Voice instruction determination method, electronic equipment and storage medium

Abstract

The invention provides a voice instruction determination method, electronic equipment and a storage medium, and the method comprises the steps: obtaining a current recognition result corresponding to a to-be-processed voice sent by a user, determining whether the current recognition result is an intermediate result, carrying out the text supplement of the intermediate result in response to the intermediate result, and obtaining a corresponding prediction text, judging whether a semantic understanding task corresponding to the prediction text exists or not, if not, obtaining a semantic instruction based on the generated semantic understanding task corresponding to the prediction text, then returning to execute the step of obtaining the current recognition result and determining whether the current recognition result is an intermediate result or not, and responding to the current recognition result as a final result. Whether the semantic understanding task corresponding to the prediction text which is the same as the final result exists or not is judged, and if yes, the semantic instruction corresponding to the prediction text serves as a final instruction, so that the starting frequency of a semantic understanding module is greatly reduced; the problems that a large number of intermediate results are subjected to semantic understanding, so that the computing power is soaring, and the vehicle command response is slow are solved.

Description

Technical Field

[0001] The present application relates to the field of vehicle computer technology, and in particular to a voice command determination method, an electronic device and a storage medium. Background Art

[0002] At present, the execution link process of car computer voice is as follows: the recorded data is converted into text through the recognition engine, the text content is passed to the semantic understanding module to be converted into various instructions, and then handed over to the specific business module to execute the instruction.

[0003] When the user is speaking, the recognition engine will generate a large number of intermediate results and a final result. The existing technologies are generally divided into two types: 1. Do not process the intermediate results of speech recognition, wait for the final result of the recognition engine to be returned, and the semantic understanding module only processes the final result; 2. Use the intermediate results as preprocessing data for the semantic understanding module, perform semantic understanding on each intermediate result, and match the final result with the preprocessed intermediate results after the final result is returned.

[0004] However, the above scheme has the following defects: 1. Since it is necessary to wait for the final result to be returned before semantic understanding can be performed, the final return time of the instruction is prolonged, which in turn leads to slow instruction execution; 2. The large number of pre-processed intermediate results will cause the semantic understanding module to perform a large number of semantic understanding tasks, which will cause a surge in computing power. Summary of the invention

[0005] In view of the above-mentioned defects or deficiencies in the prior art, the present application aims to provide a voice command determination method, an electronic device and a storage medium to solve the problem of slow instruction execution caused by waiting for the semantic understanding of the final result to be completed in the prior art, as well as to solve the problem of soaring computing power caused by semantic understanding of a large number of intermediate results.

[0006] The present invention provides a method for determining a voice command, the method comprising:

[0007] Obtaining a current recognition result corresponding to the speech to be processed issued by the user, and determining whether the current recognition result is an intermediate result;

[0008] In response to the current recognition result being an intermediate result, supplementing the intermediate result with text to obtain a predicted text corresponding to the intermediate result;

[0009] Determine whether there is a semantic understanding task corresponding to the predicted text, and if not, generate the semantic understanding task corresponding to the predicted text to obtain the semantic instruction corresponding to the predicted text;

[0010] Returning to the step of obtaining the current recognition result corresponding to the to-be-processed speech emitted by the user, and determining whether the current recognition result is an intermediate result;

[0011] In response to the current recognition result being the final result, it is determined whether there is a semantic understanding task corresponding to the same predicted text as the final result, and if so, the semantic instruction corresponding to the predicted text is used as the final instruction.

[0012] Optionally, text supplementing is performed on the intermediate result to obtain a predicted text corresponding to the intermediate result, including:

[0013] Matching the intermediate result with each common sentence text in the user's common sentence set;

[0014] The predicted text corresponding to the intermediate result is determined based on the matching results of each common sentence text.

[0015] Optionally, matching the intermediate result with each common sentence text in the common sentence set of the user includes:

[0016] Determining a first similarity between the intermediate result and each common sentence text;

[0017] Determining the predicted text corresponding to the intermediate result based on the matching results of each common sentence text includes:

[0018] Determine the commonly used sentence texts whose corresponding first similarity is greater than a preset first similarity threshold as candidate texts;

[0019] If there are multiple candidate texts, the predicted text is determined from each candidate text based on the number of times each candidate text is used, the most recent time of use, and the corresponding first similarity.

[0020] Optionally, the method further includes:

[0021] If the matching of the predicted text from the common sentence set fails, matching the intermediate result with each general instruction text in the general operation instruction set;

[0022] The predicted text corresponding to the intermediate result is determined based on the matching results of each general instruction text.

[0023] Optionally, the method further includes:

[0024] If the matching of the predicted text from the general operation instruction set fails, a semantic understanding task corresponding to the intermediate result is generated to obtain a semantic instruction corresponding to the intermediate result.

[0025] Optionally, before generating the semantic understanding task corresponding to the intermediate result and obtaining the semantic instruction corresponding to the semantic understanding task, the method further includes:

[0026] Determining the number of words in the text of the intermediate result;

[0027] Determining whether the intermediate result meets a preset semantic understanding entry condition based on the number of words in the text of the intermediate result;

[0028] If so, execute the step of generating the semantic understanding task corresponding to the intermediate result and obtaining the semantic instruction corresponding to the intermediate result.

[0029] Optionally, the method further includes:

[0030] In response to the current recognition result being a final result, determining whether there is a semantic understanding task corresponding to an intermediate result having the same number of characters as the final result;

[0031] If so, the semantic instruction corresponding to the intermediate result is used as the final instruction.

[0032] Optionally, after determining whether there is a semantic understanding task corresponding to the predicted text that is the same as the final result, the method further includes:

[0033] If there is no semantic understanding task corresponding to the predicted text identical to the final result, generating a semantic understanding task corresponding to the final result, and obtaining a semantic instruction corresponding to the final result;

[0034] The semantic instruction corresponding to the final result is used as the final instruction.

[0035] Optionally, the method further includes:

[0036] Executing the final instruction, and after the execution of the final instruction is completed, using the final instruction as the common sentence to be added;

[0037] The common sentence set is updated based on the common sentences to be added.

