speech and uses the second semantics expressed by the secondary round of speech to supplement the first semantics to obtain instructions corresponding to the complete semantics, it is specifically used to recognize the secondary round of speech and determine whether the second semantics expressed by the secondary round of speech contains target information to supplement the slot; if so, use the target information to supplement the slot missing from the non-executable instruction to form a target executable instruction, and determine the target executable instruction as the instruction corresponding to the complete semantics.

[0052] Optionally, when the clarification module selects and executes a corresponding clarification strategy according to the prediction result to receive a second-round voice from the user, if the prediction result is that the user's intention is not predicted, then select and execute a third clarification strategy; the third clarification strategy includes sending an intentional questioning voice to the user to guide the user to send a second-round voice to clarify the intention, or silently waiting for the user to send a second-round voice to clarify the intention;

[0053] When the processing module recognizes the secondary round of speech and uses the second semantics expressed by the secondary round of speech to supplement the first semantics to obtain instructions corresponding to the complete semantics, it is specifically used to recognize the secondary round of speech, splice the second semantics expressed by the secondary round of speech with the first semantics to obtain the spliced ​​semantics; and determine the instructions corresponding to the spliced ​​semantics as the instructions corresponding to the complete semantics.

[0054] In a third aspect, the present disclosure provides a computer device, including:

[0055] Memory;

[0056] Processor; and

[0057] Computer programs;

[0058] The computer program is stored in the memory and is configured to be executed by the processor to implement the method as described in any one of the first aspects.

[0059] In a fourth aspect, the present disclosure provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method as described in any one of the first aspects.

[0060] In a fifth aspect, the present disclosure provides a vehicle, comprising the voice command recognition device as described in the second aspect and / or the computer device as described in the third aspect.

[0061] The voice command recognition method, device, equipment, medium and vehicle provided by the present disclosure recognize and analyze the first round of user's voice when receiving it, determine whether the semantics expressed in the first round of voice is truncated semantics, and if so, execute a clarification strategy to guide the user to issue a second round of voice, then recognize the second round of voice and use the semantics of the second round of voice to supplement the semantics of the first round of voice, so as to obtain the complete semantics that the user wants to express, and then determine the command corresponding to the complete semantics, thereby avoiding the situation where the user's voice is truncated and the user's voice command cannot be recognized, thereby improving the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0063] In order to more clearly illustrate the embodiments of the present invention 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, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0064] Figure 1 A flow chart of a voice command recognition method provided by an embodiment of the present disclosure;

[0065] Figure 2 A schematic diagram of an application scenario provided by an embodiment of the present disclosure;

[0066] Figure 3 A schematic diagram of the structure of a voice command recognition device provided by an embodiment of the present disclosure;

[0067] Figure 4 It is a structural schematic diagram of an electronic device in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0068] In order to more clearly understand the above-mentioned objectives, features and advantages of the present disclosure, the scheme of the present disclosure will be further described below. It should be noted that the embodiments of the present disclosure and the features in the embodiments can be combined with each other without conflict.

[0069] In the following description, many specific details are set forth to facilitate a full understanding of the present disclosure, but the present disclosure may also be implemented in other ways different from those described herein; it is obvious that the embodiments in the specification are only part of the embodiments of the present disclosure, rather than all of the embodiments.

[0070] Figure 1 The present invention provides a flow chart of a method for voice command recognition provided by an embodiment of the present invention. The method can be executed by a voice command recognition device. The voice command recognition device can be implemented in software and / or hardware. The voice command recognition device can be configured in an electronic device, such as a server or a terminal, wherein the terminal includes a vehicle computer, a vehicle-mounted computer, etc. The method includes the following steps:

[0071] S101: Receive a first round of voice from a user.

[0072] For example, Figure 2 This is a schematic diagram of an application scenario provided by an embodiment of the present disclosure, which includes a vehicle 200, a vehicle computer 210, a microphone 220, and a user 230. The user 230 can perform voice interaction with the vehicle computer 210 through the microphone 220. For example, the user 230 can issue voice commands to the vehicle computer 210 through the microphone 220 to control the on-board equipment of the vehicle 200.

[0073] exist Figure 2 In the illustrated scenario, when the user 230 speaks a voice, the vehicle computer 210 receives the voice through the microphone 220 and uses the language as the first round of voice in this voice interaction.

