Voice interaction method, server and computer-readable storage medium

Abstract

The present application discloses a voice interaction method, a server, and a computer-readable storage medium, the method comprising: receiving a current voice request forwarded by a vehicle, and in the case where the vehicle control instruction corresponding to the current voice request cannot be confirmed, determining target guidance information for guiding the user to adjust the current voice request based on a large language model and the current voice request, and feeding back the target guidance information to guide the user to complete the voice interaction. In this way, the server of the present application can determine the target guidance information based on the large language model, and feed back the target guidance information to the user to guide the user to adjust the voice request, and the user's experience can be guaranteed. The present application can determine the target guidance information based on the current voice request, the target guidance information is related to the current voice request, and the reliability is guaranteed. The target guidance information is determined by the large language model, and the credibility of the target guidance information is guaranteed.

Description

Technical Field

[0001] The present application relates to the field of voice interaction technology, and in particular to a voice interaction method, a server, and a computer-readable storage medium. Background Art

[0002] When the user first starts using the vehicle, he or she may express some vague and incomplete voice commands because he or she is unfamiliar with the vehicle and its voice interaction functions. When the vehicle processes these voice commands, it has difficulty understanding the actual meaning of the voice commands and is unable to execute the corresponding functions, resulting in voice interaction failure. Summary of the invention

[0003] The present application provides a voice interaction method, a server and a computer-readable storage medium.

[0004] A voice interaction method provided in an embodiment of the present application includes:

[0005] Receive the current voice request forwarded by the vehicle;

[0006] In a case where a vehicle control instruction corresponding to the current voice request cannot be determined according to the current voice request, determining target guidance information for guiding a user to adjust the current voice request according to a large language model and the current voice request, wherein the large language model is pre-trained and is capable of generating a plurality of candidate guidance information based on a first voice request sample, and determining a predicted guidance information based on a ranking result of the plurality of candidate guidance information;

[0007] The target guidance information is fed back to guide the user to complete the voice interaction.

[0008] In the voice interaction method provided in the embodiment of the present application, the server can receive the current voice request forwarded by the vehicle, and when the vehicle control instruction corresponding to the current voice request cannot be confirmed, call a pre-trained large language model that can generate multiple candidate guidance information through a first voice request sample, and determine a predicted guidance information based on the sorting results of the multiple candidate guidance information, determine the target guidance information for guiding the user to adjust the current voice request, and feed back the target guidance information to guide the user to complete the voice interaction.

[0009] Thus, in the embodiment of the present application, the server can determine the target guidance information for guiding the user to adjust the voice request based on the pre-completed large language model when the vehicle control instruction corresponding to the voice request cannot be determined, and the target guidance information is fed back to the user to guide the user to adjust the voice request, so that when the user is unclear about how to express the voice request for controlling the vehicle, the user can be guided to adjust the voice request through the target guidance information, and the user's experience of the vehicle and the vehicle voice interaction function can be guaranteed. The embodiment of the present application can determine the target guidance information according to the current voice request, so that the target guidance information can be related to the current voice request, and then the matching of the target guidance information with the user's current use needs can be guaranteed to a certain extent, and the reliability of the target guidance information can be guaranteed. And, the embodiment of the present application can determine the target guidance information through the large language model, so the credibility of the target guidance information can be guaranteed to a certain extent. And, the large language model in the embodiment of the present application can determine the predicted guidance information in the multiple candidate guidance information according to the sorting results of the multiple candidate guidance information, so the predicted guidance information can be related to the sorting results of the multiple candidate guidance information, and the credibility of the predicted guidance information and the target guidance information can be guaranteed to a certain extent.

[0010] In certain embodiments of the present application, the step of training the large language model includes:

[0011] Determining the plurality of candidate guidance information according to the first voice request sample and a pre-trained reference model;

[0012] Determining first candidate guidance information and second candidate guidance information among the plurality of candidate guidance information according to the sorting result;

[0013] The reference model is trained according to a combination of the first candidate guidance information and the second candidate guidance information to obtain the large language model.

[0014] Thus, in an embodiment of the present application, when the reference model determines multiple candidate guide information for the first voice request sample, the server can determine the first candidate guide information and the second candidate guide information among the multiple candidate guide information according to the sorting results of the multiple candidate guide information, and complete the training of the reference model through the combination of the first candidate guide information and the second candidate guide information, so that the reference model can learn the association between the first candidate guide information and the second candidate guide information in the combination to a certain extent.

[0015] In certain embodiments of the present application, the training of the reference model to obtain the large language model according to the combination of the first candidate guidance information and the second candidate guidance information includes:

[0016] Determine, according to the combination and the reference model, a first prediction result of the reference model for first candidate guidance information in the combination and a second prediction result for second candidate guidance information in the combination;

[0017] According to the difference between the first prediction result and the second prediction result, the reference model is trained to obtain the large language model.

[0018] In this way, in an embodiment of the present application, the server can train the reference model based on the degree of difference between the first prediction result and the second prediction result when it is determined that the reference model predicts a first prediction result of the first candidate guidance information in the combination, and the reference model predicts a second prediction result of the second candidate guidance information in the combination.

[0019] In certain embodiments of the present application, the reference model includes a target model and a copy model of the target model, the degree of difference includes a first degree of difference corresponding to the target model and a second degree of difference corresponding to the copy model, and the training of the reference model according to the degree of difference between the first prediction result and the second prediction result to obtain the large language model includes:

[0020] The target model is trained according to the first difference degree and the second difference degree to obtain the large language model.

[0021] In this way, in the implementation mode of the present application, the target model can be trained based on the prediction results of the copy model, so that the training of the target model can be carried out with reference to the copy model to a certain extent, thereby reducing the possibility of performance degradation of the target model due to training to a certain extent, and the training of the target model can be carried out reliably.

[0022] In certain embodiments of the present application, the step of training the large language model includes:

[0023] The plurality of candidate guidance information are determined according to the first voice request sample, the reference model and a pre-configured prompt information template.

[0024] Thus, in the implementation manner of the present application, the reference model can perform natural language understanding and processing on the first voice request sample based on the instructions of the prompt information template, thereby generating candidate guidance information for the first voice request sample, so that the robust operation of the reference model can be guaranteed.

[0025] In certain embodiments of the present application, the step of training the large language model includes:

[0026] Acquire a second voice request sample and a guidance information tag corresponding to the second voice request sample;

[0027] According to the second voice request sample and the guide information label, a predetermined basic model is trained to determine the reference model.

[0028] Thus, in the implementation mode of the present application, the server can train the basic model through the second voice request sample that does not correspond to the vehicle control command and the guidance information label corresponding to the second voice request to obtain a reference model that can generate guidance information according to the voice request that does not correspond to the vehicle control command.

