Model training method, voice interaction method, server and storage medium

Abstract

The invention discloses a model training method, a voice interaction method, a server and a computer readable storage medium, and the method comprises the steps: generating a first function data set according to a vehicle control function name; and generating a first application scene information set associated with the first function data set according to the first function data set. And generating a second function data set according to the first application scene information set. And training the model according to the second function data set. Thus, scene mining processing and data enhancement processing are carried out on the vehicle control function name, the second function data set with diversified semantics and rich scenes is generated, the model is trained according to the second function data set, optimization of the model is achieved, generation from user requirements to user diversified instructions is completed according to the model, and user experience is improved. And moreover, the dependence on manual labeling can be reduced, so that the labeling cost is reduced, and the labeling manpower is saved.

Description

Technical Field

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

[0002] In the related art, the in-vehicle intelligent system is usually trained through the in-vehicle dialogue scene data to understand the user's demand information and complete the user's demand information. However, the existing in-vehicle intelligent system lacks corresponding feedback for different scenarios, and it is difficult to meet the diverse needs of users in different scenarios. Summary of the invention

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

[0004] The present application provides a model training method, which includes:

[0005] Generate a first function data set according to the vehicle control function name;

[0006] generating, according to the first function data set, a first application scenario information set associated with the first function data set;

[0007] Generate a second function data set according to the first application scenario information set;

[0008] The model is trained based on the second functional dataset.

[0009] In this way, the server generates a first function data set based on the vehicle control function name. Then, the server generates a first application scenario information set associated with the first function data set based on the first function data set. Then, the server generates a second function data set based on the first application scenario information set. Finally, the server trains the model based on the second function data set. In this way, the vehicle control function names are subjected to scene mining processing and data enhancement processing to generate a semantically diverse and scenario-rich second function data set, and the model is trained based on the second function data set to achieve model optimization, thereby completing the generation of user needs to user diversified instructions based on the model, and improving user experience. In addition, it can reduce reliance on manual annotation, thereby reducing annotation costs and saving annotation manpower.

[0010] In some embodiments, the first function data set includes a plurality of first function data, and generating the first function data set according to the vehicle control function name includes:

[0011] Based on a first preset large language model, generating a function name in the first function data according to the vehicle control function name;

[0012] The function description information in the first function data is generated according to a preset knowledge base and the function name.

[0013] In this way, the server generates the function name in the first function data according to the vehicle control function name based on the first preset large language model. Then, the server generates the function description information in the first function data according to the preset knowledge base and the function name. In this way, the server can generate a first function data set including the function name and the function description information, and the first function data set can generate a second function data set through a series of subsequent processing to train the model.

[0014] In some implementations, generating, according to the first function data set, a first application scenario information set associated with the first function data set includes:

[0015] Based on the second preset large language model, at least one first application scenario information is generated according to each of the first function data to obtain the first application scenario information set.

[0016] In this way, the server generates at least one first application scenario information based on each first function data based on the second preset large language model to obtain a first application scenario information set. In this way, through the autonomous association of the second preset large language model, corresponding user instructions in various scenarios can be generated, which enriches the diversity of instruction data and provides a more comprehensive data foundation for subsequent model training.

[0017] In some implementations, generating a second function data set according to the first application scenario information set includes:

[0018] Performing screening and filtering on the first application scenario information set to obtain a second application scenario information set, wherein the second application scenario information set includes at least one application scenario information, and the application scenario information includes a function name, scenario description information, environment information, and function operation information;

[0019] Generate the second function data set according to the second application scenario information set.

[0020] In this way, the server performs screening and filtering on the first application scenario information set to obtain a second application scenario information set, wherein the second application scenario information set includes at least one application scenario information, and the application scenario information includes a function name, a scenario description information, an environmental information, and a function operation information. Then, the server generates a second function data set based on the second application scenario information set. In this way, by screening and filtering the first application scenario information set, invalid and erroneous data is removed, the quality and relevance of the application scenario information are improved, and a more reliable data basis is provided for subsequent model training.