[0074] S102, identifying and analyzing a first round of speech, and determining whether a first semantic meaning expressed by the first round of speech is a truncation semantic meaning.

[0075] The vehicle computer 210 recognizes the first round of speech through ASR (Auto Speech Recognition) technology. ASR technology can convert the first round of speech into text information, and then analyze the text to determine whether the first semantics expressed by the first round of speech is truncated semantics. In one embodiment, when analyzing the text, the vehicle computer 210 can perform semantic analysis on the text information to determine whether the user intention expressed by the text information is complete. If it is incomplete, it can be determined that the semantics of the text information is truncated. For example, the text information of the first round of speech is "open". The semantic analysis of the text information can show that the text information still lacks an object that accepts the action of "open". Therefore, the intention expressed by the text information "open" is incomplete, and it can be determined that the first semantics expressed by the first round of speech is truncated semantics. Among them, semantic analysis can be performed using the NLU (Natural Language Understanding) module. The NLU module is a semantic understanding module in the task-based dialogue system. It can convert the text transcribed after the user's voice input into a structured representation that is easy for the machine to understand, including three main functions: domain recognition, intent recognition, and slot filling.

[0076] In another embodiment, identifying and analyzing a first round of speech to determine whether a first semantics expressed by the first round of speech is a truncated semantics includes: identifying the first round of speech to obtain text information; performing dependency syntactic analysis on the text information to determine whether the first semantics is a truncated semantics.

[0077] When the vehicle computer 210 recognizes the first round of speech and obtains text information, it can use the n-gram language model to convert the first round of speech into text information. The n-gram language model is a language model commonly used to implement ASR technology. After obtaining the text information of the first round of speech through the n-gram language model, the dependency parsing technology (Dependency Parsing) can be used to analyze the grammatical relationship between tokens in a sentence to determine whether the current text information expresses complete semantics. Token refers to the smallest unit of Chinese processing, which can be a word or a character.

[0078] In this case, dependency syntactic analysis is performed on the text information to determine whether the first semantics of the first round of speech expression is truncated semantics, including: performing dependency syntactic analysis on the text information to determine the syntactic structure of the text information; if the syntactic structure does not contain a subject-verb structure, determining that the first semantics is truncated semantics; if the syntactic structure contains a subject-verb structure, analyzing whether the intention expressed by the text information lacks relevant purpose parameters; if the relevant purpose parameters are missing, determining that the first semantics is truncated semantics.

[0079] In task-based dialogues in car-use scenarios, voice commands with complete semantics are mostly executable phrases containing a "verb-object structure". Therefore, by performing dependency syntactic analysis on the text information, identifying the dependency relationship between words in the sentence, and inferring the syntactic structure of the text information, and judging whether the syntactic structure contains a "verb-object structure", it can be determined whether the first semantics of the first round of voice expression is truncated semantics. When the syntactic structure of the text information does not contain a verb-object structure, it can be determined that the first semantics is truncated semantics.

[0080] For example, when performing dependency syntactic analysis, the text "open" only has an action without an object, and the text "air conditioning" is a single entity without an action. In the results of dependency syntactic analysis, the text "open" and the text "air conditioning" are just a single node, and neither side has an edge of a subject-object structure, so both belong to incomplete semantics, that is, truncated semantics. Compared with the unexecutable "single point" dependency syntactic analysis results, the results of complete semantics contain rich dependency syntactic tree relationships. For example, the dependency syntactic analysis results of the text "open the air conditioning" contain a clear subject-object relationship. For example, in the dependency syntactic analysis results of the text "open the air conditioning in the back row", in addition to the subject-object structure, there is also "back row" as the attributive component. Therefore, the syntactic structures of the text "open the air conditioning" and the text "open the air conditioning in the back row" both contain edges of subject-object structures, and both belong to complete semantics.