[0038] Optionally, updating the common sentence set based on the common sentence to be added includes:

[0039] Determining whether the commonly used sentence to be added exists in the commonly used sentence set;

[0040] If so, the usage count and the most recent usage time of the to-be-added common sentence in the common sentence set are updated.

[0041] The embodiment of the present application also provides a device for determining a command of a vehicle machine voice, the device comprising:

[0042] A recognition module, used to obtain a current recognition result corresponding to the to-be-processed speech emitted by the user, and determine whether the current recognition result is an intermediate result;

[0043] A supplementing module, configured to supplement the intermediate result with text in response to the current recognition result being an intermediate result, so as to obtain a predicted text corresponding to the intermediate result;

[0044] A judgment module, used to judge whether there is a semantic understanding task corresponding to the predicted text, and if not, generate the semantic understanding task corresponding to the predicted text to obtain the semantic instruction corresponding to the predicted text;

[0045] A return module, used to return the current recognition result corresponding to the speech to be processed issued by the user, and determine whether the current recognition result is an intermediate result;

[0046] The final module is used to respond to the current recognition result being the final result, determine whether there is a semantic understanding task corresponding to the predicted text that is the same as the final result, and if so, use the semantic instruction corresponding to the predicted text as the final instruction.

[0047] An embodiment of the present application further provides an electronic device, the electronic device comprising:

[0048] Processor and memory;

[0049] The processor is used to execute the steps of the voice instruction determination method provided in any embodiment of the present application by calling the program or instruction stored in the memory.

[0050] An embodiment of the present application also provides a computer-readable storage medium, which stores a program or instruction, and the program or instruction enables a computer to execute the steps of the voice instruction determination method provided in any embodiment of the present application.

[0051] In summary, the present application proposes a method for determining a voice instruction, which obtains a current recognition result corresponding to a to-be-processed voice issued by a user, determines whether it is an intermediate result, and in response to the current recognition result being an intermediate result, performs text supplementation on the intermediate result to obtain the corresponding predicted text, and determines whether there is a semantic understanding task corresponding to the predicted text. If not, a semantic understanding task corresponding to the predicted text is generated to obtain a semantic instruction corresponding to the predicted text, and then returns to execute the steps of obtaining the current recognition result and determining whether it is an intermediate result. In response to the current recognition result being a final result, it is determined whether there is a semantic understanding task corresponding to the predicted text that is the same as the final result. If so, the semantic instruction corresponding to the predicted text is used as the final instruction. The method can predict the text of the voice that the user may subsequently initiate by performing text supplementation on each intermediate result, and through Determine whether there is a semantic understanding task corresponding to the supplemented predicted text, and perform semantic understanding on the predicted text if not, so as to avoid semantic understanding of all intermediate results, greatly reduce the number of startup times of the semantic understanding module, solve the problem of computing power surge caused by semantic understanding of a large number of intermediate results, and solve the problem of excessive computing power usage during voice command recognition, thereby avoiding vehicle computer freezes caused by voice parsing. Moreover, after obtaining the final result, the method determines whether there is a semantic understanding task corresponding to the predicted text consistent with it. If so, directly use the semantic instruction corresponding to the predicted text as the final instruction, and there is no need to perform semantic understanding on the final result. This solves the problem of long final return of the instruction due to waiting for the final result to be returned and performing semantic understanding, thereby solving the problem of slow vehicle computer instruction response. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the specific implementation methods of the present application or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0053] Figure 1 is a flow chart of a method for determining a voice instruction provided in an embodiment of the present application;

[0054] Figure 2 This is a schematic diagram of a semantic understanding process of vehicle machine speech provided by an embodiment of the present application;

[0055] Figure 3 It is a structural schematic diagram of a vehicle machine voice command determination device provided in an embodiment of the present application;

[0056] Figure 4It is a structural schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0057] The present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the relevant invention, rather than to limit the invention. It is also necessary to explain that, for ease of description, only the parts related to the invention are shown in the accompanying drawings.

[0058] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0059] As mentioned in the background technology, in response to the problems in the prior art, this application proposes a method for determining a voice command. Figure 1 is a flow chart of a method for determining a voice command provided by an embodiment of the present application. Figure 1 , the voice instruction determination method specifically includes:

[0060] S110, obtaining a current recognition result corresponding to the to-be-processed speech emitted by the user, and determining whether the current recognition result is an intermediate result.

[0061] In an embodiment of the present application, the user can initiate a pending voice through a voice assistant. For example, the user can wake up the voice assistant through voice, triggering controls on the vehicle interface or buttons on the steering wheel. After the voice assistant is awakened, the user can send a pending voice to control the vehicle through the pending voice, such as opening the passenger window, turning on the seat heating function, turning off the air conditioning external circulation, etc., or control the vehicle through the pending voice, such as playing music, lowering the volume, turning on navigation, etc.

[0062] Specifically, when the user sends out the speech to be processed, the speech data can be collected in real time through the recording device, and then the speech data can be subjected to speech recognition in real time, such as ASR (Automatic Speech Recognition), so as to obtain the current recognition result corresponding to the speech to be processed.

[0063] The current recognition result can be understood as the result obtained by real-time recognition of voice data in the process of real-time acquisition of the voice to be processed, and specifically can be the text recognized by the voice data. For example, when the user speaks the sentence "open the car window", the voice recognition is performed in real time, and the current recognition results obtained are: "open", "open", "open the car", and "open the car window".

[0064] In the embodiment of the present application, the recognition engine can perform speech recognition on the speech data acquired in real time to obtain the current recognition result. Further, the recognition engine can determine whether the current recognition result is an intermediate result or a final result.

[0065] Among them, the intermediate result can be the result of voice recognition before the end of the user's voice initiation. For example, the user speaks the sentence "open the car window", and the intermediate results may include: "hit", "open", "open the car", "open the car window". The final result can be the result of voice recognition after the end of the user's voice initiation. For example, the user speaks the sentence "open the car window", and after waiting for a period of time to confirm that the user does not continue to speak, voice recognition can be performed to obtain the final result: "open the car window".

