Vehicle end control model optimization method and device, equipment and medium

Abstract

The invention discloses an optimization method and device for a vehicle end control model, equipment and a medium, and relates to the technical field of vehicles, and the method comprises the steps: obtaining the voice information of a user; identifying the voice information to obtain text information; inputting the text information into a vehicle end control model to obtain a vehicle end identification result, and inputting the text information into a cloud end control model to obtain a cloud end identification result; judging whether the vehicle end recognition result is consistent with the cloud end recognition result or not to obtain a judgment result; and if the judgment result represents that the vehicle end recognition result is inconsistent with the cloud end recognition result, optimizing the vehicle end control model according to the text information, the cloud end recognition result and the vehicle end recognition result. According to the method, the vehicle end control model can be automatically optimized, so that the optimization efficiency is improved.

Description

Technical Field

[0001] The present application relates to the technical field of vehicles, and particularly to an optimization method, device, equipment and medium for a vehicle-end control model. Background Art

[0002] With the development of vehicle technology, vehicles are equipped with intelligent cockpit systems that support users to perform voice control on the vehicles. For example, voice control of navigation, voice control of song switching, voice control of air conditioning, voice control of windows, etc. The implementation of the above-mentioned voice control-related functions mostly depends on the voice control model deployed on the vehicle terminal. The voice control model of the vehicle terminal may have inaccurate recognition, so it is necessary to optimize this voice control model.

[0003] Currently, the optimization schemes for end-side models are all to manually collect problems of the end-side model, reconstruct a new data set, and then perform targeted optimization on the end-side model. The optimization process is time-consuming, laborious and cumbersome, and the optimization efficiency is poor. Summary of the Invention

[0004] The present application provides an optimization method, device, equipment and medium for a vehicle-end control model, which can automatically optimize the vehicle-end control model, thereby improving the optimization efficiency.

[0005] To achieve the above object, the present application adopts the following technical solutions: In a first aspect, the present application provides an optimization method for a vehicle-end control model, including: Obtaining voice information of a user; Identifying the voice information to obtain text information; Inputting the text information into the vehicle-end control model to obtain a vehicle-end recognition result, and inputting the text information into the cloud control model to obtain a cloud recognition result; Judging whether the vehicle-end recognition result is consistent with the cloud recognition result to obtain a judgment result; If the judgment result indicates that the vehicle-end recognition result is inconsistent with the cloud recognition result, optimizing the vehicle-end control model according to the text information, the cloud recognition result and the vehicle-end recognition result.

[0006] Optionally, judging whether the vehicle-end recognition result is consistent with the cloud recognition result to obtain a judgment result includes: If the vehicle-end recognition result is a vehicle control instruction and the cloud recognition result is a rejection recognition instruction, or the vehicle-end recognition result is a rejection recognition instruction and the cloud recognition result is a vehicle control instruction, obtaining a judgment result that the vehicle-end recognition result is inconsistent with the cloud recognition result; If the vehicle-end recognition result is a vehicle control instruction and the cloud-end recognition result is a vehicle control instruction, or, if the vehicle-end recognition result is a rejection recognition instruction and the cloud-end recognition result is a rejection recognition instruction, a judgment result that the vehicle-end recognition result is consistent with the cloud-end recognition result is obtained.

[0007] Optionally, optimizing the vehicle-end control model according to the text information, the cloud-end recognition result, and the vehicle-end recognition result includes: Adding the text information, the corresponding cloud-end recognition result, and the vehicle-end recognition result to an optimization sample data set; Based on semantic similarity, clustering the text information in the optimization sample data set to obtain multiple categories; In the case where there is a target category with a sample quantity greater than a quantity threshold among the multiple categories, optimizing the vehicle-end control model by using the target text information in the target category, the corresponding target vehicle-end recognition result, and the target cloud-end recognition result.