[0081] When the syntactic structure of the text information does not contain a verb-object structure, the intention expressed by the text information is identified through semantic analysis, and it is determined whether the text information lacks the purpose parameter related to the intention. If the purpose parameter is missing, it can be determined that the first semantics is truncated semantics. Taking the text information of language input as "turn on the air conditioner" and "adjust the air conditioner" as examples, in natural language analysis, the intention of "turn on the air conditioner" will be divided into a class that does not require a purpose parameter, and the intention of "adjust the air conditioner" will be divided into a class that requires a purpose parameter. Because "turn on the air conditioner" belongs to the intention that does not require a purpose parameter, it can be directly determined that it does not lack the relevant purpose parameters, and "adjust the air conditioner" belongs to the intention that requires a purpose parameter, so it will be determined whether the text information lacks the purpose parameter related to the intention. The purpose parameter refers to the necessary information or conditions related to the execution of the intention. For the intention of "adjusting the air conditioner", the purpose parameters may include target temperature, adjustment mode, wind speed, etc. If these relevant parameters are missing in the text information, it can be determined that the first semantics of the first round of voice input is truncated semantics. Therefore, the semantics of the text information "turn on the air conditioner" is complete semantics, while the semantics of the text information "adjust the air conditioner" is truncated semantics. In the embodiments of the present disclosure, the intent recognition and classification of text information and the subsequent determination of whether there is a lack of purpose parameters related to the intent can be achieved by a deep learning model.

[0082] The disclosed embodiment obtains the corresponding text information by identifying the first round of speech, and then performs dependency syntactic analysis on the text information to determine whether the syntactic structure of the text information contains a subject-object structure. When the subject-object structure is not contained, it can be determined that the first semantics expressed by the first round of speech is a truncated semantics, thereby quickly determining the truncated semantics of the first semantics expressed by the first round of speech that lacks the subject-object structure. When the subject-object structure is contained, whether the first semantics expressed by the first round of speech is a truncated semantics is determined by analyzing whether the intention expressed by the text information lacks relevant purpose parameters, thereby improving the accuracy of truncated semantic recognition.

[0083] If yes, execute S103.

[0084] S103: Execute a clarification strategy to receive a second round of speech from the user.

[0085] After determining that the first semantics of the first round of speech expression of the user 230 is a truncated semantics, the vehicle computer 210 can execute a clarification strategy of asking the user a follow-up question, and send a follow-up question through the microphone 220, thereby guiding the user 230 to reply, or it can also use a silent clarification strategy to wait for the user 230 to reply, and the vehicle computer 210 uses the received speech of the user 230 as the second round of speech, so that the semantics of the first round of language expression can be supplemented by the semantics of the second round of speech expression. In addition to using follow-up questions as a clarification strategy, waiting for the user to speak again can also be used as a clarification strategy.

[0086] S104: After receiving the second-round speech, recognize the second-round speech, and use the second semantics expressed by the second-round speech to supplement the first semantics, so as to obtain an instruction corresponding to the complete semantics.

[0087] After receiving the second round of speech from the user 230 through the microphone 220, the vehicle computer 210 recognizes the second semantics expressed by the second round of speech, and uses the second semantics to supplement the first semantics, thereby obtaining the complete semantics of the user 230 and determining the command corresponding to the complete semantics. For example, the first semantics of the first round of speech is "turn on", and the second semantics of the second round of speech is "air conditioning". After the second semantics are used to supplement the first semantics, the complete semantics of the user is obtained as "turn on the air conditioning". The vehicle computer can determine that the command the user wants to execute is to turn on the air conditioning, thus avoiding the situation where the first round of speech is truncated and the first round of speech and the second round of speech are respectively identified to fail to obtain the command with complete semantics.

[0088] The disclosed embodiment recognizes and analyzes the first round of speech from the user when receiving the first round of speech, determines whether the semantics expressed in the first round of speech is truncated semantics, and if so, executes a clarification strategy to guide the user to issue a second round of speech, then recognizes the second round of speech and uses the semantics of the second round of speech to supplement the semantics of the first round of speech, thereby obtaining the complete semantics that the user wants to express, and then determines the instructions corresponding to the complete semantics, thereby avoiding the situation where the user's voice is truncated and the user's voice instructions cannot be recognized, thereby improving the user experience.

[0089] In some embodiments, a clarification strategy is executed to receive a second round of voice issued by the user, including: obtaining the user's historical behavior data and the current state of the in-vehicle device; predicting the user's intention based on the first semantics, the historical behavior data and the current state of the in-vehicle device to obtain a prediction result; selecting and executing a corresponding clarification strategy based on the prediction result to receive a second round of voice issued by the user.