[0066] S120 . In response to the current recognition result being an intermediate result, text supplement is performed on the intermediate result to obtain a predicted text corresponding to the intermediate result.

[0067] Specifically, if the current recognition result is determined to be an intermediate result, it means that the user has not completed the voice initiation at this time, that is, the user has not finished saying this sentence. Therefore, the intermediate result can be supplemented with text to predict the text that the user expects to say later, that is, to predict the complete voice content initiated by the user.

[0068] Among them, the text supplement of the intermediate result can be to input the intermediate result into the text prediction model to obtain the predicted text output by the text prediction model; or, it can be combined with common sentences or general operation instructions, matched with the intermediate result, and perform text supplement based on the matched content.

[0069] In a specific implementation, text supplementation is performed on the intermediate result to obtain a predicted text corresponding to the intermediate result, including the following steps:

[0070] Step 11: Match the intermediate result with the text of each common sentence in the user's common sentence set;

[0071] Step 12: Determine the predicted text corresponding to the intermediate result based on the matching results of each common sentence text.

[0072] The common user set may be composed of common sentence texts corresponding to each common sentence of the user. For example, the user may be guided to initiate each common sentence through the vehicle interface in advance, and then the corresponding common sentence set may be constructed; or the corresponding common sentence set may be constructed according to the instruction triggered by the user on the vehicle interface.

[0073] In an embodiment of the present application, a commonly used sentence set can be set for the same vehicle; alternatively, corresponding commonly used sentence sets can be set for different users, respectively, and the user can be identified by facial image or login information, thereby using the commonly used sentence set of the identified user to improve the accuracy of predicting subsequent text of the current recognition result.

[0074] Specifically, in step 11, the text vector corresponding to each common sentence text can be determined, and the text vector corresponding to the intermediate result can be determined, and then the similarity between the text vectors is calculated as the first similarity between the common sentence text and the intermediate result, so as to achieve matching between the intermediate result and the common sentence text.

[0075] Further, in step 12, the predicted text corresponding to the intermediate result is determined based on the matching results of each common sentence text, which can be: according to the first similarity corresponding to each common sentence text, the predicted text is selected from each common sentence text. For example, the common sentence text whose first similarity is greater than a preset first similarity threshold is selected as the predicted text.

[0076] Through the above steps 11 and 12, the text of the intermediate result can be supplemented through the user's common sentences to predict the subsequent text of the intermediate result, so that the predicted text can be as consistent as possible with the user's common sentences to meet the user's usage needs.

[0077] It should be noted that, considering that there may be a situation where the first similarities corresponding to multiple common sentence texts are greater than the preset first similarity threshold, for such a situation, the predicted text can be determined in combination with the number of times the common sentence text is used, the most recent use time, etc.

[0078] Optionally, matching the intermediate result with each common sentence text in the user's common sentence set includes: determining a first similarity between the intermediate result and each common sentence text;

[0079] Determining the predicted text corresponding to the intermediate result based on the matching results of each common sentence text includes the following steps:

[0080] Step 121, determining the commonly used sentence texts whose corresponding first similarity is greater than a preset first similarity threshold as candidate texts;

[0081] Step 122: If there are multiple candidate texts, a predicted text is determined from each candidate text based on the number of times each candidate text is used, the most recent time of use, and the corresponding first similarity.

[0082] Specifically, the intermediate result and the commonly used sentence texts may be matched by calculating the first similarity between the intermediate result and each commonly used sentence text. If a large number of matched texts are obtained, the number of times the text is used, the most recent time of use, etc. may be further considered.

[0083] In step 121, some common sentence texts may be first screened out as candidate texts based on the first similarity, for example, common sentence texts with a first similarity greater than a preset first similarity threshold may be selected as candidate texts.

[0084] Furthermore, if there are multiple candidate texts, such as, the intermediate result is "open", the commonly used sentence texts whose first similarity is greater than the preset first similarity threshold include: "open the window", "open settings", "turn on high beam", "turn on navigation"; at this time, the predicted text can be determined based on the number of times each candidate text is used, the time of most recent use and the corresponding first similarity.

[0085] The number of times used may be the total number of times the text is executed as a command, or may be the number of times the text is executed as a command within a set time period (such as two weeks). The most recent time of use may be the most recent time when the text is executed as a command.

[0086] Exemplarily, corresponding weights can be set in advance for the number of times used, the time of most recent use, and the first similarity, respectively. Then, the number of times used, the time of most recent use, and the first similarity are weighted in combination with the corresponding weights to obtain a weighted result for describing the prediction accuracy of the candidate text, and the predicted text is determined among the candidate texts based on the weighted result.

[0087] Through the above steps 121 and 122, the predicted text is determined in combination with the number of times used and the most recent time of use, which can further ensure that the predicted text better meets the user's needs.

[0088] In addition, it is considered that there may be a situation where text supplementation of the intermediate result based on the common sentence set fails, that is, the matching of the predicted text from the common sentence set fails. Exemplarily, the failure to match the predicted text from the common sentence set may be: the number of candidate texts is zero (that is, the first similarity corresponding to each common sentence text is less than the preset first similarity threshold), or the number of candidate texts is single and the candidate text is consistent with the intermediate result. For such situations, the subsequent text of the intermediate result can be further predicted in combination with the general operation instruction set. Optionally, the method provided in the embodiment of the present application also includes the following steps:

[0089] Step 123: if the predicted text is not matched in the common sentence set, the intermediate result is matched with each general instruction text in the general operation instruction set;

[0090] Step 124: Determine the predicted text corresponding to the intermediate result based on the matching results of each general instruction text.