[0029] In certain embodiments of the present application, when a vehicle control instruction corresponding to the current voice request cannot be determined according to the current voice request, determining target guidance information for guiding the user to adjust the current voice request according to the large language model and the current voice request includes:

[0030] When a vehicle control instruction corresponding to the current voice request cannot be determined according to the current voice request, a recommended voice request corresponding to the current voice request is generated according to the large language model and the current voice request to determine the target guidance information.

[0031] In this way, in the implementation mode of the present application, the server can infer the recommended voice request corresponding to the current voice request based on the large language model, and form target guidance information based on the recommended voice request, and then the user can adjust the current voice request based on the recommended voice request, so that the validity of the target guidance information can be guaranteed.

[0032] In certain embodiments of the present application, when a vehicle control instruction corresponding to the current voice request cannot be determined according to the current voice request, determining target guidance information for guiding the user to adjust the current voice request according to the large language model and the current voice request includes:

[0033] When the vehicle control instruction corresponding to the current voice request cannot be determined based on the large language model, the current voice request and the first sub-prompt information template in the pre-configured thought chain prompt information template, the target guidance information is determined according to the large language model, the current voice request and the second sub-prompt information template in the thought chain prompt information template.

[0034] Thus, in the implementation mode of the present application, the large language model can confirm the correspondence between the current voice request and the vehicle control instruction based on the first sub-prompt information template, the second sub-prompt information template and the current voice request in the thought chain prompt information template, and when it is confirmed that the current voice request corresponds to the vehicle control instruction, confirm the target guidance information corresponding to the current voice request, thereby realizing the reasoning of the large language model based on the thought chain prompt information template, and thus the reasoning accuracy of the large language model can be guaranteed to a certain extent.

[0035] An embodiment of the present application provides a server, including a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the above-mentioned voice interaction method is implemented.

[0036] An embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by one or more processors, the above-mentioned voice interaction method is implemented.

[0037] The server and computer-readable storage medium of the embodiment of the present application can determine the target guidance information for guiding the user to adjust the voice request based on the pre-completed large language model when the vehicle control instruction corresponding to the voice request cannot be determined, and the target guidance information is fed back to the user to guide the user to adjust the voice request, so that when the user is unclear about how to express the voice request for controlling the vehicle, the user can be guided to adjust the voice request through the target guidance information, and the user's experience of the vehicle and the vehicle voice interaction function can be guaranteed. The embodiment of the present application can determine the target guidance information according to the current voice request, so that the target guidance information can be related to the current voice request, and then the matching of the target guidance information with the user's current use needs can be guaranteed to a certain extent, and the reliability of the target guidance information can be guaranteed. And, the embodiment of the present application can determine the target guidance information through the large language model, so the credibility of the target guidance information can be guaranteed to a certain extent. And, the large language model in the embodiment of the present application can determine the predicted guidance information in the multiple candidate guidance information according to the sorting results of multiple candidate guidance information, so the predicted guidance information can be related to the sorting results of multiple candidate guidance information, and the credibility of the predicted guidance information and the target guidance information can be guaranteed to a certain extent.

[0038] Additional aspects and advantages of the embodiments of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the embodiments of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0040] Figure 1 A flowchart of a voice interaction method in certain implementation modes of the present application;

[0041] Figure 2 A flowchart of a voice interaction method in certain implementation modes of the present application;

[0042] Figure 3 A flowchart of a voice interaction method in certain implementation modes of the present application;

[0043] Figure 4 A flowchart of a voice interaction method in certain implementation modes of the present application;

[0044] Figure 5 A flowchart of a voice interaction method in certain implementation modes of the present application;

[0045] Figure 6 A flowchart of a voice interaction method in certain implementation modes of the present application;

[0046] Figure 7 A flowchart of a voice interaction method in certain implementation modes of the present application;

[0047] Figure 8 A flowchart of a voice interaction method in certain implementation modes of the present application;

[0048] Fig. 9 A flowchart of a voice interaction method in certain implementation modes of the present application;

[0049] Fig.10 This is a flowchart of the voice interaction method in certain implementation modes of the present application. DETAILED DESCRIPTION

[0050] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals represent the same or similar elements or elements having the same or similar functions from beginning to end. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the embodiments of the present application, and cannot be understood as limiting the embodiments of the present application.

[0051] When users first use a vehicle or its voice interaction functions, they are relatively unfamiliar with the objects and functions in the vehicle that can be controlled by voice. Therefore, they may express some voice commands with ambiguous meanings, confusing sentence components or missing components, such as "adjust the air conditioner" or "turn on the temperature adjuster".

[0052] It is understandable that these voice commands have one or more problems such as unclear intention, missing components (or ambiguity), incorrect collocation of sentence components, etc. For example, it can be considered that the collocation relationship between 'adjust' and 'air conditioner' in "adjust air conditioner" is inappropriate, and it can also be considered that the intention corresponding to "adjust air conditioner" includes increasing and decreasing the cooling temperature / heating temperature of the air conditioner, switching the operating mode of the air conditioner, etc., so the intention of "adjust air conditioner" is unclear.

[0053] Therefore, when processing these voice commands, it is difficult to determine the specific vehicle control instructions used to implement or complete these voice commands due to one or more problems such as unclear intentions, missing components (or ambiguity), incorrect sentence component matching, etc., which leads to the failure of the voice interaction with the user.

[0054] Further, to improve this situation, some solutions are timed dynamic advertisements to play preset text or video information for guiding users to use specific functions of the vehicle to teach users how to use the vehicle. However, this solution is difficult to meet the current usage needs of users.

[0055] In other solutions, the user is requested to clarify the voice commands, or in other words, when the user expresses a voice command with ambiguous meanings or incomplete sentence components, the user is asked a question in return. For example, when the user's voice command is "adjust the air conditioner", the vehicle can play a voice message saying "We didn't understand what you meant, please speak more clearly" to request the user to clarify the true intention of the voice command.

[0056] However, when requesting clarification from the user, the interaction with the user is usually based on fixed sentences, such as the above-mentioned "I don't understand what you mean, please speak more clearly". In this case, if the user is using the vehicle and the voice interaction function installed in the vehicle for the first time or has just started to use it, and the user is relatively unfamiliar with the objects and functions in the vehicle that can be controlled by voice, then when the user faces the counter-questions (or clarification requests) raised by the vehicle, the user may not know how to clarify, or in other words, do not know "how to express the voice command so that the vehicle can perform the corresponding function", resulting in the user having to have multiple rounds of dialogue and clarification with the vehicle before being able to control the vehicle through correct (or, understandable) voice commands.