[0090] The user's historical behavior data can reflect the user's car-using habits. The car computer can predict the user's intention by combining the first semantics of the first round of voice expression, historical behavior data, and the current state of the in-car equipment. For example, when the user just gets in the car, the user says "turn on". The car-using habits reflected by the historical behavior data show that most users will turn on the air conditioner after getting in the car, and the air conditioner in the car is not turned on at this time. The car computer can predict that the user's intention is to turn on the air conditioner; when the air conditioner in the car is turned on, excluding the intention to turn on the air conditioner, the car-using habits reflected by the historical behavior data show that the user may open the curtains after getting in the car. At this time, the car computer will predict that the user's intention is to open the curtains. In short, the car computer can summarize the probability of each user's car-using habit through historical behavior data, and then combine the current state of the in-car equipment to predict some car-using habits that have occurred or will not occur, and then use the remaining car-using habits with the highest probability as the predicted user intention, which can be achieved through machine learning models in specific implementation. In addition, when predicting intentions, you can also combine the user's current behavior to make predictions.

[0091] After obtaining the prediction result of the user's intention, the vehicle computer will select and execute the corresponding clarification strategy according to the prediction result. Taking into account the three situations of the prediction results, the embodiments of the present disclosure provide three different follow-up clarification strategies respectively. After implementing different clarification strategies, the specific processing methods in S104 are also different. The following introduces the clarification strategies selected for the three different situations of the prediction results, as well as the specific processing methods of S104 after implementing different clarification strategies.

[0092] In one embodiment, a corresponding clarification strategy is selected and executed based on the prediction result to receive a second round of voice issued by the user, including: if the prediction result is an executable instruction, a first clarification strategy is selected and executed; the first clarification strategy includes issuing a confirmation voice of the executable instruction to the user based on the executable instruction to guide the user to issue a second round of voice in response to the confirmation voice.

[0093] The prediction result is an executable instruction, which indicates that the predicted user intention is complete, that is, the semantics expressed by the prediction result is complete. For example, when the prediction result is "turn on the air conditioner", the vehicle computer determines that the semantics of "turn on the air conditioner" are complete, so the first clarification strategy is selected for execution. The first clarification strategy is to send a follow-up voice to the user to confirm the executable instruction, that is, to ask the user whether the executable instruction is yes or no. For example, the confirmation voice corresponding to "turn on the air conditioner" can be "Do you want me to help you turn on the air conditioner?", clarifying to the user "Do you want me to help you turn on the air conditioner?" to guide the user to reply.

[0094] In this case, the second round of speech is recognized, and the second semantics expressed by the second round of speech is used to supplement the first semantics to obtain instructions corresponding to the complete semantics, including: recognizing the second round of speech, determining whether the second semantics expressed by the second round of speech is represented as executing executable instructions; if so, determining the executable instructions as instructions corresponding to the complete semantics.

[0095] After receiving the second round of voice reply from the user, the car computer identifies the second semantics expressed in the second round of voice, and determines whether the second semantics indicates that the executable instruction can be executed. If so, the car computer uses the executable instruction as the instruction corresponding to the user's complete semantics. For example, if the second round of voice reply from the user is a phrase indicating affirmation, such as "yes", "okay" or "turn on", the car computer uses "turn on the air conditioner" as the instruction with the complete semantics that the user wants to express.

[0096] If the second semantics of the second round of voice expression indicates that the executable instruction cannot be executed, that is, the second round of voice reply from the user is a negative phrase, such as "no", "don't open", etc., then the first round of voice recognition of the next cycle is performed.

[0097] In one embodiment, a corresponding clarification strategy is selected and executed based on the prediction result to receive a second round of voice issued by the user, including: if the prediction result is an unexecutable instruction with a missing slot, a second clarification strategy is selected and executed; the second clarification strategy includes sending a slot inquiry voice to the user with slot corresponding information based on the slot where the unexecutable instruction is missing, so as to guide the user to issue a second round of voice in response to the slot inquiry voice.