[0021] In some implementations, the filtering the first application scenario information set to obtain the second application scenario information set includes:

[0022] The first application scenario information set is screened and filtered based on preset tags to obtain the second application scenario information set, wherein the preset tags include preset invalid environment tags and preset invalid function operation tags.

[0023] In this way, based on the preset tags, the server performs screening and filtering on the first application scenario information set to obtain the second application scenario information set, and the preset tags include the preset invalid environment tags and the preset invalid function operation tags. In this way, by using the preset tags for screening, irrelevant or invalid scenario information can be removed, ensuring that the generated user instructions are more accurate and practical.

[0024] In some implementations, generating the second function data set according to the second application scenario information set includes:

[0025] Based on preset standard rules, standardization processing is performed according to the scene description information and the environment information to determine standard application scene information;

[0026] Generate the demand mapping information according to the scenario description information and the environment information;

[0027] The second functional data set is generated according to the standard application scenario information, the requirement mapping information and the functional operation information.

[0028] In this way, the server performs normalization processing based on the scene description information and environmental information based on preset normative rules to determine the standard application scenario information. Next, the server generates demand mapping information based on the scene description information and environmental information. Finally, the server generates a second functional data set based on the standard application scenario information, demand mapping information and functional operation information. In this way, through normalization processing, the consistency and comparability of the data are ensured, which facilitates model learning and generation of accurate output. In addition, the demand mapping information helps the model better understand the needs of users in different scenarios, thereby improving the accuracy of function recommendations.

[0029] In some implementations, generating demand mapping information according to the scenario description information and the environment information includes:

[0030] Performing annotation processing on the scene description information and the environment information to generate annotation information;

[0031] The annotation information is extracted to generate demand mapping information.

[0032] In this way, the server annotates the scene description information and environmental information to generate annotation information. Next, the server extracts the annotation information to generate demand mapping information. In this way, through annotation processing, key elements in the scene description and environmental information can be accurately identified and extracted, thereby improving the accuracy of subsequent processing. In addition, by extracting key elements and tags from the annotation information, demand mapping information is generated to help the model better understand the needs of users in different scenarios, thereby improving the accuracy of function recommendations.

[0033] The present application provides a voice interaction method, the method comprising:

[0034] Receive voice requests forwarded by the vehicle;

[0035] Based on the model trained by the above-mentioned model training method, generating a vehicle control instruction corresponding to the voice request according to the voice request;

[0036] The vehicle control command is sent to the vehicle to complete the voice interaction method.

[0037] In this way, the server receives the voice request forwarded by the vehicle. Then, based on the model trained by the above-mentioned model training method, the server generates a vehicle control instruction corresponding to the voice request according to the voice request. Finally, the server sends the vehicle control instruction to the vehicle to complete the voice interaction method. In this way, the trained model can quickly and accurately understand the user's voice request and generate corresponding vehicle control instructions, thereby improving the response speed and accuracy of the system.

[0038] An embodiment of the present application provides a server, which includes a processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the above method is implemented.

[0039] The embodiment of the present application provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps of the above method are implemented.

[0040] 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

[0041] 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:

[0042] Figure 1 It is one of the flowcharts of the model training method of certain embodiments of the present application;

[0043] Figure 2 is a schematic diagram of a flow chart for generating model training data in certain embodiments of the present application;

[0044] Figure 3 This is the second flow chart of the model training method of certain embodiments of the present application;

[0045] Figure 4 This is the third flow chart of the model training method of certain implementation modes of the present application;

[0046] Figure 5 This is the fourth flow chart of the model training method of certain embodiments of the present application;

[0047] Figure 6 This is the fifth flow chart of the model training method of certain embodiments of the present application;

[0048] Figure 7 This is the sixth flow chart of the model training method of certain embodiments of the present application;

[0049] Figure 8 This is the seventh flow chart of the model training method of certain implementation modes of the present application;

[0050] Fig. 9 It is a flowchart of the voice interaction method of certain embodiments of the present application. DETAILED DESCRIPTION

[0051] 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.