[0057] Based on the above problems you may encounter, please refer to Figure 1 , the present application embodiment provides a voice interaction method, including:

[0058] 01: Receive the current voice request forwarded by the vehicle;

[0059] 02: when a vehicle control instruction corresponding to the current voice request cannot be determined according to the current voice request, target guidance information for guiding the user to adjust the current voice request is determined according to the large language model and the current voice request, wherein the large language model is pre-trained and is capable of generating multiple candidate guidance information based on the first voice request sample, and determining a predicted guidance information based on the sorting results of the multiple candidate guidance information;

[0060] 03: Provide target guidance information to guide users to complete voice interaction.

[0061] The embodiment of the present application provides a voice interaction device. The voice interaction method of the embodiment of the present application can be implemented by the voice interaction device of the embodiment of the present application. Specifically, the voice interaction device includes a receiving module, a determining module and an interaction module. Among them, the acquisition module is used to receive the current voice request forwarded by the vehicle. The determination module is used to determine the target guidance information for guiding the user to adjust the current voice request according to the large language model and the current voice request when the vehicle control instruction corresponding to the current voice request cannot be determined according to the current voice request. The large language model is pre-trained and can generate multiple candidate guidance information based on the first voice request sample, and determine a predicted guidance information based on the sorting results of multiple candidate guidance information. The interaction module is used to feedback the target guidance information to guide the user to complete the voice interaction.

[0062] The embodiment of the present application also provides a server, which includes a memory and a processor. The voice interaction method of the embodiment of the present application can be implemented by the server of the embodiment of the present application. Specifically, a computer program is stored in the memory, and the processor is used to receive the current voice request forwarded by the vehicle, and to determine the target guidance information for guiding the user to adjust the current voice request according to the large language model (LLM) and the current voice request when the vehicle control instruction corresponding to the current voice request cannot be determined according to the current voice request. The large language model is pre-trained and can generate multiple candidate guidance information based on the first voice request sample, and determine a predicted guidance information based on the sorting results of multiple candidate guidance information, and is used to feedback the target guidance information to guide the user to complete the voice interaction.

[0063] Specifically, in the embodiment of the present application, when the user wants to control the vehicle to perform a specific operation by voice, the corresponding current voice request is triggered at the current moment, and the vehicle can receive and forward the current voice request to the server. The server can receive the current voice request forwarded by the vehicle and perform corresponding natural language processing on the current voice request. However, in the case where the vehicle control instruction corresponding to the current voice request cannot be determined based on the natural language processing result of the current voice request, the server can input the current voice request into the pre-trained large language model to output the target guidance information that can be used to guide the user to adjust the current voice request through the large language model, and then, the server can obtain the target guidance information through the large language model, and then send the target guidance information to the vehicle, so as to feedback the target guidance information to the user through the vehicle, and guide the user to adjust the triggered voice request (that is, the current voice request triggered at the last moment), so that the vehicle can determine the effective voice request corresponding to the vehicle control instruction, and the vehicle can complete the interaction with the user through the effective voice request.

[0064] It can be understood that the voice request in the implementation mode of the present application can be understood as a voice instruction with one or more problems such as unclear intention, missing components (or ambiguity), and wrong combination of sentence components, such as "adjust the air conditioner" and "turn on 'the temperature adjuster'".

[0065] It can also be understood that the server of the implementation mode of the present application also performs natural language understanding of voice requests based on pre-set programs, codes or pre-completed natural language processing models, and generates vehicle control instructions through the natural language understanding results.

[0066] Furthermore, in the event that a vehicle control instruction corresponding to the current voice request cannot be generated, or in other words, in the event that a vehicle control instruction for implementing the current voice request can be generated (such as because the intention of "adjusting the air conditioner" is unclear and the corresponding vehicle control instruction cannot be generated), the server can call a pre-trained large language model and input the current voice request into the large language model to generate target guidance information through the large language model to guide the user to adjust the current voice request.

[0067] For example, the target guidance information for "adjusting the air conditioner" may include but is not limited to "I can't understand what you mean, here are some standard points you can say, such as: adjust the air conditioner to 20 degrees, adjust the air volume to the third gear, open the air conditioner page." and other forms.

[0068] It can be understood that since the target guidance information is generated by the large language model according to the current voice request, the target guidance information is related to the current voice request, and the target guidance information can guide the user to adjust and clarify the current voice request to a certain extent.

[0069] It can also be understood that the large language model in the implementation manner of the present application has completed corresponding training before being called by the server, so that it can output corresponding target guidance information when receiving the input of the current voice request.

[0070] Furthermore, during the training process of the large language model, the large language model may receive an input of a first voice request sample, and predict (or infer) the guidance information based on the first voice request sample to predict a plurality of candidate guidance information.

[0071] It is clear that for a voice request with one or more of the following problems, such as "unclear intention, missing components (or ambiguity), incorrect collocation of sentence components, etc.", different guidance information with different words, expressions, and statements can be used to guide the user to adjust the voice request. However, when guiding the user based on guidance information with different words, expressions, and statements, the words, expressions, and statements can directly affect the user's understanding and acceptance of the guidance information. It is understandable that if the guidance information is relatively blunt, it may cause user disgust, and if the guidance information is relatively detailed, it may be annoying to the user due to its tediousness.

[0072] In this regard, during the training process of the large language model, multiple predictions are made for each received first voice request sample based on the large language model (or each first voice request sample is repeatedly input into the large language model multiple times) to obtain multiple candidate guidance information for each first voice request sample. The server of the implementation mode of the present application can train the large language model based on the sorting results of the multiple candidate guidance information, so that the large language model can generate different candidate guidance information and determine the pros and cons of the candidate guidance information during the training process.

[0073] Optionally, during the training process of the large language model in certain embodiments of the present application, when the server obtains multiple candidate guide information for each first voice request sample, it can receive sorting and labeling instructions from the labeling personnel for the multiple candidate guide information for each first voice request sample, and determine the sorting results of the multiple candidate guide information for each first voice request sample based on the sorting and labeling instructions.

[0074] To more clearly illustrate the implementation of this application, please refer to Figure 2 , Figure 2 This is a schematic diagram of application scenarios in certain embodiments of the present application. Figure 2As shown, in an embodiment of the present application, a large language model that has not yet completed training can perform guidance information prediction for "adjust the air conditioner" multiple times after receiving the first voice request sample "adjust the air conditioner" to output three candidate guidance information, namely res1, res2 and res3.

[0075] Among them, res1 may include: "I can't understand what you mean. You can say it in a more standard way, such as: adjust the air conditioner to 20 degrees and adjust the air volume to the third gear." res2 may include: "I'm sorry, I can't understand what you mean. I suggest you say it, such as: adjust the air conditioner to 20 degrees and adjust the air volume to the third gear." res3 may include: "I'm sorry, I can't understand what you mean. If you want to adjust the temperature, you can say: adjust the air conditioner to 20 degrees. If you want to adjust the air volume, you can say: adjust the air volume to the third gear."