[0098] The prediction result is that the unexecutable instruction with a missing slot indicates that the predicted user intention is incomplete. Slot refers to the information unit used to extract user intention and parameters in natural language understanding, that is, part of the user's intention is predicted, but due to the lack of slots, some of the intentions in the prediction result cannot be clearly executed. For example, the first semantics of the user's first round of voice expression is "adjust the air conditioner", and the car usage habits reflected in the historical behavior data are mostly to adjust the air conditioner temperature, but the specific degree of the temperature is uncertain, so the prediction result is "adjust the air conditioner temperature to XX degrees", and "XX degrees" is the missing slot in the unexecutable instruction. So the car computer selects the second clarification strategy for execution. The second clarification strategy is to send a follow-up voice to the user to supplement the slot information for the missing slot in the unexecutable instruction. For example, the slot follow-up voice can be "How many degrees do you want to adjust the air conditioner temperature to?", thereby guiding the user to reply to supplement the information of the missing slot.

[0099] In this case, the second round of speech is recognized, and the second semantics expressed by the second round of speech is used to supplement the first semantics to obtain instructions corresponding to the complete semantics, including: recognizing the second round of speech, determining whether the second semantics expressed by the second round of speech contains target information for supplementing the slots; if so, using the target information to supplement the missing slots of the non-executable instructions to form target executable instructions, and determining the target executable instructions as instructions corresponding to the complete semantics.

[0100] After receiving the second round of voice reply from the user, the car computer identifies the second semantics expressed in the second round of voice, determines whether the second semantics contains the target information for supplementing the slots, and if so, uses the target information to supplement the slots missing from the non-executable instruction, and the non-executable instruction can become an executable instruction after the missing slots are completed. For example, if the second round of voice reply from the user is "25", the car computer will recognize that "25" can be used as the target information for supplementing the slots, so it uses "25" to supplement the missing slots of the non-executable instruction, forming the target executable instruction "adjust the air conditioning temperature to 25 degrees", and uses "adjust the air conditioning temperature to 25 degrees" as the instruction corresponding to the user's complete semantics.

[0101] If there is no target information that can fill the slot in the second semantics of the second round of speech expression, the first round of speech recognition in the next cycle is performed.

[0102] In one embodiment, a corresponding clarification strategy is selected and executed based on the prediction result to receive a second round of speech from the user, including: if the prediction result is that the user's intention is not predicted, then a third clarification strategy is selected and executed; the third clarification strategy includes sending an intentional questioning voice to the user to guide the user to send a second round of speech to clarify the intention.

[0103] When neither of the above two predictions yields a result, that is, the result of the prediction is neither an executable instruction nor an instruction that cannot be executed due to lack of a slot, it means that the user's intention was not predicted when the user intention was predicted. The vehicle computer selects the third clarification strategy for execution. The third clarification strategy is to send a voice message directly asking the user about the intention or silently wait for the user to send a second round of voice message. For example, the first semantic meaning expressed by the user in the first round of voice is "air conditioning". The vehicle computer does not predict the user's intention, so it directly sends a voice message to the user asking about the intention, "Tell me more about it", to guide the user to supplement or explain the specific intention when replying; or, when the vehicle computer does not predict the user's intention, it no longer sends a voice message, and waits for the user to say a second round of voice message to supplement or explain the specific intention within a preset time.

[0104] In this case, the second round of speech is recognized, and the second semantics expressed by the second round of speech is used to supplement the first semantics to obtain instructions corresponding to the complete semantics, including: recognizing the second round of speech, splicing the second semantics expressed by the second round of speech with the first semantics to obtain the spliced ​​semantics; and determining the instructions corresponding to the spliced ​​semantics as the instructions corresponding to the complete semantics.

[0105] After receiving the second round of voice reply from the user, the car computer identifies the second semantics expressed in the second round of voice, and then splices the second semantics with the first semantics. Specifically, when splicing the first semantics and the second semantics, the NLU module is used to process the first semantics and the second semantics respectively, extract their dialogue intentions and corresponding slot information, and then merge the dialogue intentions and slot information of the first semantics with the dialogue intentions and slot information of the second semantics to generate a new semantics containing the information of the two rounds of voice. In addition, context modeling can also be performed by adding some special marks or symbols to the semantic representation after splicing to indicate the order of information in different rounds. For example, a sequence model (such as a recurrent neural network or a Transformer) can be used to introduce a context attention mechanism in the model to model and fuse the context to obtain a better semantic representation. After the splicing is completed, the car computer analyzes whether the spliced ​​semantics is a complete semantics. If so, the instructions corresponding to the spliced ​​semantics are determined as the instructions corresponding to the complete semantics. The method of analyzing semantics has been introduced in the above embodiment, so it will not be repeated here.