[0052] In current in-vehicle intelligent system technology, in order to enable the system to understand and respond to user needs, it is usually trained using in-vehicle dialogue scenario data. This training method involves using a large amount of in-vehicle dialogue scenario data, which includes examples of users interacting with the vehicle in a variety of different scenarios. By analyzing this data, the in-vehicle intelligent system can learn the user's demand patterns, so that when the user makes a request, it can provide corresponding services or operations.

[0053] However, existing in-vehicle intelligent systems often have certain limitations when responding to user needs in different scenarios. Specifically, these systems may lack in-depth understanding and adaptive feedback for specific scenarios. This means that when the user is in a new scenario that is different from the training data, the in-vehicle intelligent system may not be able to accurately identify the user's needs or provide the most appropriate response.

[0054] For example, when a user is driving in snowy weather, the road is slippery and covered with snow, and the user proposes a user demand of "help me improve the stability of the vehicle", but the in-vehicle intelligent system may only use the training data of "scenario name: rough road; user demand: improve vehicle stability; function operation: ['raise suspension']". The in-vehicle intelligent system will directly perform the function operation of "raise suspension", but in the snowy weather scenario, in order to ensure safety, the function operation to be performed is "switch the driving mode to escape mode and raise the suspension".

[0055] Based on the above questions, please refer to Figure 1 , the embodiment of the present application provides a model training method, the method comprising:

[0056] 01: Generate the first function data set according to the vehicle control function name;

[0057] 02: Generate a first application scenario information set associated with the first function data set according to the first function data set;

[0058] 03: Generate a second function data set according to the first application scenario information set;

[0059] 04: Train the model according to the second functional data set.

[0060] The embodiment of the present application also provides a server, including 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 generate a first function data set according to the vehicle control function name. And based on the first function data set, a first application scenario information set associated with the first function data set is generated. The processor is also used to generate a second function data set according to the first application scenario information set. And the model is trained according to the second function data set.

[0061] The embodiment of the present application also provides a model training device. The voice interaction method of the embodiment of the present application can be implemented by the model training device of the embodiment of the present application. Specifically, the model training device includes a generation module and a training module. The generation module is used to generate a first function data set according to the vehicle control function name. The generation processing module is also used to generate a first application scenario information set associated with the first function data set according to the first function data set. The generation processing module is also used to generate a second function data set according to the first application scenario information set. The training module is used to train the model according to the second function data set.

[0062] Specifically, the vehicle control function name refers to the names of various controls and function names on the vehicle, which can be used to describe and identify the different controls and functions of the vehicle so that users and the vehicle system can understand and interact. The control name refers to the name of the physical or virtual elements used to interact with the vehicle, including the steering wheel, accelerator pedal, brake pedal, turn signal lever, center console buttons, icons on the touch screen, etc. The function name refers to the name of the function or service that the control can achieve, including automatic emergency braking (AEB), adaptive cruise control (ACC), lane keeping assist (LKA), automatic air conditioning, navigation system, in-vehicle entertainment system, etc.

[0063] See also Figure 2 , Figure 2 1 is a schematic diagram of a flow chart for generating model training data in certain embodiments of the present application. In which, the server first obtains the names of vehicle controls and functions, such as "driving mode", "air purification", etc.

[0064] Then, based on the first preset large language model, the server generates a first function data set according to the vehicle control function name. The first function data set includes a plurality of first function data.

[0065] Next, based on the second preset large language model, the server generates a first application scenario information set associated with the first function data set according to the first function data set, such as "in a city traffic jam", "on a long journey", "driving on a rugged mountain road", and "driving in the snow".

[0066] Then, the server generates a second functional data set based on the first application scenario information set. Finally, the server trains the model based on the second functional data set so that it can understand the needs of users in different scenarios and generate diversified instructions.