[0091] Among them, the matching prediction text from the common sentence set fails, specifically, it can be: the number of candidate texts is zero, the number of candidate texts is single and the candidate text is consistent with the intermediate result. The number of candidate texts is zero, which means that there is no common sentence text that is similar to the intermediate result; the number of candidate texts is single and the candidate text is consistent with the intermediate result, which means that the common sentence text matched based on the common sentence set is the same as the intermediate result, for example, the candidate text and the current recognition result are both: "Open the window", and the subsequent text of the intermediate result cannot be predicted.

[0092] Specifically, if the number of candidate texts is zero, or the number of candidate texts is single and the candidate text is consistent with the intermediate result, it can be determined that the subsequent text of the intermediate result based on the common sentence set has failed to be predicted at this time, and the subsequent text of the intermediate result can be further predicted based on the general operation instruction set.

[0093] The general operation instruction set is composed of general instruction texts corresponding to each cockpit general operation instruction. For example, a large number of user's vehicle machine voice data can be collected in advance, and the cockpit general operation instructions can be analyzed based on the vehicle machine voice data to construct the general operation instruction set.

[0094] Specifically, in step 123, the text vector of each general instruction text in the general operation instruction set and the text vector of the intermediate result can be determined, and then the similarity between the text vectors is calculated as the second similarity between the general instruction text and the intermediate result, so as to achieve matching between the general instruction text and the intermediate result.

[0095] Furthermore, in step 124, the predicted text corresponding to the intermediate result is determined based on the matching results of each general instruction text, which may be: determining the predicted text by the second similarity corresponding to each general instruction text. For example, a general instruction text having a second similarity greater than a preset second similarity threshold is screened from all general instruction texts, and if the number of screened general instruction texts is one, the general instruction text may be directly used as the predicted text.

[0096] In addition, if there are multiple general command texts selected, one of the general command texts can be randomly selected as the predicted text, or the general command text can be selected as the predicted text in combination with the status of each component of the vehicle. For example, the general command texts selected include: "Turn on the air conditioner" and "Turn on the high beam". At this time, it can be judged whether the vehicle air conditioner and high beam are turned on. If the air conditioner is not turned on, "Turn on the air conditioner" can be used as the predicted text; or the general command texts selected include: "Lower the playback volume" and "Lower the window opening". At this time, it can be judged whether the vehicle computer is playing audio and whether the window is open. If the window is not open, or the vehicle computer is playing music, navigation and other audio, "Lower the playback volume" can be used as the predicted text.

[0097] Through the above steps 123 and 124, when the subsequent text of the intermediate result fails to be predicted based on the common sentence set, the subsequent text of the intermediate result can be further predicted in combination with the general operation instruction set, which can avoid the prediction failure and enter the semantic understanding as much as possible, and improve the efficiency of vehicle voice command recognition.

[0098] Considering that there may be a situation where the predicted text cannot be matched from the general operation instruction set, for example, the second similarities corresponding to all general instruction texts in the general operation instruction set are less than the preset second similarity threshold, for such a situation, optionally, the method provided in the embodiment of the present application further includes the following steps:

[0099] Step 125: If the matching of the predicted text from the general operation instruction set fails, a semantic understanding task corresponding to the intermediate result is generated to obtain a semantic instruction corresponding to the intermediate result.

[0100] Specifically, if the second similarities corresponding to all general instruction texts are less than the preset second similarity threshold, it can be determined that the predicted text matching from the common sentence set and the general operation instruction set fails, and the predicted text cannot be obtained, then the semantic understanding task corresponding to the intermediate result can be directly generated to obtain the corresponding semantic instruction. Among them, semantic understanding can be to extract the actions and action subjects that need to be executed in the text through structured processing, and construct instructions.

[0101] When prediction based on both the commonly used sentence set and the general operation instruction set fails, that is, no predicted text is obtained, by performing semantic understanding on the intermediate results and obtaining the semantic instructions corresponding to the intermediate results, real-time semantic understanding can be achieved when the user initiates a completely new voice that is different from the commonly used sentences and the general cockpit instructions, thereby ensuring control response to the user's new voice.

[0102] Taking into account the possibility that the user's new voice may have too many words, in order to avoid semantic understanding of each intermediate result, word skipping processing can also be performed, that is, semantic understanding is only performed on intermediate results of a set number of words.

[0103] With respect to the above step 125, in one example, before generating the semantic understanding task corresponding to the intermediate result and obtaining the semantic instruction corresponding to the semantic understanding task, the following steps are also included:

[0104] Step 1251, determining the number of words in the text of the intermediate result;

[0105] Step 1252: judging whether the intermediate result meets the preset semantic understanding entry condition based on the number of words in the text of the intermediate result;

[0106] Step 1253: If yes, execute the semantic understanding task corresponding to the intermediate result, and obtain the semantic instruction corresponding to the intermediate result.

[0107] The preset semantic understanding entry condition may be that the number of characters in the text reaches a preset number of characters, such as the number of characters in the text reaches 4 characters, 6 characters, 8 characters, etc.

[0108] In the embodiment of the present application, the number of preset word counts can be multiple, and each preset word count can be separated by a certain word skipping step length, and the word skipping step length can be a fixed value, or the corresponding word skipping step length can be set for different usage scenarios, and the current usage scenario is determined according to the current recognition result, and the corresponding word skipping step length is obtained, thereby determining each preset word count. For example, if the current recognition result is: "I want to listen to the album of singer A", then the current usage scenario can be listening to music. Considering that the song title generally has more words, the word skipping step length corresponding to listening to music can be 3.

[0109] Specifically, the number of characters in the text of the intermediate result can be determined first. If the number of characters in the text is the same as the preset number of characters, it can be determined that the intermediate result meets the preset semantic understanding entry condition. Further, the above-mentioned step of generating the semantic understanding task corresponding to the intermediate result and obtaining the semantic instruction corresponding to the intermediate result can be performed, that is, the intermediate result enters the semantic understanding.

[0110] Of course, if the number of characters in the text of the intermediate result is different from all the preset numbers of characters, the process can directly return to S110, that is, continue to obtain the current recognition result corresponding to the to-be-processed speech emitted by the user, and then continue to judge the next recognition result.