[0106] For example, after sending the intended follow-up voice message "Tell me more about this", the user's second-round voice message is "turn off". The second semantics "turn off" and the first semantics "air conditioner" are concatenated, and the concatenated semantics is "turn off the air conditioner". The concatenated semantics contains a subject-object structure and is a complete semantics. Therefore, "turn off the air conditioner" is used as the instruction corresponding to the user's complete semantics.

[0107] The three clarification strategies introduced in the above embodiment can also configure the respective priorities of the three clarification strategies for specific speech recognition products, so as to implement execution according to the priority of the clarification strategies in different scenarios. For example, in some scenarios, the third clarification strategy is more effective, so the third clarification strategy can be configured to be executed first in the speech recognition products applied to these scenarios.

[0108] The disclosed embodiment predicts the user's intention by combining the first semantics of the first round of voice expression, historical behavior data and the current state of the in-vehicle device to obtain a prediction structure, and then provides three corresponding clarification strategies for the three situations of the prediction results, and performs corresponding processing when the three clarification strategies subsequently complete the semantic first round of voice, thereby improving the user experience and increasing the flexibility and applicability of the solution.

[0109] Figure 3The structure diagram of the voice command recognition device provided in the embodiment of the present disclosure is as follows. The voice command recognition device may be a terminal as in the above embodiment, or the voice command recognition device may be a component or assembly in the terminal. The voice command recognition device provided in the embodiment of the present disclosure may execute the processing flow provided in the voice command recognition method embodiment, such as Figure 3 As shown, the voice command recognition device 300 includes:

[0110] The receiving module 301 is used to receive the first round of voice sent by the user;

[0111] An analysis module 302 is used to identify and analyze the first round of speech to determine whether the first semantics expressed by the first round of speech is a truncation semantics;

[0112] The clarification module 303 is used to execute the clarification strategy to receive the second round of speech sent by the user;

[0113] The processing module 304 is used to recognize the second-round speech after receiving the second-round speech, and use the second semantics expressed by the second-round speech to supplement the first semantics to obtain an instruction corresponding to the complete semantics.

[0114] In some embodiments, when the analysis module 302 recognizes and analyzes the first round of speech and determines whether the first semantics expressed by the first round of speech is a truncated semantics, it is specifically used to recognize the first round of speech and obtain text information; perform dependency syntactic analysis on the text information to determine whether the first semantics is a truncated semantics.

[0115] In some embodiments, when the analysis module 302 performs dependency syntactic analysis on the text information to determine whether the first semantics of the first round of speech expression is truncated semantics, it is specifically used to perform dependency syntactic analysis on the text information to determine the syntactic structure of the text information; if the syntactic structure does not contain a subject-verb structure, the first semantics is determined to be truncated semantics; if the syntactic structure contains a subject-verb structure, it is analyzed whether the intention expressed by the text information lacks relevant purpose parameters; if the relevant purpose parameters are missing, the first semantics is determined to be truncated semantics.

[0116] In some embodiments, when the clarification module 303 executes the clarification strategy to receive the second round of voice issued by the user, it is specifically used to obtain the user's historical behavior data and the current status of the in-vehicle device; predict the user's intention based on the first semantics, historical behavior data and the current status of the in-vehicle device to obtain a prediction result; select and execute the corresponding clarification strategy according to the prediction result to receive the second round of voice issued by the user.

[0117] In some embodiments, when the clarification module 303 selects and executes a corresponding clarification strategy according to the prediction result to receive a second-round voice from the user, it is specifically configured to select and execute a first clarification strategy if the prediction result is an executable instruction; the first clarification strategy includes sending a confirmation voice of the executable instruction to the user according to the executable instruction to guide the user to send a second-round voice in reply to the confirmation voice;

[0118] Correspondingly, when the processing module 304 recognizes the secondary round of speech and uses the second semantics expressed by the secondary round of speech to supplement the first semantics to obtain the instructions corresponding to the complete semantics, it is specifically used to recognize the secondary round of speech and determine whether the second semantics expressed by the secondary round of speech is represented as an executable instruction to execute; if so, the executable instruction is determined as the instruction corresponding to the complete semantics.