[0067] In summary, in the model optimization method and server provided in the implementation manner of the present application, the server generates a first function data set based on the vehicle control function name. Next, the server generates a first application scenario information set associated with the first function data set based on the first function data set. Then, the server generates a second function data set based on the first application scenario information set. Finally, the server trains the model based on the second function data set. In this way, the vehicle control function names are subjected to scene mining processing and data enhancement processing to generate a semantically diverse and scenario-rich second function data set, and the model is trained based on the second function data set to optimize the model, thereby completing the generation of user needs to user diversified instructions based on the model, and improving the user experience. In addition, it can reduce the reliance on manual annotation, thereby reducing annotation costs and saving annotation manpower.

[0068] See also Figure 3 In some embodiments, the first function data set includes a plurality of first function data, and step 01 (generating the first function data set according to the vehicle control function name) includes:

[0069] 011: Generate a function name in the first function data based on the first preset large language model and the function name of the vehicle control;

[0070] 012: Generate function description information in the first function data according to the preset knowledge base and the function name.

[0071] In some embodiments, the generating module is further used to generate the function name in the first function data according to the vehicle control function name based on the first preset large language model, and to generate the function description information in the first function data according to the preset knowledge base and the function name.

[0072] In some embodiments, the processor is further configured to generate a function name in the first function data according to the vehicle control function name based on the first preset large language model, and to generate function description information in the first function data according to the preset knowledge base and the function name.

[0073] Specifically, the first preset large language model refers to a pre-trained large language model (Large Language Model, LLM), which can understand and process vehicle control function names and generate natural language text describing these vehicle control function names.

[0074] The preset knowledge base refers to a database or knowledge collection that includes vehicle function description information, vehicle user manuals and other related knowledge. In some embodiments, the preset knowledge base also includes technical information such as vehicle configuration lists and vehicle technical parameters.

[0075] The server generates a function name in the first function data based on the first preset large language model and the vehicle control function name. Then, the server generates function description information in the first function data based on the preset knowledge base and the function name.

[0076] The following is an example to illustrate the model training method of the embodiment of the present application. First, the server provides the vehicle control function name to the first preset large language model. For example, automatic anti-collision emergency braking system, adaptive cruise control system, lane keeping assist and departure warning system, air purifier and driving mode, etc.

[0077] Subsequently, the large language model processes the received vehicle control function names to obtain corresponding function names. For example, the function name corresponding to the automatic anti-collision emergency braking system is automatic emergency braking, the function name corresponding to the adaptive cruise control system is adaptive cruise, the function name corresponding to the air purifier is air purification, and the function name corresponding to the driving mode is driving mode.

[0078] Next, the server generates the function description information in the first function data set based on the preset knowledge base and function information, and the function description information provides a detailed explanation of each function. For example, for "air purification", the function description information may explain how it is applicable to the situation where PM2.5 pollution exceeds the standard in the car, and how it quickly filters fine particles in the air.

[0079] In some embodiments, the first function data in the generated first function data set may be as follows: "{'Function':'Air purification', 'Description':'Applicable to scenarios where PM2.5 pollution in the car exceeds the standard and the air needs to be purified. When turned on, it can quickly filter fine particles in the air in the car, eliminate air pollution, and provide a healthy breathing environment for travel.'}", or "{'Function':'Driving mode', 'Description':'Driving modes include standard, comfort, sports, ejection, geek, escape, and trailer RV modes. Standard or comfortable can be used for daily driving. Users can select the driving mode by turning on, switching, and turning off.}".

[0080] In this way, the server can generate a first function data set including a function name and function description information, and the first function data set can generate a second function data set through a subsequent series of processing to train the model.

[0081] See also Figure 4 In some implementations, step 02 (generating a first application scenario information set associated with the first function data set according to the first function data set) includes:

[0082] 021: Based on the second preset large language model, generate at least one first application scenario information according to each first function data to obtain a first application scenario information set.

[0083] In some embodiments, the generating module is further used to generate at least one first application scenario information based on the second preset large language model and according to each first function data to obtain a first application scenario information set.