[0008] Optionally, optimizing the vehicle-end control model by using the target text information in the target category, the corresponding target vehicle-end recognition result, and the target cloud-end recognition result includes: Inputting each target text information in the target category into the cloud-end control model to obtain the corresponding target cloud-end recognition result for each target text information; In the case where the target cloud-end recognition results corresponding to each target text information are all consistent, optimizing the vehicle-end control model by using the target text information, the target vehicle-end recognition result, and the target cloud-end recognition result.

[0009] Optionally, the method further includes: In the case where there are inconsistencies in the target cloud-end recognition results corresponding to each target text information, obtaining the quantities corresponding to the respective target cloud-end recognition results; Determining the text information corresponding to the recognition result with the largest quantity among the multiple target cloud-end recognition results as the remaining text information; Optimizing the vehicle-end control model by using the remaining text information, the corresponding remaining cloud-end recognition result, and the remaining vehicle-end recognition result.

[0010] Optionally, after optimizing the vehicle-end control model, the method further includes: Obtaining test samples; Testing the accuracy rate of the optimized vehicle-end control model by using the test samples; When the accuracy rate is greater than the accuracy rate threshold, deploy the optimized vehicle - end control model to the vehicle end.

[0011] Optionally, optimizing the vehicle - end control model includes: Optimize the vehicle - end control model at a preset period.

[0012] In a second aspect, the present application provides an optimization device for a vehicle - end control model, including: An acquisition module, configured to acquire the voice information of a user; An identification module, configured to identify the voice information to obtain text information; input the text information into the vehicle - end control model to obtain a vehicle - end identification result, and input the text information into the cloud - end control model to obtain a cloud - end identification result; A judgment module, configured to judge whether the vehicle - end identification result is consistent with the cloud - end identification result to obtain a judgment result; An optimization module, configured to, if the judgment result indicates that the vehicle - end identification result is inconsistent with the cloud - end identification result, optimize the vehicle - end control model according to the text information, the cloud - end identification result, and the vehicle - end identification result.

[0013] In a third aspect, the present application provides a computing device, including a memory and a processor; Wherein, one or more computer programs are stored in the memory, and the one or more computer programs include instructions; when the instructions are executed by the processor, the computing device is enabled to execute the method according to any one of the first aspect.

[0014] In a fourth aspect, the present application provides a computer - readable storage medium, which is used to store a computer program, and the computer program is used to execute the method according to any one of the first aspect.

[0015] It can be seen from the above technical solutions that the present application has at least the following beneficial effects: The present application provides an optimization method for a vehicle - end control model. The method includes: acquiring the voice information of a user, then converting it into text information, and respectively inputting it into the vehicle - end control model and the cloud - end control model to obtain a vehicle - end identification result and a cloud - end identification result respectively, and then comparing whether the vehicle - end identification result is consistent with the cloud - end identification result. If they are inconsistent, it means that there is a problem with the vehicle - end identification result and the vehicle - end control model needs to be updated. Then, use the text information, the vehicle - end identification result, and the cloud - end identification result to optimize the vehicle - end control model, thereby realizing the automatic optimization of the vehicle - end control model. Compared with the traditional optimization scheme, the optimization scheme of the present application does not require manual collection of end - side data and participation in the model optimization process. Therefore, this method can improve the optimization efficiency.

[0016] It should be understood that the description of technical features, technical solutions, beneficial effects or similar language in this application does not imply that all features and advantages can be achieved in any single embodiment. On the contrary, it can be understood that the description of features or beneficial effects means that specific technical features, technical solutions or beneficial effects are included in at least one embodiment. Therefore, the description of technical features, technical solutions or beneficial effects in this specification does not necessarily refer to the same embodiment. Furthermore, the technical features, technical solutions and beneficial effects described in this embodiment can be combined in any appropriate manner. Those skilled in the art will understand that an embodiment can be implemented without one or more specific technical features, technical solutions or beneficial effects of a specific embodiment. In other embodiments, additional technical features and beneficial effects can also be identified in specific embodiments that do not embody all embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flowchart of an optimization method for a vehicle-end control model provided by an embodiment of this application; Figure 2 It is a schematic diagram of an optimization device for a vehicle-end control model provided by an embodiment of this application; Figure 3 It is a schematic diagram of a computing device provided by an embodiment of this application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] The terms "first", "second", "third", etc. in the specification and drawings of this application are used to distinguish different objects, rather than to limit a specific order.