[0111] For example, the complete speech uttered by the user is: "I want to listen to AAA's album BBB's CC". Through the above steps, semantic understanding can be performed only on the intermediate results "I want to listen", "I want to listen to AAA", "I want to listen to AAA's album", "I want to listen to AAA's album BBB", and "I want to listen to AAA's album BBB's CC" that are not matched to the predicted text.

[0112] Through the above steps 1251 to 1253, when the predicted text corresponding to the intermediate results is not matched, semantic understanding can be performed only on part of the intermediate results through word skipping processing, thereby avoiding processing all the intermediate results for the new voice uttered by the user, thereby reducing computing power and improving the efficiency of voice response.

[0113] S130, determining whether there is a semantic understanding task corresponding to the predicted text; if not, generating a semantic understanding task corresponding to the predicted text to obtain a semantic instruction corresponding to the predicted text.

[0114] Specifically, if the predicted text corresponding to the current recognition result can be determined, further, it can be determined whether there is a semantic understanding task corresponding to the predicted text.

[0115] Among them, if there is a semantic understanding task corresponding to the predicted text, it means that the predicted text has entered semantic understanding. For example, the predicted text is the same as the predicted text corresponding to the previous intermediate result, and the same predicted text has undergone semantic understanding. If there is no semantic understanding task corresponding to the predicted text, it means that the predicted text has not entered semantic understanding. For example, the predicted text is the predicted text corresponding to the first intermediate result, or the predicted text is different from the predicted text corresponding to the previous intermediate result.

[0116] Exemplarily, for the voice to be processed, if the intermediate result is the first intermediate result, at this time, none of the intermediate results have entered semantic understanding, and there is no semantic understanding task corresponding to the predicted text. At this time, a semantic understanding task corresponding to the predicted text can be directly generated to obtain the semantic instruction corresponding to the predicted text.

[0117] If the intermediate result is not the first intermediate result, then at this time, the predicted text corresponding to the intermediate result may or may not have entered semantic understanding. It is necessary to determine whether there is a semantic understanding task corresponding to the predicted text. If not, it means that the predicted text is different from the predicted text corresponding to the previous intermediate result. At this time, a semantic understanding task corresponding to the predicted text can be directly generated to obtain the semantic instruction corresponding to the predicted text. If so, it means that the predicted text has been predicted in the previous intermediate result and has undergone semantic understanding. Therefore, at this time, S140 can be directly executed to return to S110 to continue obtaining a new current recognition result, and then determine whether it is an intermediate result.

[0118] Exemplarily, before the user finishes speaking the whole sentence, the predicted text corresponding to the first intermediate result "hit" is "open the window", then a corresponding semantic understanding task is generated to obtain the semantic instruction; the predicted text corresponding to the second intermediate result "open" is "open the window". At this time, it can be determined that there is a semantic understanding task corresponding to the predicted text, and S140 can be directly executed to return to S110 to continue obtaining the third recognition result.

[0119] For another example, the predicted text corresponding to the third intermediate result "open set" is "open settings", and there is no semantic understanding task corresponding to "open settings". At this time, a corresponding semantic understanding task can be generated to obtain the semantic instruction.

[0120] S140: Return to the step of obtaining the current recognition result corresponding to the voice to be processed issued by the user and determining whether the current recognition result is an intermediate result.

[0121] Specifically, if there is no semantic understanding task corresponding to the predicted text, a semantic understanding task corresponding to the predicted text can be generated, and the semantic instruction corresponding to the predicted text can be obtained, and then the current recognition result can be returned to continue to be obtained, and it can be determined whether the current recognition result is an intermediate result, so as to continue to judge whether the next recognition result is an intermediate result. If so, the text is supplemented, and it is determined whether there is a semantic understanding task corresponding to the supplemented predicted text.

[0122] S150: In response to the current recognition result being the final result, determine whether there is a semantic understanding task corresponding to the predicted text that is the same as the final result; if so, take the semantic instruction corresponding to the predicted text as the final instruction.

[0123] In the process of returning to S110, if it is determined that the current recognition result is not an intermediate result, that is, the current recognition result is the final result, it can be determined whether there is a semantic understanding task corresponding to the predicted text that is the same as the final result. If so, the semantic instruction corresponding to the predicted text is used as the final instruction, avoiding semantic understanding of the final result and improving the response efficiency of the vehicle voice.

[0124] In an embodiment of the present application, if the predicted text is not successfully matched due to the large number of words in the intermediate result, after semantic understanding is performed on it, it can be directly determined based on the number of words whether semantic understanding of the final result is required.

[0125] In one example, the method provided in the embodiment of the present application further includes:

[0126] In response to the current recognition result being the final result, determine whether there is a semantic understanding task corresponding to an intermediate result having the same number of words as the final result; if so, take the semantic instruction corresponding to the intermediate result as the final instruction.

[0127] Among them, if the prediction text matching for the intermediate result based on the common sentence set and the general operation instruction set fails, the intermediate result can directly enter the semantic understanding, and it can also be judged whether it needs to enter the semantic understanding based on the number of words in the text.

[0128] Furthermore, for the final result, it is possible to directly determine whether there is a semantic understanding task corresponding to an intermediate result with the same number of words as the final result. If so, it means that an intermediate result with the same number of words has been semantically understood, and the text content of the intermediate result with the same number of words is the same as that of the final result. Therefore, there is no need to perform semantic understanding on the final result. At this time, the semantic instruction corresponding to the intermediate result can be directly used as the final instruction.

[0129] It is understandable that if there is no semantic understanding task corresponding to the intermediate result having the same number of words as the final result, a semantic understanding task corresponding to the final result can be generated to obtain a semantic instruction corresponding to the final result.