[0076] Furthermore, when res1, res2 and res3 are obtained, the annotator can manually annotate and sort res1, res2 and res3, or in other words, the server can receive the annotator's annotation and sorting instructions for res1, res2 and res3, so as to determine the quality of res1, res2 and res3, and determine the sorting result that res3 is the highest, res2 is the second, and res1 is the lowest.

[0077] Furthermore, the server can train the large language model based on the sorting result, so that the large language model can output high-quality target guidance information according to the voice request during the reasoning process after landing. For example, if the large language model is landed and the current voice request received is "adjust the air conditioner", the target guidance information (or guidance information prediction) generated by the large language model through "adjust the air conditioner" is res3.

[0078] It can be understood that, in the implementation manner of the present application, the sorting method of the plurality of candidate guidance information is content that can be set according to actual conditions, and manual sorting is only one of the feasible methods.

[0079] It is also understandable that after the server generates the target guidance information of the current voice request through the pre-trained large language model, the target guidance information can be sent to the vehicle. When the vehicle receives the target guidance information, it can play the target guidance information based on the display component or the sound playback component to guide the user to adjust or clarify the current voice request previously expressed, thereby adjusting the current voice request that cannot determine the vehicle control command to a valid voice request that can determine the vehicle control command, and allowing the vehicle to execute the vehicle control command corresponding to the valid voice request, and finally completing the voice interaction.

[0080] Optionally, in certain embodiments of the present application, the server may also send target guidance information to the user's terminal, so that the user can adjust the current voice request through the target guidance information played by the terminal.

[0081] In summary, in the implementation of the present application, the server can determine the target guidance information for guiding the user to adjust the voice request based on the pre-completed large language model when the vehicle control instruction corresponding to the voice request cannot be determined, and the target guidance information is fed back to the user to guide the user to adjust the voice request. Therefore, when the user is unclear about how to express the voice request for controlling the vehicle, the user can be guided to adjust the voice request through the target guidance information, and the user's experience of the vehicle and the vehicle voice interaction function can be guaranteed. The implementation of the present application can determine the target guidance information according to the current voice request, so that the target guidance information can be related to the current voice request, and then the target guidance information can be matched with the user's current use needs to a certain extent, and the reliability of the target guidance information can be guaranteed. And, the implementation of the present application can determine the target guidance information through the large language model, so the credibility of the target guidance information can be guaranteed to a certain extent. And, the large language model in the implementation of the present application can determine the predicted guidance information in the multiple candidate guidance information according to the sorting results of the multiple candidate guidance information, so the predicted guidance information can be related to the sorting results of the multiple candidate guidance information, and the credibility of the predicted guidance information and the target guidance information can be guaranteed to a certain extent.

[0082] See also Figure 3 In certain embodiments of the present application, the training step of the large language model includes:

[0083] 04: Determine multiple candidate guidance information according to the first voice request sample and the pre-trained reference model;

[0084] 05: Determine first candidate guidance information and second candidate guidance information from the plurality of candidate guidance information according to the sorting result;

[0085] 06: According to the combination of the first candidate guidance information and the second candidate guidance information, the reference model is trained to obtain a large language model.

[0086] The speech interaction device of the embodiment of the present application further includes a first guide information determination module, a second guide information determination module and a reference model training module. The first guide information determination module is used to determine the first candidate guide information and the second candidate guide information among multiple candidate guide information according to the sorting result. The second guide information determination module is used to determine the first candidate guide information and the second candidate guide information among multiple candidate guide information according to the sorting result. The reference model training module is used to train the reference model according to the combination of the first candidate guide information and the second candidate guide information to obtain a large language model.

[0087] The processor of the embodiment of the present application is also used to determine multiple candidate guide information based on the first voice request sample and the pre-trained reference model, and to determine the first candidate guide information and the second candidate guide information among the multiple candidate guide information based on the sorting result, and to train the reference model based on the combination of the first candidate guide information and the second candidate guide information to obtain a large language model.

[0088] Specifically, for the training process of the large language model that has not yet been trained, or for the training process of the reference model, the server of the embodiment of the present application can input the first voice request sample into the reference model so that the reference model performs multiple guidance information predictions on the first voice request sample, thereby outputting multiple candidate guidance information. Therefore, each first voice request sample corresponds to multiple candidate guidance information.

[0089] Furthermore, when multiple candidate guidance information corresponding to each first voice request sample is obtained, for each first voice request sample, according to the sorting results of the multiple candidate guidance information corresponding to the first voice request sample, the implementation mode of the present application can determine the first candidate guidance information with the highest quality among the multiple candidate guidance information, and multiple second candidate guidance information with lower quality than the first candidate guidance information.

[0090] Furthermore, the server may combine the first candidate guidance information with each second candidate guidance information to form multiple combinations. It can be understood that if there are n candidate guidance information corresponding to a first voice request sample, then: there is 1 first candidate guidance information and n-1 second candidate guidance information among the n candidate guidance information, and by combining 1 first candidate guidance information with each second candidate guidance information, a total of n-1 candidate guidance information combinations can be obtained.

[0091] To more clearly illustrate the implementation method of this application, please refer to Figure 3 , and see Figure 4 , Figure 4Schematic diagram of the combination of candidate guidance information in some embodiments of the present application. That is, let the first candidate guidance information be chosen, and let the second candidate guidance information be rejected, then: Figure 3 In the sorting result shown, the first candidate guide information is res3, and the second candidate guide information includes res1 and res2, so the following can be formed: Figure 4 There are two candidate guidance information combinations shown, in which res3 is chosen and res1 is rejected in one combination, and res3 is chosen and res2 is rejected in the other combination.

[0092] It should also be noted that when a combination of candidate guidance information consisting of one first candidate guidance information and one second candidate guidance information is obtained, the server of the implementation mode of the present application can use the combination to train the reference model so that the reference model can predict the better candidate guidance information in the combination (i.e., the first candidate guidance information) during the training process.

[0093] Optionally, in certain embodiments of the present application, the server is based on "a combination of 1 first candidate guidance information and 1 second candidate guidance information", so that the reference model can predict the first candidate guidance information and the second candidate guidance information in the combination, while also ensuring that the probability of the reference model predicting the first candidate guidance information in the combination is higher than the probability of predicting the second candidate guidance information, thereby ensuring that the trained reference model (or large language model) can output the first candidate guidance information as a result when it is landed and receives the first voice request sample.