[0084] In some embodiments, the processor is further configured to generate at least one first application scenario information based on each first function data based on the second preset large language model to obtain a first application scenario information set.

[0085] Specifically, the second preset large language model refers to the first application scenario information related to the vehicle control function. The second preset large language model is similar to the first preset large language model and can also understand and generate natural language text, but it focuses on extracting and generating information related to the application scenario from the first function data. It should be noted that the second preset large language model and the first preset large language model can be of the same scale or of different scales.

[0086] The first application scenario information refers to a set of first application scenario information generated based on the second preset large language model according to the first function data in the first function data set, including descriptive information of specific application scenarios related to the vehicle control function. The first application scenario information set includes multiple first application scenario information, each of which includes a first function name, first scenario description information, first environment information, and first function operation information.

[0087] Based on the second preset large language model (LLM), the server analyzes each first function data and generates at least one first application scenario information related thereto. For example, for the "air purification" function, the first application scenario information that may be generated includes "the air quality in the car decreases when urban traffic is congested", "dust enters the car when passing a construction site", or "suitable for scenarios where PM2.5 pollution in the car exceeds the standard and the air needs to be purified. After turning it on, it can quickly filter fine particles in the air in the car, eliminate air pollution, and provide a healthy breathing environment for travel." For the "driving mode" function, the application scenario information that may be generated includes "the road conditions in the snow are very bad, and the user wants to drive the vehicle out of the snow" or "the user is driving on a mountain road and needs to improve the vehicle's passability and suspension stability."

[0088] In some implementations, the prompt of the second preset large language model is constructed as follows:

[0089] Based on the description of a certain function in the car that I provided, 1. Combined with environmental factors such as weather, temperature, and time, list the situations or scenarios in which the user needs to use this function, and output the possible corresponding adjustment or switch operations. Please note that the scene description can be diverse. Please output the results of at least 5 scenes. Output format: {'Scene description': ", 'Environment': ", 'Function operation': ["]}

[0090] For example: [input]: {'Function': 'Driving Mode', 'Description': 'Driving modes include standard, comfort, sport, ejection, geek, escape, and trailer. Standard or comfort can be used for daily driving. Users can select the driving mode by turning on, switching on, and turning off.}

[0091] [Output]:{"Scene description":"The road conditions are very bad in the snow, and the user wants to drive the vehicle out of the snow","Environment":"[Snowy day","Winter"],"Function operation":["Switch the driving mode to escape mode","Raise the suspension"]}

[0092] Please answer, and make sure to list as many output scenarios as possible, and output the json format file:

[0093] It should be noted that after the prompt is constructed, the temperature coefficient and sampling method of the large language model are continuously adjusted to perform multiple call reasoning of the large language model to output different [descriptions | environments | function operations] for the same function.

[0094] In this way, through the autonomous association of the second preset large language model, corresponding user commands in a variety of scenarios can be generated, enriching the diversity of command data and providing a more comprehensive data foundation for subsequent model training.

[0095] See also Figure 5 In some implementations, step 03 (generating a second function data set according to the first application scenario information set) includes:

[0096] 031: Screen and filter the first application scenario information set to obtain a second application scenario information set;

[0097] 032: Generate a second function data set according to the second application scenario information set.

[0098] In some embodiments, the generating module is further used to perform screening and filtering on the first application scenario information set to obtain a second application scenario information set, and to generate a second function data set according to the second application scenario information set.

[0099] In some embodiments, the processor is further configured to perform screening and filtering on the first application scenario information set to obtain a second application scenario information set, and to generate a second function data set according to the second application scenario information set.

[0100] Specifically, the screening and filtering process refers to reviewing and selecting the first application scenario information set to extract the most relevant and useful information, remove invalid or vehicle-unavailable information, and generate a second application scenario information set. It should be noted that, compared with the first application scenario information set, the content of the scenario information that has not been removed in the second application scenario information set is consistent with that in the first application scenario information set.