[0130] Through the above examples, considering that the predicted text is not successfully matched usually because the intermediate result has a large number of words, when the intermediate result does not successfully match the predicted text, it can be determined whether there is a semantic understanding task corresponding to the intermediate result with the same number of words as the final result. If so, it means that the intermediate result with the same text content as the final result has been semantically understood. At this time, there is no need to perform semantic understanding on the final result. The semantic instructions of the intermediate result can be directly called to ensure the execution efficiency of the instructions. In addition, since the intermediate result directly entered into the semantic understanding does not successfully obtain the text supplement, if it has the same number of words as the final result, it can be considered that the text contents of the two are the same, and there is no need to compare the texts, which can further improve the execution efficiency of the instructions.

[0131] In a specific implementation, after determining whether there is a semantic understanding task corresponding to the predicted text that is the same as the final result, the method further includes:

[0132] If there is no semantic understanding task corresponding to the predicted text that is the same as the final result, a semantic understanding task corresponding to the final result is generated to obtain a semantic instruction corresponding to the final result; and the semantic instruction corresponding to the final result is used as the final instruction.

[0133] Specifically, if there is no semantic understanding task corresponding to the predicted text that is the same as the final result, it means that the prediction of the intermediate result may deviate from the final result, and the semantic instructions generated by the predicted text cannot be used as the final instructions. At this time, the final result can be input into the semantic understanding module to perform semantic understanding on the final result and obtain the semantic instructions corresponding to the final result.

[0134] Furthermore, the semantic instruction corresponding to the final result can be used as a final instruction and sent to a corresponding control center to execute the final instruction, for example, a vehicle control center, a media control center, a navigation control center, etc.

[0135] Through the above implementation, when there is no semantic understanding task corresponding to the same predicted text as the final result, the final result can be semantically understood, thereby avoiding the situation where a response error occurs due to directly using a semantic instruction corresponding to a predicted text.

[0136] In order to improve the accuracy of predicting the subsequent text of the intermediate result, the common sentence set can also be updated according to each final instruction executed. Optionally, the method provided in the embodiment of the present application also includes the following steps:

[0137] Step 21, execute the final instruction, and after the final instruction is executed, use the final instruction as the common sentence to be added;

[0138] Step 22: Update the common sentence set based on the common sentences to be added.

[0139] Specifically, after the final instruction is determined and executed, the final instruction may be used as the common sentence to be added.

[0140] Furthermore, it can be determined whether there is a common sentence to be added in the common sentence set. If there is no common sentence to be added in the common sentence set, the text corresponding to the common sentence to be added can be saved as the common sentence text in the common sentence set, and the number of times the common sentence to be added is used and the most recent use time can be set.

[0141] In one example, updating the common sentence set based on the common sentences to be added includes:

[0142] It is determined whether there is a common sentence to be added in the common sentence set; if so, the number of times the common sentence to be added in the common sentence set is used and the most recent time of use are updated.

[0143] Specifically, if there are common sentences to be added in the common sentence set, the usage count and the most recent usage time of the common sentence to be added in the common sentence set can be directly updated, that is, the usage count is increased by one, and the most recent usage time can be updated to the time when the final result is recognized or the time when the final instruction is executed.

[0144] Through the above implementation, after the final instruction is determined and executed, the common sentence set can be updated to improve the accuracy of subsequent text supplementation of the intermediate results, so that the supplemented text can better meet the user's control needs.

[0145] The voice instruction determination method provided in the embodiment of the present application obtains the current recognition result corresponding to the to-be-processed voice issued by the user, determines whether it is an intermediate result, and in response to the current recognition result being an intermediate result, performs text supplement on the intermediate result to obtain the corresponding predicted text, and determines whether there is a semantic understanding task corresponding to the predicted text. If not, generates a semantic understanding task corresponding to the predicted text, obtains the semantic instruction corresponding to the predicted text, and then returns to execute the steps of obtaining the current recognition result and determining whether it is an intermediate result. In response to the current recognition result being a final result, determines whether there is a semantic understanding task corresponding to the predicted text that is the same as the final result. If so, uses the semantic instruction corresponding to the predicted text as the final instruction. The method can predict the text of the voice that the user may subsequently initiate by performing text supplement on each intermediate result, and determines whether it is Whether there is a semantic understanding task corresponding to the supplemented predicted text, if not, semantic understanding is performed on the predicted text, which can avoid semantic understanding of all intermediate results, greatly reducing the number of startup times of the semantic understanding module, solving the problem of computing power surge caused by semantic understanding of a large number of intermediate results, and solving the problem of excessive computing power usage during voice command recognition, thereby avoiding vehicle computer freezes caused by voice parsing. In addition, after obtaining the final result, the method determines whether there is a semantic understanding task corresponding to the predicted text that is consistent with it. If so, the semantic instruction corresponding to the predicted text is directly used as the final instruction, and there is no need to perform semantic understanding on the final result. This solves the problem of long final return of the instruction due to waiting for the final result to return and performing semantic understanding, thereby solving the problem of slow response of vehicle computer instructions.

[0146] Figure 2 is a schematic diagram of a semantic understanding process of a vehicle machine voice provided by an embodiment of the present application, such as Figure 2 As shown, the current recognition result is taken as an example to illustrate the processing flow of the current recognition result. First, the current recognition result can be obtained through ASR recognition, and it is determined whether it is an intermediate result, that is, whether it is an intermediate result or a final result.

[0147] If it is an intermediate result, the common sentence text is matched for the current recognition result through the common sentence set. If it is matched, it is directly used as the predicted text. If it is not matched, the general instruction text is matched for the current recognition result based on the general operation instruction set. If it is matched, it is directly used as the predicted text. Furthermore, it can be determined whether there is a corresponding semantic understanding task. If not, semantic understanding is performed to obtain the corresponding semantic instruction.

[0148] If the general instruction text is not matched based on the general operation instruction set, it can be determined whether the number of words in the text of the current recognition result meets the preset number of words. If so, semantic understanding is performed to obtain the corresponding semantic instruction. If the number of words in the text does not meet the preset number of words, the next recognition result can be determined and executed. Figure 2 The process shown.

[0149] If the current recognition result is the final result, it can be determined whether there is a semantic understanding task corresponding to the same predicted text. If not, the semantic understanding is performed to obtain the corresponding semantic instruction. If so, the semantic instruction corresponding to the same predicted text can be directly executed to complete the vehicle voice response and end the process. After executing the instruction, the user's common sentence set can also be updated based on the executed instruction.