[0094] Thus, in an embodiment of the present application, when the reference model determines multiple candidate guide information for the first voice request sample, the server can determine the first candidate guide information and the second candidate guide information among the multiple candidate guide information according to the sorting results of the multiple candidate guide information, and complete the training of the reference model through the combination of the first candidate guide information and the second candidate guide information, so that the reference model can learn the association between the first candidate guide information and the second candidate guide information in the combination to a certain extent.

[0095] See also Figure 5 In certain embodiments of the present application, step 06 includes:

[0096] 060: Determine, according to the combination and the reference model, a first prediction result of the reference model for the first candidate guidance information in the combination, and a second prediction result for the second candidate guidance information in the combination;

[0097] 061: According to the difference between the first prediction result and the second prediction result, the reference model is trained to obtain a large language model.

[0098] The reference model training module of the implementation mode of the present application is also used to determine the first prediction result of the reference model for the first candidate guidance information in the combination and the second prediction result for the second candidate guidance information in the combination based on the combination and the reference model, and to train the reference model according to the degree of difference between the first prediction result and the second prediction result to obtain a large language model.

[0099] The processor of the embodiment of the present application is also used to determine, based on the combination and the reference model, a first prediction result of the reference model for the first candidate guidance information in the combination, and a second prediction result for the second candidate guidance information in the combination, and to train the reference model according to the degree of difference between the first prediction result and the second prediction result to obtain a large language model.

[0100] Specifically, in order to make the "probability of predicting the first candidate guidance information in the combination" of the reference model higher than the "probability of predicting the second candidate guidance information in the combination" during the training process, the implementation mode of the present application can determine the first prediction result of the reference model predicting the first candidate guidance information of the combination through the first voice request sample through the prediction result generated by the reference model based on the first voice request sample during the training process, and determine the second prediction result of the reference model predicting the second candidate guidance information of the combination through the first voice request sample.

[0101] Among them, the prediction results in the implementation of the present application can be understood as classification probability or confidence probability, etc. Therefore, in the implementation of the present application, "the reference model predicts the first prediction result of the combined first candidate guidance information through the first voice request sample" can be understood as the reference model receiving the input of the first voice request sample and outputting the probability P1 of the first candidate guidance information.

[0102] Correspondingly, "the reference model predicts the second prediction result of the combined second candidate guidance information through the first voice request sample" can be understood as the reference model receiving the input of the first voice request sample and outputting the probability P2 of the second candidate guidance information.

[0103] Furthermore, let the first prediction result be the above-mentioned P1 and the second prediction result be the above-mentioned P2, then the server can train the reference model according to the difference between P1 and P2, so that during the training process of the model, P1 gradually increases and P2 gradually decreases, and the difference between P1 and P2 can increase with the increase of training rounds until it stabilizes.

[0104] In this way, in an embodiment of the present application, the server can train the reference model based on the degree of difference between the first prediction result and the second prediction result when it is determined that the reference model predicts a first prediction result of the first candidate guidance information in the combination, and the reference model predicts a second prediction result of the second candidate guidance information in the combination.

[0105] In some embodiments of the present application, the reference model includes a target model and a replica model of the target model, and the difference degree includes a first difference degree corresponding to the target model and a second difference degree corresponding to the replica model, and step 061 includes:

[0106] According to the first difference degree and the second difference degree, the target model is trained to obtain a large language model.

[0107] The reference model training module of the embodiment of the present application is also used to train the target model according to the first difference degree and the second difference degree to obtain a large language model.

[0108] The processor of the embodiment of the present application is also used to train the target model according to the first difference degree and the second difference degree to obtain a large language model.

[0109] Specifically, in order to avoid excessive divergence in the guidance information prediction results of the reference model during the training process, which would lead to a decrease in the guidance information prediction reasoning ability originally possessed by the reference model training (i.e., degradation of the target model), the server of the implementation method of the present application may load a large language model that has not yet been trained twice, thereby obtaining a target model and a copy of the target model (i.e., a copy model).

[0110] To more clearly illustrate the implementation of this application, please refer to Figure 6 , Figure 6 This is a flowchart of the voice interaction method in certain embodiments of the present application. That is, when the server obtains the target model and the replica model, it can input the first voice request sample to the target model and the replica model. Furthermore, based on the combination of previously constructed candidate guidance information, and the results output by the target model and the replica model through the first voice request sample respectively, it can be determined that the target model predicts the first prediction result of the first candidate guidance information in the combination and predicts the second prediction result of the second candidate guidance information in the combination, and it can also be determined that the replica model predicts the first prediction result of the first candidate guidance information in the combination and predicts the second prediction result of the second candidate guidance information in the combination.

[0111] To clearly illustrate the implementation of the present application, let the target model predict the first prediction result of the first candidate guidance information in the combination as P 1-1 , the target model predicts that the second prediction result of the second candidate guidance information in the combination is P1-2 , the replica model predicts that the first prediction result of the first candidate guidance information in the combination is P 2-1 , the replica model predicts that the second prediction result of the second candidate guidance information in the combination is P 2-2 , then: According to P 1-1 Relative to P 1-2 The first difference degree Diff1, and P 2-1 Relative to P 2-2 The second degree of difference Diff2 is used to train the target model to obtain the large language model required by the implementation mode of the present application.

[0112] Optionally, in the implementation manner of the present application, the server is based on the above P 1-1 , P 1-2 , P 2-1 , P 2-2 , Diff1 and Diff2 to train the target model. The training objectives may include: 1-1 Can be higher than P 1-2 , and P 1-1 With P 1-2 The difference between Diff1 and Diff2 increases until convergence, and the similarity between Diff1 and Diff2 is maintained within a preset range.

[0113] Optionally, in some other implementations of the present application, the server is based on the above P 1-1 , P 1-2 , P 2-1 , P 2-2 , Diff1 and Diff2 to train the target model. The training objectives may include: for the target model, P 1-1 Higher than P 1-2 , and P 1-1 With P 1-2 The difference between the two increases until convergence. For the replica model, P 2-2 Higher than P 2-1 , and P 2-2 With P 2-1 The difference between the two increases until it converges.

[0114] Optionally, in certain embodiments of the present application, the server may calculate the similarity between the above Diff1 and the above Diff2, and train the target model based on the similarity, wherein the similarity may be calculated based on the KL divergence (Kullback-Leibler Divergence) calculation formula.

[0115] In this way, in the implementation mode of the present application, the target model can be trained based on the prediction results of the copy model, so that the training of the target model can be carried out with reference to the copy model to a certain extent, thereby reducing the possibility of performance degradation of the target model due to training to a certain extent, and the training of the target model can be carried out reliably.

[0116] In certain embodiments of the present application, the step of training the large language model further includes:

[0117] A plurality of candidate guide information is determined according to the first voice request sample, a reference model and a pre-configured prompt information template.