[0119] In some embodiments, when the clarification module 303 selects and executes a corresponding clarification strategy according to the prediction result to receive a second round of voice from the user, it is specifically configured to select and execute a second clarification strategy if the prediction result is an unexecutable instruction lacking a slot; the second clarification strategy includes sending a slot inquiry voice to the user according to the slot lacking the unexecutable instruction, so as to guide the user to send a second round of voice in response to the slot inquiry voice;

[0120] Correspondingly, when the processing module 304 recognizes the secondary round of speech and uses the second semantics expressed by the secondary round of speech to supplement the first semantics to obtain the instructions corresponding to the complete semantics, it is specifically used to recognize the secondary round of speech and determine whether the second semantics expressed by the secondary round of speech contains the target information for supplementing the slots; if so, the target information is used to supplement the missing slots of the non-executable instructions to form target executable instructions, and the target executable instructions are determined as the instructions corresponding to the complete semantics.

[0121] In some embodiments, when the clarification module 303 selects and executes a corresponding clarification strategy according to the prediction result to receive a second-round voice from the user, it is specifically configured to select and execute a third clarification strategy if the prediction result is that the user's intention is not predicted; the third clarification strategy includes sending an intentional questioning voice to the user to inquire about the intention, so as to guide the user to send a second-round voice to clarify the intention;

[0122] Correspondingly, when the processing module 304 recognizes the secondary round of speech and uses the second semantics expressed by the secondary round of speech to supplement the first semantics to obtain the instructions corresponding to the complete semantics, it is specifically used to recognize the secondary round of speech, splice the second semantics expressed by the secondary round of speech with the first semantics to obtain the spliced ​​semantics; and determine the instructions corresponding to the spliced ​​semantics as the instructions corresponding to the complete semantics.

[0123] Figure 3 The voice command recognition device of the illustrated embodiment can be used to execute the technical solution of the above-mentioned method embodiment, and its implementation principle and technical effect are similar and will not be repeated here.

[0124] Figure 4 The electronic device provided in the embodiment of the present disclosure may be a terminal as described in the above embodiment. The electronic device provided in the embodiment of the present disclosure may execute the processing flow provided in the embodiment of the voice command recognition method, such as Figure 4 As shown, the electronic device 400 includes: a memory 401, a processor 402, a computer program and a communication interface 403; wherein the computer program is stored in the memory 401 and is configured to be executed by the processor 402 as described above. The voice command recognition method. In a specific embodiment, the memory 401 can be a non-volatile solid-state memory. In a specific embodiment, the memory 401 includes a read-only memory ROM (Read-Only Memory). In appropriate cases, the ROM can be a mask-programmed ROM, a programmable ROM (Programmable ROM, PROM), an erasable PROM (Electrical Programmable ROM, EPROM), an electrically erasable PROM (Electrically Erasable Programmable ROM, EEPROM), an electrically alterable ROM (Electrically Alterable ROM, EAROM) or a flash memory, or a combination of two or more of these.

[0125] In addition, the embodiment of the present disclosure further provides a computer-readable storage medium on which a computer program is stored, and the computer program is executed by a processor to implement the voice command recognition method described in the above embodiment. The storage medium can be a non-volatile / non-temporary computer-readable storage medium, for example, the non-volatile / non-temporary computer-readable storage medium can be a ROM, a random access memory (Random Access Memory, RAM), a compact disc read-only memory (Compact Disc ROM, CD-ROM), a magnetic tape, a floppy disk, and an optical data storage device, etc.

[0126] The embodiments of the present disclosure also provide a vehicle, which includes the voice command recognition device and / or computer equipment as described in the above embodiments.

[0127] It should be noted that, in this article, relational terms such as "first" and "second" 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 term "comprises" or any other variant thereof is 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 existence of other identical elements in the process, method, article or device including the elements.

Claims

1. A method for voice command recognition, characterized in that: include: Receive the first round of voice from the user; Identifying and analyzing the first round of speech, and determining whether the first semantics expressed by the first round of speech is truncation semantics; If yes, a clarification strategy is executed to receive a second-round speech from the user; After receiving the second-round voice, the second-round voice is recognized, and the first semantics is supplemented by the second semantics expressed by the second-round voice to obtain instructions corresponding to the complete semantics.