[0150] The voice command determination method provided in the embodiment of the present application can effectively intercept and filter the intermediate results to the semantic understanding module, reducing the number of intermediate results entering the semantic understanding module, thereby reducing the number of execution times of the semantic understanding task, or reducing the number of times the semantic understanding module is called and started, solving the problem of excessive computing power usage during vehicle-computer voice recognition, and avoiding vehicle-computer freezes and slow responses to vehicle-computer commands during voice recognition.

[0151] Figure 3 3 is a schematic diagram of a vehicle machine voice command determination device provided in an embodiment of the present application. The device includes a recognition module 310, a supplement module 320, a judgment module 330, a return module 340 and a final module 350, wherein:

[0152] The recognition module 310 is used to obtain the current recognition result corresponding to the speech to be processed issued by the user, and determine whether the current recognition result is an intermediate result;

[0153] A supplementing module 320, configured to supplement the intermediate result with text in response to the current recognition result being an intermediate result, to obtain a predicted text corresponding to the intermediate result;

[0154] A judgment module 330 is used to judge whether there is a semantic understanding task corresponding to the predicted text. If not, a semantic understanding task corresponding to the predicted text is generated to obtain a semantic instruction corresponding to the predicted text.

[0155] Return module 340, returns to the step of obtaining the current recognition result corresponding to the to-be-processed speech emitted by the user, and determining whether the current recognition result is an intermediate result;

[0156] The final module 350 is used to determine whether there is a semantic understanding task corresponding to the same predicted text as the final result in response to the current recognition result being the final result, and if so, take the semantic instruction corresponding to the predicted text as the final instruction.

[0157] On the basis of the above-mentioned implementation modes, optionally, the supplement module 320 includes a common sentence matching unit, which is used to match the intermediate result with each common sentence text in the user's common sentence set; and determine the predicted text corresponding to the intermediate result based on the matching results of each common sentence text.

[0158] On the basis of the above implementations, optionally, the common sentence matching unit is specifically used to:

[0159] Determine the first similarity between the intermediate result and each commonly used sentence text; determine the commonly used sentence text whose corresponding first similarity is greater than a preset first similarity threshold as a candidate text; if there are multiple candidate texts, determine the predicted text in each candidate text based on the number of times each candidate text is used, the most recent use time and the corresponding first similarity.

[0160] On the basis of the above-mentioned embodiments, optionally, the supplement module 320 also includes a general instruction matching unit, which is used to match the intermediate result with each general instruction text in the general operation instruction set if the matching of the predicted text from the common sentence set fails; and determine the predicted text corresponding to the intermediate result based on the matching result of each general instruction text.

[0161] On the basis of the above-mentioned embodiments, optionally, the device also includes a matching failure module, which is used to generate a semantic understanding task corresponding to the intermediate result if the matching of the predicted text from the general operation instruction set fails, and obtain the semantic instruction corresponding to the intermediate result.

[0162] Based on the above-mentioned embodiments, optionally, the matching failure module includes a skip-word processing unit, which is used to determine the number of text words in the intermediate result; based on the number of text words in the intermediate result, determine whether the intermediate result meets the preset semantic understanding entry conditions; if so, execute the semantic understanding task corresponding to the intermediate result to obtain the semantic instruction corresponding to the intermediate result.

[0163] On the basis of the above-mentioned embodiments, optionally, the final module 350 is also used to judge whether there is a semantic understanding task corresponding to an intermediate result with the same number of words as the final result in response to the current recognition result being the final result; if so, the semantic instruction corresponding to the intermediate result is used as the final instruction.

[0164] On the basis of the above-mentioned implementation modes, optionally, the final module 350 is also used to generate a semantic understanding task corresponding to the final result if there is no semantic understanding task corresponding to the predicted text identical to the final result, and obtain a semantic instruction corresponding to the final result; and use the semantic instruction corresponding to the final result as the final instruction.

[0165] Based on the above embodiments, optionally, the device also includes a common sentence updating module, which is used to execute the final instruction and, after the execution of the final instruction, use the final instruction as the common sentence to be added; and update the common sentence set based on the common sentence to be added.

[0166] On the basis of the above implementation modes, optionally, the common sentence updating module is specifically used for:

[0167] It is determined whether the commonly used sentence to be added exists in the commonly used sentence set; if so, the usage count and the most recent usage time of the commonly used sentence to be added in the commonly used sentence set are updated.

[0168] The vehicle-mounted computer voice command determination device provided in the embodiment of the present application can execute the steps in the voice command determination method provided in the method embodiment of the present application, and the execution steps and beneficial effects are no longer repeated here.

[0169] Figure 4 Schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Figure 4 As shown, the electronic device 400 includes one or more processors 401 and a memory 402 .

[0170] The processor 401 may be a central processing unit (CPU) or other forms of processing units having data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device 400 to perform desired functions.

[0171] The memory 402 may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, a random access memory (RAM) and / or a cache memory (cache), etc. The non-volatile memory may include, for example, a read-only memory (ROM), a hard disk, a flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor 401 may run the program instructions to implement the voice instruction determination method of any embodiment of the present application described above and / or other desired functions. Various contents such as initial external parameters, thresholds, etc. may also be stored in the computer-readable storage medium.

[0172] In one example, the electronic device 400 may further include: an input device 403 and an output device 404, which are interconnected via a bus system and / or other forms of connection mechanisms (not shown). The input device 403 may include, for example, a keyboard, a mouse, etc. The output device 404 may output various information to the outside, including early warning information, braking force, etc. The output device 404 may include, for example, a display, a speaker, a printer, a communication network and a remote output device connected thereto, etc.

[0173] Of course, to simplify, Figure 4 Only some of the components related to the present application in the electronic device 400 are shown, and components such as a bus, an input / output interface, etc. are omitted. In addition, according to specific application conditions, the electronic device 400 may also include any other appropriate components.