[0118] The implementation manner of the present application determines multiple candidate guidance information based on the first voice request sample, the reference model and the pre-configured prompt information template.

[0119] The processor of the implementation mode of the present application is also used to determine multiple candidate guidance information based on the first voice request sample, the reference model and the pre-configured prompt information template.

[0120] Specifically, in order to ensure that the reference model can understand that it needs to predict the guidance information for the first voice request sample when it receives the first voice request sample, the implementation method of the present application can input a pre-configured prompt (prompt or instruction) information template into the reference model at the same time as the first voice request sample is input into the reference model.

[0121] It can be understood that the prompt information template can be understood as a natural language text segment, including "natural language text that describes the downstream tasks that the model needs to handle" and "explanatory text that describes how the model should reason about the input information."

[0122] It can also be understood that the prompt information template in the implementation manner of the present application is content that can be set according to actual conditions.

[0123] Exemplarily, in one example, the prompt information template may include: "If you are an intelligent voice assistant, the first step is to select an API that can be executed or provide information based on the instructions provided by the user, and extract the key information in the instructions for API search for relevant information. The returned results include, API name: api, key information: arguments. Step 2: If the API is unclear, recommend relevant annotation instructions to the user. If it is not unclear, directly return the API result."

[0124] Among them, the "instruction" in the example of the above prompt information template can be understood as the voice request in the implementation of the present application. "If the API is unclear" can be understood as the large language model receiving the voice request and recognizing that the voice request cannot be implemented through the pre-defined API (Application interface), such as the semantically ambiguous voice request of "adjusting the air conditioner". There is no API for implementing ambiguous vehicle functions, so "adjusting the air conditioner" cannot be implemented through the defined API, and then the API of the voice request is confirmed as unclear.

[0125] Furthermore, the reference model of the implementation manner of the present application can, upon receiving the first voice request sample and the prompt information template, infer the first voice request sample according to the prompt information template, and upon inferring that the API of the first voice request sample is unclear, predict the guidance information of the first voice request sample, thereby obtaining candidate guidance information for the first voice request sample.

[0126] Optionally, in certain embodiments of the present application, the server may combine the prompt information template and the first voice request sample to form prompt information, and then input the prompt information into a reference model, so that the reference model completes the guidance information reasoning work of the first voice request sample based on the prompt information template and the first voice request sample in the prompt information.

[0127] Thus, in the implementation manner of the present application, the reference model can perform natural language understanding and processing on the first voice request sample based on the instructions of the prompt information template, thereby generating candidate guidance information for the first voice request sample, so that the robust operation of the reference model can be guaranteed.

[0128] See also Figure 7 In certain embodiments of the present application, the training step of the large language model includes:

[0129] 07: Obtain a second voice request sample and a guidance information tag corresponding to the second voice request sample;

[0130] 08: According to the second voice request sample and the guide information label, a predetermined basic model is trained to determine a reference model.

[0131] The sample acquisition module and basic model training module of the implementation mode of the present application. The sample acquisition module is used to obtain the second voice request sample and the guide information label corresponding to the second voice request sample. The basic model training module is used to train a predetermined basic model according to the second voice request sample and the guide information label to determine a reference model.

[0132] The processor of the embodiment of the present application is also used to obtain a second voice request sample and a guide information label corresponding to the second voice request sample, and to train a predetermined basic model according to the second voice request sample and the guide information label to determine a reference model.

[0133] Specifically, in order to reduce the number of module deployments and make full use of the trainable parameters in the large language model, the server of the implementation mode of the present application can train a basic model with certain natural language processing capabilities based on a second voice request sample that is received in advance and does not correspond to a vehicle control instruction, and a prompt information label corresponding to the second voice request sample, so that the basic model can learn the characteristics of the "voice request that does not correspond to the vehicle control instruction" and can be applicable to downstream tasks of inferring guidance information.

[0134] It can be understood that the second voice request sample and the guidance information label corresponding to the second voice request sample are contents that can be designed according to actual conditions. For example, the second voice request sample can be "car window", and the guidance information label corresponding to the sample can include voice requests such as "open car window" and "close car window", and then the complete guidance information label can be "I can't understand what you mean, you can say: open car window. Or: close car window".

[0135] Thus, in the implementation mode of the present application, the server can train the basic model through the second voice request sample that does not correspond to the vehicle control command and the guidance information label corresponding to the second voice request to obtain a reference model that can generate guidance information according to the voice request that does not correspond to the vehicle control command.

[0136] Optionally, in certain embodiments of the present application, the basic model (or base model) can be understood as a model that has certain natural language processing capabilities and can be applied to conventional natural language processing tasks in general fields.

[0137] Optionally, in certain embodiments of the present application, since the basic model is applicable to conventional natural language processing tasks in the general field, but is relatively unfamiliar with voice request processing tasks in the vehicle field, the server of the embodiments of the present application may also perform corresponding pre-training and / or knowledge injection on the basic model so that the basic model can understand the semantics of unique characters, words and sentences in the vehicle field, thereby improving the adaptability of the basic model to voice request processing tasks in the vehicle field.

[0138] In certain embodiments of the present application, step 02 includes:

[0139] When a vehicle control instruction corresponding to the current voice request cannot be determined according to the current voice request, a recommended voice request corresponding to the current voice request is generated according to the large language model and the current voice request to determine target guidance information.

[0140] The determination module of the implementation mode of the present application is also used to generate a recommended voice request corresponding to the current voice request based on the large language model and the current voice request to determine the target guidance information when the vehicle control instruction corresponding to the current voice request cannot be determined based on the current voice request.

[0141] The processor of the embodiment of the present application is also used to generate a recommended voice request corresponding to the current voice request based on the large language model and the current voice request to determine the target guidance information when the vehicle control instruction corresponding to the current voice request cannot be determined based on the current voice request.

[0142] Specifically, in order to ensure that the target guidance information can reliably guide the user to adjust the current voice request, the target guidance information in the implementation mode of the present application can also be generated based on the "recommended voice request" corresponding to the current voice request (or semantically similar) so that the user can complete the reliable adjustment of the current voice request based on the recommended voice request used as an example in the target guidance information.

[0143] For example, when the current voice request is "adjust the air conditioner", the large language model receives the current voice request, and the recommended voice requests inferred based on the current voice request may be "adjust the air conditioner to 20 degrees" and "adjust the wind volume to the third gear".

[0144] Furthermore, when a recommended voice request is received, the large language model improves or completes the recommended voice request to form corresponding target guidance information, such as "I cannot understand what you mean. You can say standard points, such as: adjust the air conditioner to 20 degrees, adjust the air volume to the third gear, and open the air conditioner page."