2. The method according to claim 1, characterized in that The identifying and analyzing the first-round speech to determine whether the first semantics expressed by the first-round speech is a truncation semantics includes: Recognize the first round of speech to obtain text information; Perform dependency syntax analysis on the text information to determine whether the first semantics is truncation semantics.

3. The method according to claim 2, characterized in that The performing dependency syntactic analysis on the text information to determine whether the first semantics of the first round of speech expression is truncation semantics includes: Performing dependency syntactic analysis on the text information to determine the syntactic structure of the text information; If the syntactic structure does not include a subject-verb structure, determining that the first semantics is a truncation semantics; If the syntactic structure includes a subject-verb structure, analyzing whether the intention expressed by the text information lacks relevant purpose parameters; If the relevant destination parameter is missing, it is determined that the first semantics is truncation semantics.

4. The method according to claim 1, characterized in that The executing of the clarification strategy to receive a second-round speech from the user includes: Obtain the user's historical behavior data and the current status of the in-vehicle equipment; Predicting the user's intention based on the first semantics, the historical behavior data, and the current state of the in-vehicle device to obtain a prediction result; A corresponding clarification strategy is selected and executed according to the prediction result to receive a second round of speech from the user.

5. The method according to claim 4, characterized in that The selecting and executing a corresponding clarification strategy according to the prediction result to receive a second round of speech from the user includes: If the prediction result is an executable instruction, selecting and executing a first clarification strategy; The first clarification strategy includes sending a confirmation voice of the executable instruction to the user according to the executable instruction, so as to guide the user to send a second round of voice in response to the confirmation voice; The step of recognizing the second-round speech and supplementing the first semantics with the second semantics expressed by the second-round speech to obtain the instruction corresponding to the complete semantics includes: Recognize the second-round speech, and determine whether the second semantics expressed by the second-round speech is characterized by executing the executable instruction; If so, the executable instruction is determined to be an instruction corresponding to the complete semantics.

6. The method according to claim 4, characterized in that The selecting and executing a corresponding clarification strategy according to the prediction result to receive a second round of speech from the user includes: If the prediction result is an unexecutable instruction lacking a slot, selecting and executing a second clarification strategy; The second clarification strategy includes sending a slot inquiry voice to the user according to the slot missing from the unexecutable instruction, so as to guide the user to send a second round of voice in response to the slot inquiry voice; The step of recognizing the second-round speech and supplementing the first semantics with the second semantics expressed by the second-round speech to obtain the instruction corresponding to the complete semantics includes: Recognize the second-round speech, and determine whether the second semantics expressed by the second-round speech includes target information to supplement the slot; If so, the target information is used to supplement the missing slots of the non-executable instructions to form target executable instructions, and the target executable instructions are determined as instructions corresponding to complete semantics.

7. The method according to claim 4, characterized in that The selecting and executing a corresponding clarification strategy according to the prediction result to receive a second round of speech from the user includes: If the prediction result is that the user's intention is not predicted, selecting and executing a third clarification strategy; The third clarification strategy includes sending a follow-up voice to the user to guide the user to send a second round of voice to clarify the intention, or silently waiting for the user to send a second round of voice to clarify the intention; The step of recognizing the second-round speech and supplementing the first semantics with the second semantics expressed by the second-round speech to obtain the instruction corresponding to the complete semantics includes: Recognizing the second-round speech, and concatenating the second semantics expressed by the second-round speech with the first semantics to obtain concatenated semantics; The spliced ​​semantically corresponding instructions are determined as complete semantically corresponding instructions.

8. A voice command recognition device, characterized in that: include: A receiving module, used to receive the first round of voice sent by the user; An analysis module, used for identifying and analyzing the first-round speech, and determining whether the first semantics expressed by the first-round speech is a truncation semantics; A clarification module, used for executing a clarification strategy to receive a second-round speech from a user; The processing module is used to recognize the second-round voice after receiving the second-round voice, and use the second semantics expressed by the second-round voice to supplement the first semantics to obtain instructions corresponding to the complete semantics.

9. A computer device comprising: Memory; processor; as well as Computer programs; The computer program is stored in the memory and is configured to be executed by the processor to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

11. A vehicle, characterized in that: It comprises the voice instruction recognition device as claimed in claim 8 and / or the computer device as claimed in claim 9.

Images