[0101] The server performs screening and filtering on the first application scenario information set to remove irrelevant or duplicate information and ensure the accuracy and relevance of the information, including removing ambiguous scenario descriptions, merging similar scenarios, and removing functional operations that are not in line with reality.

[0102] Subsequently, through screening and filtering processing, the server obtains a second application scenario information set, the second application scenario information set includes at least one application scenario information, and each scenario information includes a function name, scenario description information, environment information and function operation information.

[0103] Next, the server generates a second functional data set according to the second application scenario information set, and the second functional data set can be used to train the model.

[0104] In this way, by screening and filtering the first application scenario information set, invalid and erroneous data can be removed, the quality and relevance of the application scenario information can be improved, and a more reliable data basis can be provided for subsequent model training.

[0105] See also Figure 6 In some implementations, step 031 (screening and filtering the first application scenario information set to obtain a second application scenario information set) includes:

[0106] 0311: Based on the preset tag, the first application scenario information set is screened and filtered to obtain a second application scenario information set.

[0107] In some implementations, the generation module is used to perform screening and filtering processing on the first application scenario information set based on preset tags to obtain the second application scenario information set.

[0108] In some implementations, the processor is further configured to perform screening and filtering processing on the first application scenario information set based on a preset tag to obtain a second application scenario information set.

[0109] Specifically, the preset tag refers to a standard or classification used to screen and filter the first application scenario information set. The preset tag usually includes a preset invalid environment tag and a preset invalid function operation tag, which can identify and exclude scene information that does not meet specific standards or is not important to the operation of the in-vehicle intelligent system.

[0110] The preset invalid environment labels can be used to identify those environmental factors that have no effect on or are irrelevant to the operation of vehicle functions. For example, if the vehicle control function is irrelevant to weather conditions, the preset invalid environment labels may include weather-related labels such as "sunny" and "rainy".

[0111] The preset invalid function operation labels can be used to identify those operations that have no effect on the vehicle function operation or are irrelevant to the vehicle function operation. For example, if the vehicle function is irrelevant to the vehicle speed, the preset invalid function operation labels may include labels related to the vehicle speed, such as "high speed driving" and "low speed driving".

[0112] The server uses these preset tags to filter the first application scenario information set, that is, to check the scene description information, environment information and function operation information in each first application scenario information, and remove those containing preset invalid tags. For example, if the scene description contains "nighttime", and "nighttime" is one of the preset invalid environment tags, then the first application scenario information will be filtered out.

[0113] Through screening and filtering, the server obtains a second application scenario information set. This second application scenario information set includes screened application scenario information that does not contain preset invalid tags, and is more concise and relevant. This second application scenario information set can help the system understand how to use the vehicle's functions to meet user needs in different driving environments and situations.

[0114] In this way, by using preset tags for filtering, irrelevant or invalid scene information can be removed, ensuring that the generated user instructions are more accurate and practical.

[0115] See also Figure 7 In some implementations, step 032 (generating a second function data set according to the second application scenario information set) includes:

[0116] 0321: Based on preset standard rules, standardize the scene description information and environmental information to determine the standard application scene information;

[0117] 0322: Generate requirement mapping information based on scenario description information and environment information;

[0118] 0323: Generate a second functional data set according to the specification application scenario information, requirement mapping information and functional operation information.

[0119] In some embodiments, the generation module is further used to perform normalization processing based on the scene description information and the environment information based on the preset normative rules to determine the normative application scene information, and to generate the requirement mapping information based on the scene description information and the environment information, and to generate the second function data set based on the normative application scene information, the requirement mapping information and the function operation information.

[0120] In some embodiments, the processor is further configured to perform normalization processing based on the scene description information and the environment information based on the preset normative rules to determine the normative application scene information, and to generate the requirement mapping information based on the scene description information and the environment information, and to generate the second function data set based on the normative application scene information, the requirement mapping information and the function operation information.

[0121] Specifically, preset specification rules refer to predefined rules or standards that are used to guide how to format and standardize scene description information and environmental information, ensure the uniformity and consistency of information, and enable the in-vehicle intelligent system to understand and process this information more accurately.