[0145] In this way, in the implementation mode of the present application, the server can infer the recommended voice request corresponding to the current voice request based on the large language model, and form target guidance information based on the recommended voice request, and then the user can adjust the current voice request based on the recommended voice request, so that the validity of the target guidance information can be guaranteed.

[0146] In certain embodiments of the present application, step 02 includes:

[0147] When the vehicle control instruction corresponding to the current voice request cannot be determined based on the large language model, the current voice request and the first sub-prompt information template in the pre-configured thinking chain prompt information template, the target guidance information is determined according to the large language model, the current voice request and the second sub-prompt information template in the thinking chain prompt information template.

[0148] The determination module of the implementation mode of the present application is also used to determine the target guidance information according to the large language model, the current voice request and the second sub-prompt information template in the thinking chain prompt information template when the vehicle control instruction corresponding to the current voice request cannot be determined based on the large language model, the current voice request and the first sub-prompt information template in the pre-configured thinking chain prompt information template.

[0149] The processor of the embodiment of the present application is also used to determine the target guidance information based on the large language model, the current voice request and the second sub-prompt information template in the thinking chain prompt information template when the vehicle control instruction corresponding to the current voice request cannot be determined based on the large language model, the current voice request and the first sub-prompt information template in the pre-configured thinking chain prompt information template.

[0150] Specifically, in order to enable the large language model to reliably and accurately complete the target-guided reasoning, the server of the implementation mode of the present application can input the current voice request and the pre-built Chain-of-Thought Prompt (COT) information template into the large language model, so that the large language model can perform corresponding reasoning on the current voice request according to the reasoning steps indicated by the Chain-of-Thought Prompt information template, and then output the target-guided information.

[0151] It can be understood that the specific form of the thought chain prompt can be set according to the actual situation. For example, in some embodiments, the thought chain prompt may include "Let's think step by step".

[0152] Correspondingly, in the implementation mode of the present application, since the large language model is used to identify whether the current voice request can correspond to a vehicle control instruction, and is also used to generate target guidance information of the current voice request, the thinking chain prompt information of the implementation mode of the present application can be divided into two parts. One part can be understood as a first sub-prompt information template used to prompt the large language model to identify "whether the current voice request can correspond to a vehicle control instruction", and the other part can be understood as a second sub-prompt information template used to prompt the large language model to generate "target guidance information corresponding to the current voice request".

[0153] In one example, a thinking chain prompt information template including a first sub-prompt information template and a second sub-prompt information template includes "If you are an intelligent voice assistant, the first step is to select an API that can be executed or provide information based on the instructions provided by the user, and extract the key information in the instructions for API search for relevant information. The returned results include, API name: api, key information: arguments. Step 2: If the API is unclear, recommend relevant annotation instructions to the user. If it is not unclear, directly return the API results."

[0154] Further, it can be understood that the "first step, according to the instructions provided by the user, select the API that can be executed or provide information, and extract the key information in the instructions for API search for related information. The returned results include, API name: api, key information: arguments" in the above example can be the first sub-prompt information template, and the "if the API is unclear, recommend relevant annotation instructions to the user. If it is not unclear, directly return the API results" in the above example can be the second sub-prompt information template.

[0155] For example, see Figure 8 , Figure 8 The flowchart of the voice interaction method in certain embodiments of the present application is shown in FIG. That is, when the current voice request forwarded by the vehicle received by the server is "adjust the air conditioner", the server inputs "adjust the air conditioner" and the above-mentioned thought chain prompt information template into the large language model. When the large language model confirms that the API corresponding to "adjust the air conditioner" is unclear based on the first sub-prompt information template, the large language model can infer the recommended voice request corresponding to "adjust the air conditioner" (i.e., "related annotation instructions") based on the second sub-prompt information template.

[0156] Furthermore, in the case of inferred recommended voice requests such as "adjust the air conditioner to 20 degrees", "adjust the air volume to the third gear" and "open the air conditioner page", the large language model can complete and improve these recommended voice requests to form complete target guidance information, such as "I can't understand what you mean. You can say standard points, such as: adjust the air conditioner to 20 degrees, adjust the air volume to the third gear, open the air conditioner page".

[0157] Furthermore, after the target guidance information is sent to the vehicle, the vehicle can play the target guidance information to prompt the user to complete the adjustment of the current voice request "adjust the air conditioner" based on the target guidance information.

[0158] Optionally, it is understandable that in order to ensure the reliability of the large language model, the server of the implementation mode of the present application can perform model training based on the thinking chain prompt information template, samples and sample labels during the early training process of the large language model, and in the post-landing reasoning process of the large language model, the large language model performs corresponding reasoning work through the thinking chain prompt information template and voice request. It is understandable that the thinking chain prompt information template used in the early training process and the thinking chain prompt information template used in the post-landing reasoning process can be the same.

[0159] Thus, in the implementation mode of the present application, the large language model can confirm the correspondence between the current voice request and the vehicle control instruction based on the first sub-prompt information template, the second sub-prompt information template and the current voice request in the thought chain prompt information template, and when it is confirmed that the current voice request corresponds to the vehicle control instruction, confirm the target guidance information corresponding to the current voice request, thereby realizing the reasoning of the large language model based on the thought chain prompt information template, and thus the reasoning accuracy of the large language model can be guaranteed to a certain extent.

[0160] Optional, see Fig. 9 , Fig. 9 The flowchart of the voice interaction method in some embodiments of the present application is shown in FIG. That is, in order to ensure that the large language model can identify whether the current voice request corresponds to the vehicle control instruction, the embodiment of the present application is provided with corresponding downstream tasks and prompts, so that the large language model can distinguish between "voice requests corresponding to vehicle control instructions" and "voice requests not corresponding to vehicle control instructions" and perform natural language processing respectively.

[0161] Specifically, Fig. 9 As shown, the server can train the basic model based on the third voice request sample, the API label corresponding to the third voice request sample, and the prompt corresponding to the third voice request sample so that the basic model can distinguish between "voice requests with API label unclear" and "voice requests of other APIs other than label unclear".

[0162] It can be understood that the prompt corresponding to the third voice request sample is "If you are an intelligent voice assistant, you can select the API that can execute this information according to the instructions provided by the user, and extract the key information in the instruction for API search related information. The returned results include: API name: api, key information: arguments. Command: Help me open the car window. Output: {API:WindowsOpen, arguments:{drivce:car window}}". And, the "Command: Help me open the car windowOutput: {API:WindowsOpen, arguments:{drivce:car window}}" in the prompt can be understood as an example.