[0174] In addition to the above-mentioned methods and devices, an embodiment of the present application may also be a computer program product, which includes computer program instructions, which, when executed by a processor, enable the processor to execute the steps of the voice instruction determination method provided in any embodiment of the present application.

[0175] The computer program product may be written in any combination of one or more programming languages ​​to write program codes for performing the operations of the embodiments of the present application, including object-oriented programming languages, such as Java, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user computing device, as an independent software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0176] In addition, an embodiment of the present application may also be a computer-readable storage medium on which computer program instructions are stored. When the computer program instructions are executed by a processor, the processor executes the steps of the voice instruction determination method provided in any embodiment of the present application.

[0177] The computer readable storage medium can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can include, for example, but is not limited to, a system, device or device of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination of the above. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0178] It should be noted that the terms used in this application are only for describing specific embodiments, rather than limiting the scope of this application. As shown in the specification and claims of this application, unless the context clearly indicates an exception, the words "one", "a", "a kind of" and / or "the" do not specifically refer to the singular, but may also include the plural. The terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method or device. In the absence of further restrictions, the elements defined by the sentence "including one..." do not exclude the presence of other identical elements in the process, method or device including the elements.

[0179] It should also be noted that the terms "center", "up", "down", "left", "right", "vertical", "horizontal", "inside", "outside", etc., indicating the orientation or position relationship, are based on the orientation or position relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present application. Unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", etc. should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be an indirect connection through an intermediate medium, or it can be a connection between the two elements. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0180] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and its core ideas of this application. The above is only the preferred implementation method of this application. It should be pointed out that due to the limitations of textual expression and the objective existence of infinite specific structures, ordinary technicians in this technical field can make several improvements, modifications or changes without departing from the principles of this application, and can also combine the above technical features in an appropriate manner; these improvements, modifications, changes or combinations, or the direct application of the concept and technical solution of the invention to other occasions without improvement, should be regarded as the scope of protection of this application.

Claims

1. A method for determining a voice instruction, characterized in that: include: Obtaining a current recognition result corresponding to the speech to be processed issued by the user, and determining whether the current recognition result is an intermediate result; In response to the current recognition result being an intermediate result, supplementing the intermediate result with text to obtain a predicted text corresponding to the intermediate result; Determine whether there is a semantic understanding task corresponding to the predicted text, and if not, generate the semantic understanding task corresponding to the predicted text to obtain the semantic instruction corresponding to the predicted text; Returning to the step of obtaining the current recognition result corresponding to the to-be-processed speech emitted by the user, and determining whether the current recognition result is an intermediate result; In response to the current recognition result being the final result, it is determined whether there is a semantic understanding task corresponding to the same predicted text as the final result, and if so, the semantic instruction corresponding to the predicted text is used as the final instruction.

2. The method according to claim 1, characterized in that Supplementing the intermediate result with text to obtain a predicted text corresponding to the intermediate result includes: Matching the intermediate result with each common sentence text in the user's common sentence set; The predicted text corresponding to the intermediate result is determined based on the matching results of each common sentence text.

3. The method according to claim 2, characterized in that Matching the intermediate result with each common sentence text in the common sentence set of the user includes: Determining a first similarity between the intermediate result and each common sentence text; Determining the predicted text corresponding to the intermediate result based on the matching results of each common sentence text includes: Determine the commonly used sentence texts whose corresponding first similarity is greater than a preset first similarity threshold as candidate texts; If there are multiple candidate texts, the predicted text is determined from each candidate text based on the number of times each candidate text is used, the most recent time of use, and the corresponding first similarity.

4. The method according to claim 3, characterized in that The method further comprises: If the matching of the predicted text from the common sentence set fails, matching the intermediate result with each general instruction text in the general operation instruction set; The predicted text corresponding to the intermediate result is determined based on the matching results of each general instruction text.

5. The method according to claim 4, characterized in that The method further comprises: If the matching of the predicted text from the general operation instruction set fails, a semantic understanding task corresponding to the intermediate result is generated to obtain a semantic instruction corresponding to the intermediate result.

6. The method according to claim 5, characterized in that Before generating the semantic understanding task corresponding to the intermediate result and obtaining the semantic instruction corresponding to the semantic understanding task, the method further includes: Determining the number of words in the text of the intermediate result; Determining whether the intermediate result meets a preset semantic understanding entry condition based on the number of words in the text of the intermediate result; If so, execute the step of generating the semantic understanding task corresponding to the intermediate result and obtaining the semantic instruction corresponding to the intermediate result.

7. The method according to claim 5, characterized in that The method further comprises: In response to the current recognition result being a final result, determining whether there is a semantic understanding task corresponding to an intermediate result having the same number of characters as the final result; If so, the semantic instruction corresponding to the intermediate result is used as the final instruction.

8. The method according to claim 1, characterized in that After determining whether there is a semantic understanding task corresponding to the predicted text that is the same as the final result, the method further includes: If there is no semantic understanding task corresponding to the predicted text identical to the final result, generating a semantic understanding task corresponding to the final result, and obtaining a semantic instruction corresponding to the final result; The semantic instruction corresponding to the final result is used as the final instruction.

9. The method according to claim 2, characterized in that: The method further comprises: Executing the final instruction, and after the execution of the final instruction is completed, using the final instruction as the common sentence to be added; The common sentence set is updated based on the common sentences to be added.

10. The method according to claim 9, characterized in that Updating the common sentence set based on the common sentence to be added includes: Determining whether the commonly used sentence to be added exists in the commonly used sentence set; If so, the usage count and the most recent usage time of the to-be-added common sentence in the common sentence set are updated.

11. An electronic device, characterized in that: The electronic device comprises: Processor and memory; The processor is used to execute the steps of the voice instruction determination method according to any one of claims 1 to 10 by calling the program or instruction stored in the memory.

12. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a program or instruction, and the program or instruction enables a computer to execute the steps of the voice instruction determination method according to any one of claims 1 to 10.

Images