[0019] In the embodiments of this application, words such as "exemplary" or "for example" are used to represent examples, illustrations or explanations. Any embodiment or design solution described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or having more advantages than other embodiments or design solutions. Rather, the use of words such as "exemplary" or "for example" is intended to present relevant concepts in a specific manner.

[0020] Currently, the voice assistants of the intelligent cockpit system of vehicles are divided into two parts: the end side and the cloud side. The end side and the cloud side work together. The resources of the end side are limited, and the resources of the cloud side are rich, but they are affected by the network. Therefore, the end side has some simple functions and can give responses quickly, while the responses given by the cloud side are slow, but more accurate and perfect.

[0021] Currently, the optimization method of the end-side model is to collect problems of the end-side model and construct a data set manually, and the update process is cumbersome and requires manual operation by users. Currently, there is a lack of a solution for automatically optimizing the end-side model.

[0022] In view of this, an optimization method for a vehicle-end control model is provided in an embodiment of the present application. This method can be applied to a processing device, which can be a terminal or a server. The terminal includes but is not limited to a smart phone, a tablet computer, a laptop computer, a personal digital assistant, or a smart wearable device, etc. The server can be a cloud server, such as the central server in a central cloud computing cluster or the edge server in an edge cloud computing cluster. Of course, the server can also be a server in a local data center. The local data center refers to a data center directly controlled by a user. Compared with the traditional optimization scheme, the optimization scheme of the present application does not require manual collection of end-side data and participation in the model optimization process. Therefore, this method can improve the optimization efficiency.

[0023] In order to make the technical solution of the present application clearer and easier to understand, the technical solution of the present application will be introduced below with reference to the accompanying drawings. As Figure 1 shown, this figure is a flowchart of an optimization method for a vehicle-end control model provided in an embodiment of the present application. This method includes: S101. The processing device obtains the voice information of the user.

[0024] The user can be a vehicle occupant, for example, a driver or a passenger. The voice information of the user can be the content spoken by the user, for example, "turn on the air conditioner". In some embodiments, the user can first wake up the vehicle's voice assistant and then say voice information such as "turn on the air conditioner". Then the vehicle can obtain the voice information of the user through a sound pickup device such as a microphone and transmit the voice information to the processing device. Then the processing device can obtain the voice information of the user.

[0025] S102. The processing device identifies the voice information to obtain text information.

[0026] In some examples, the processing device can first obtain a voice recognition model, which can be a voice recognition model deployed on the vehicle end. The processing device inputs the voice information of the user into the voice recognition model for recognition to obtain the text information corresponding to the voice information.

[0027] S103. The processing device inputs the text information into the vehicle-end control model to obtain a vehicle-end recognition result.

[0028] The vehicle-end control model is a control model deployed at the vehicle end to identify the instructions issued by the user. For example, it identifies the control instructions corresponding to the above text information. To make the environment variables the same, the processing device needs to obtain the same vehicle-end control model as that of the vehicle end. That is to say, the vehicle-end control model into which the processing device inputs the text information is the same as the vehicle-end control model deployed on the vehicle. After the processing device obtains the text information, it can input the text information to the vehicle-end control model to obtain the vehicle-end recognition result.

[0029] Among them, the vehicle-end recognition result includes two cases. The first case is the vehicle control instruction, and the second case is the rejection recognition instruction. Among them, the vehicle control instruction refers to the instruction used to control the related functions of the vehicle, such as the air-conditioning function, window function, seat function, etc. The rejection recognition instruction refers to the instruction that rejects the response to the user's control. The vehicle-end control model will not respond to non-vehicle control instructions, such as instructions that require network connection, such as playing music and navigation destination.

[0030] S104. The processing device inputs the text information into the cloud control model to obtain the cloud recognition result.