[0122] Normalization processing refers to the process of organizing and converting information according to the above-mentioned preset standard rules.

[0123] Based on preset normative rules, the server normalizes the scene description information and environmental information to obtain standardized application scene information, so as to convert the original information into a unified format for subsequent processing and analysis. For example, all time information is converted to a 24-hour system, temperature information is converted to Celsius or Fahrenheit, speed information is converted to kilometers per hour or miles per hour, etc. Alternatively, non-standard typesetting is optimized to make it standardized.

[0124] Next, the server generates demand mapping information based on the scenario description information and environment information. The demand mapping information associates the user's actual needs with the vehicle function operation, and can know that the model understands the vehicle functions that the user may need in a specific scenario.

[0125] Finally, the server generates a second functional data set based on the standard application scenario information, requirement mapping information and functional operation information. The second functional data set includes descriptions of user requirements and related functional operations for different scenarios and can be used to train the model.

[0126] Please refer to Table 1 and Table 2. Table 1 is an example of part of the content in the second application scenario information set, and Table 2 is an example of part of the content in the second function data set.

[0127] Table 1

[0128]

[0129] Table 2

[0130]

[0131] Among them, the content in Table 2 is obtained according to Table 1.

[0132] In this way, the consistency and comparability of data are ensured through normalization, which facilitates model learning and generates accurate output. In addition, demand mapping information helps the model better understand user needs in different scenarios, thereby improving the accuracy of function recommendations.

[0133] See also Figure 8 In some implementations, step 0322 (generating requirement mapping information according to the scene description information and the environment information) includes:

[0134] 03221: Annotate the scene description information and environment information to generate annotation information;

[0135] 03222: Extract annotation information and generate requirement mapping information.

[0136] In some implementations, the generation module is used to perform annotation processing on the scene description information and the environment information to generate annotation information, and to extract the annotation information to generate demand mapping information.

[0137] In some implementations, the processor is further configured to perform annotation processing on the scene description information and the environment information to generate annotation information, and to extract the annotation information to generate demand mapping information.

[0138] Specifically, the annotation process refers to the process of marking or annotating the scene description information and the environment information so that the server can better understand and process the information. In some embodiments, the scene description information and the environment information can be annotated manually.

[0139] The server annotates the scene description information and environmental information. That is, the server marks the key elements in the scene description (such as time, location, activity, etc.) and environmental information (such as weather, temperature, lighting, etc.). For example, the scene description "the air quality in the car decreases when the city traffic is congested" may be annotated as {time: rush hour, location: city road, activity: vehicle driving}, and the environmental information "the weather is clear and the temperature is moderate" may be annotated as {weather: clear, temperature: moderate}.

[0140] Through the annotation processing, the server generates annotation information containing tags. Subsequently, the server extracts the annotation information and generates demand mapping information. The demand mapping information converts the key elements and tags in the annotation information into a description of the user's needs. For example, if the annotation information indicates that it is night driving and visibility is low, the demand mapping information may indicate that the user needs "enhanced lighting".

[0141] In this way, through annotation processing, key elements in scene descriptions and environmental information can be accurately identified and extracted, thereby improving the accuracy of subsequent processing. In addition, by extracting key elements and tags from the annotation information, demand mapping information is generated to help the model better understand user needs in different scenarios, thereby improving the accuracy of function recommendations.

[0142] See also Fig. 9 In some embodiments, the present application provides a voice interaction method, the method comprising:

[0143] 05: Receive the voice request forwarded by the vehicle;

[0144] 06: Based on the model trained by the above-mentioned model training method, a vehicle control instruction corresponding to the voice request is generated according to the voice request;

[0145] 07: Send vehicle control instructions to the vehicle to complete the voice interaction method.

[0146] The embodiment of the present application also provides a server, including 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 a voice request forwarded by a vehicle. And based on the model trained by the above-mentioned model training method, a vehicle control instruction corresponding to the voice request is generated according to the voice request. And the vehicle control instruction is sent to the vehicle to complete the voice interaction method.