[0163] Furthermore, the server may train the basic model based on the fourth voice request sample, the vehicle control instruction label corresponding to the fourth voice request sample, and the prompt corresponding to the third voice request sample so that the basic model can infer the vehicle control instructions corresponding to the "voice requests of different APIs".

[0164] It can be understood that the prompt corresponding to the fourth voice request sample is "If you are an intelligent voice assistant, you can generate standard user instructions based on the function points, API and key information arguments provided by the user. The input is: {API: WindowsOpen, Arguments: {device: car window}}. Output: standard instruction: open the car window, open the car window." And, the prompt "The input is: {API: WindowsOpen, Arguments: {device: car window}}. Output: standard instruction: open the car window, open the car window" can be understood as an example.

[0165] In addition, the server can train the basic model based on the "fifth voice request sample that does not correspond to the vehicle control command", the recommended voice request label corresponding to the fifth voice request sample, and the prompt corresponding to the third voice request sample so that when the basic model faces a voice request with an API of unclear, it can infer the recommended voice request corresponding to the voice request (or in other words, a semantically similar voice request) (that is, Fig. 9 'Related Standard Directives' in the .

[0166] It can be understood that the prompt corresponding to the fifth voice request sample is "If you are an intelligent voice assistant, you can provide standard instructions to users based on the incomplete or unclear instructions provided by users, and provide users with choices. Input: car window. Output: standard instructions: open car window, close car window." And, the "Input: car window. Output: standard instructions: open car window, close car window" in the prompt can be understood as an example.

[0167] It can also be understood that after completing the training of the basic model based on the above-mentioned third voice request sample, fourth voice request sample and fifth voice request sample, and the labels and prompts corresponding to these three voice request samples respectively, the reference model in the implementation mode of the present application can be obtained.

[0168] Optionally, in certain embodiments of the present application, please refer to Fig.10 , Fig.10 This is a flowchart of the voice interaction method in some embodiments of the present application, and please refer to Figure 2 , 4 and 6.

[0169] That is, the server inputs the first voice request sample ("adjust the air conditioner") into the reference model for each first voice request sample received in advance, so that the reference model infers the guidance information according to the first voice request sample, thereby forming multiple candidate guidance information, such as Figure 2 res1, res2 and res3 in it.

[0170] Then, the server determines the sorting results of the plurality of candidate guide information according to the received sorting operation instruction, such as Figure 2 The order represented by "res3-res2-res1" in the file.

[0171] Then, the server determines the first candidate guidance information and the second candidate guidance information from the plurality of candidate guidance information according to the sorting result of the plurality of candidate guidance information, and constructs a candidate guidance information combination consisting of the first candidate guidance information and the second candidate guidance information. Figure 4 shown.

[0172] Afterwards, the server may train the reference model using the candidate guidance information combination, so that when the reference model is trained, the first prediction result (or probability) of the first candidate guidance information in the candidate guidance information combination can be higher than the second prediction result (or probability) of the second candidate guidance information in the combination predicted by the reference model. For details, please refer to Figure 6 shown.

[0173] Finally, when the predefined training termination conditions are met (such as the number of training rounds reaches a preset number, the loss function value converges, etc.), it can be confirmed that the reference model training is completed, thereby obtaining a large language model.

[0174] The present application also provides a computer-readable storage medium storing a computer program, which implements the above-mentioned voice interaction method when the computer program is executed by one or more processors.

[0175] In the description of this specification, the descriptions with reference to the terms "specifically", "further", "particularly", "understandably", etc. are intended to mean that the specific features, structures, materials or characteristics described in conjunction with the embodiments or examples are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms are not intended to refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, unless they are contradictory.

[0176] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code that includes one or more executable instructions for implementing the steps of a specific logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present application belong.

[0177] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A voice interaction method, characterized in that: include: Receive the current voice request forwarded by the vehicle; In a case where a vehicle control instruction corresponding to the current voice request cannot be determined according to the current voice request, determining target guidance information for guiding a user to adjust the current voice request according to a large language model and the current voice request, wherein the large language model is pre-trained and is capable of generating a plurality of candidate guidance information based on a first voice request sample, and determining a predicted guidance information based on a ranking result of the plurality of candidate guidance information; Feedback the target guidance information to guide the user to complete the voice interaction; The training steps of the large language model include: Determining the plurality of candidate guidance information according to the first voice request sample and a pre-trained reference model; Determining first candidate guidance information and second candidate guidance information among the plurality of candidate guidance information according to the sorting result; The reference model is trained according to a combination of the first candidate guidance information and the second candidate guidance information to obtain the large language model.

2. The method according to claim 1, characterized in that The step of training the reference model according to the combination of the first candidate guidance information and the second candidate guidance information to obtain the large language model includes: Determine, according to the combination and the reference model, a first prediction result of the reference model for first candidate guidance information in the combination and a second prediction result for second candidate guidance information in the combination; According to the difference between the first prediction result and the second prediction result, the reference model is trained to obtain the large language model.

3. The method according to claim 2, characterized in that The reference model includes a target model and a replica model of the target model, the difference degree includes a first difference degree corresponding to the target model and a second difference degree corresponding to the replica model, and training the reference model according to the difference degree between the first prediction result and the second prediction result to obtain the large language model includes: The target model is trained according to the first difference degree and the second difference degree to obtain the large language model.

4. The method according to claim 1, characterized in that: The training steps of the large language model include: The plurality of candidate guidance information are determined according to the first voice request sample, the reference model and a pre-configured prompt information template.

5. The method according to claim 1, characterized in that: The training steps of the large language model include: Acquire a second voice request sample and a guidance information tag corresponding to the second voice request sample; According to the second voice request sample and the guide information label, a predetermined basic model is trained to determine the reference model.

6. The method according to claim 1, characterized in that The method of determining target guidance information for guiding a user to adjust the current voice request according to the large language model and the current voice request when a vehicle control instruction corresponding to the current voice request cannot be determined according to the current voice request includes: In a case where a vehicle control instruction corresponding to the current voice request cannot be determined according to the current voice request, a recommended voice request corresponding to the current voice request is generated according to the large language model and the current voice request to determine the target guidance information.

7. The method according to claim 1, characterized in that The method of determining target guidance information for guiding a user to adjust the current voice request according to the large language model and the current voice request when a vehicle control instruction corresponding to the current voice request cannot be determined according to the current voice request includes: When the vehicle control instruction corresponding to the current voice request cannot be determined based on the large language model, the current voice request and the first sub-prompt information template in the pre-configured thought chain prompt information template, the target guidance information is determined according to the large language model, the current voice request and the second sub-prompt information template in the thought chain prompt information template.

8. A server, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the method according to any one of claims 1 to 7 is implemented.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by one or more processors, the method according to any one of claims 1 to 7 is implemented.

Images