[0147] The embodiment of the present application also 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 generating module and a sending module. The receiving module is used to receive a voice request forwarded by a vehicle. The generating processing module is also used to generate a vehicle control instruction corresponding to the voice request based on the model trained by the above-mentioned model training method according to the voice request. The sending processing module is also used to send the vehicle control instruction to the vehicle to complete the voice interaction method.

[0148] Specifically, suppose a driver says, "The road feels a bit slippery and not easy to drive on" while driving in snowy winter. The vehicle will receive this voice request and forward it to the server.

[0149] The server uses a pre-trained model to analyze the voice request. The model recognizes that the driver's need is to "improve the stability of the vehicle" and knows to drive in snowy winter. Based on the above analysis, the server generates the corresponding vehicle control command "Switch the driving mode to escape mode and raise the suspension", and sends the vehicle control command to complete the voice interaction process.

[0150] In this way, the trained model can quickly and accurately understand the user's voice requests and generate corresponding vehicle control instructions, improving the system's response speed and accuracy.

[0151] The present application also provides a computer-readable storage medium. When the computer program is executed by one or more processors, the one or more processors execute the method of the present application.

[0152] It is understood that a computer program includes computer program code. The computer program code may be in source code form, object code form, executable file or some intermediate form. Computer readable storage media may include: any entity or device capable of carrying computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), and software distribution medium.

[0153] 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.

[0154] 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.

[0155] 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 model training method, characterized in that: The method comprises: Generate a first function data set according to the vehicle control function name; generating, according to the first function data set, a first application scenario information set associated with the first function data set; Generate a second function data set according to the first application scenario information set; The model is trained based on the second functional dataset.

2. The method according to claim 1, characterized in that The first function data set includes a plurality of first function data, and the first function data set is generated according to the vehicle control function name, including: Based on a first preset large language model, generating a function name in the first function data according to the vehicle control function name; The function description information in the first function data is generated according to a preset knowledge base and the function name.

3. The method according to claim 2, characterized in that The step of generating, according to the first functional data set, a first application scenario information set associated with the first functional data set includes: Based on the second preset large language model, at least one first application scenario information is generated according to each of the first function data to obtain the first application scenario information set.

4. The method according to claim 3, characterized in that The step of generating a second function data set according to the first application scenario information set includes: Performing screening and filtering on the first application scenario information set to obtain a second application scenario information set, wherein the second application scenario information set includes at least one application scenario information, and the application scenario information includes a function name, scenario description information, environment information, and function operation information; Generate the second function data set according to the second application scenario information set.

5. The method according to claim 4, characterized in that The filtering and screening of the first application scenario information set to obtain the second application scenario information set includes: The first application scenario information set is screened and filtered based on preset tags to obtain the second application scenario information set, wherein the preset tags include preset invalid environment tags and preset invalid function operation tags.

6. The method according to claim 4, characterized in that The step of generating the second function data set according to the second application scenario information set includes: Based on preset standard rules, standardization processing is performed according to the scene description information and the environment information to determine standard application scene information; Generate demand mapping information according to the scenario description information and the environment information; The second functional data set is generated according to the standard application scenario information, the requirement mapping information and the functional operation information.

7. The method according to claim 6, characterized in that The generating demand mapping information according to the scenario description information and the environment information includes: Performing annotation processing on the scene description information and the environment information to generate annotation information; The annotation information is extracted to generate the demand mapping information.

8. A voice interaction method, characterized in that: The method comprises: Receive voice requests forwarded by the vehicle; Based on the model trained by the model training method according to any one of claims 1 to 7, generating a vehicle control instruction corresponding to the voice request according to the voice request; The vehicle control command is sent to the vehicle to complete the voice interaction method.

9. A server, characterized in that: The server includes a processor and a memory, wherein a computer program is stored in the memory. When the computer program is executed by the processor, the method according to any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images