[0031] The cloud control model refers to the control model deployed in the cloud. This cloud control model corresponds to the vehicle. Since the cloud control model is deployed in the cloud with stronger computing power, its prediction accuracy is higher than that of the vehicle-end control model.

[0032] To make the test environment the same, the processing device needs to obtain the same cloud control model as that of the vehicle end. That is to say, the cloud control model into which the processing device inputs the text information is the same as the cloud control model corresponding to the vehicle. After the processing device obtains the text information, it can input the text information to the cloud control model to obtain the cloud recognition result.

[0033] Among them, the cloud recognition result is similar to the vehicle-end recognition result. The cloud recognition result includes two cases. The first case is the vehicle control instruction, and the second case is the rejection recognition instruction.

[0034] It should be noted that this application does not specifically limit the execution order of S103 and S104. In some other embodiments, the processing device may also execute S104 first and then S103.

[0035] S105. The processing device determines whether the vehicle-end recognition result is consistent with the cloud recognition result.

[0036] The processing device determines whether the vehicle-end recognition result is consistent with the cloud recognition result. If the vehicle-end recognition result is consistent with the cloud recognition result, the process ends. If the vehicle-end recognition result is inconsistent with the cloud recognition result, S106 is executed.

[0037] If the processing device determines that the vehicle-end recognition result is consistent with the cloud-end recognition result, it indicates that the prediction result of the vehicle-end control model for this text information is relatively accurate and there is no need to optimize the vehicle-end control model relying on this text information. In the case where it is determined that the vehicle-end recognition result is inconsistent with the cloud-end recognition result, it indicates that the prediction result of the vehicle-end control model for this text information is inaccurate and the vehicle-end control model needs to be optimized using this text information.

[0038] In some embodiments, if the vehicle-end recognition result is a vehicle control instruction and the cloud-end recognition result is also a vehicle control instruction, or the vehicle-end recognition result is a rejection recognition instruction and the cloud-end recognition result is a rejection recognition instruction, then a judgment result that the vehicle-end recognition result is consistent with the cloud-end recognition result is obtained. If the vehicle-end recognition result is a vehicle control instruction and the cloud-end recognition result is a rejection recognition instruction, or the vehicle-end recognition result is a rejection recognition instruction and the cloud-end recognition result is a vehicle control instruction, then a judgment result that the vehicle-end recognition result is inconsistent with the cloud-end recognition result is obtained.

[0039] In some embodiments, the processing device can pre-obtain a vehicle control instruction set. After the processing device inputs the text information into the vehicle-end control model (or the cloud-end control model), an output control instruction can be obtained, and then this control instruction is matched with the vehicle control instruction set. If this control instruction hits in the vehicle control instruction set, it indicates that this control instruction is a vehicle control instruction; otherwise, it is a rejection recognition instruction.

[0040] S106. The processing device optimizes the vehicle-end control model according to the text information, the cloud-end recognition result, and the vehicle-end recognition result.

[0041] When the processing device determines that the vehicle-end recognition result is inconsistent with the cloud-end recognition result, it can save the data in this inconsistent situation. For example, the text information, as well as the vehicle-end recognition result and the cloud-end recognition result corresponding to this text information, so as to facilitate subsequent optimization of the vehicle-end control model using this text information, as well as the vehicle-end recognition result and the cloud-end recognition result corresponding to this text information.

[0042] In some embodiments, the processing device can add the text information, as well as the cloud-end recognition result and the vehicle-end recognition result corresponding to this text information, to the optimization sample data set. Among them, the samples included in the optimization sample data set are all composed of the text information when the cloud-end recognition result is inconsistent with the vehicle-end recognition result, as well as the cloud-end recognition result and the vehicle-end recognition result corresponding to this text information.

[0043] After obtaining the optimized sample data set, text information in the optimized sample data set can be clustered based on semantic similarity to obtain multiple categories, and then it is determined whether there is a target category in the multiple categories where the number of samples is greater than the quantity threshold. If there is such a target category, the target text information in the target category, as well as the corresponding target vehicle-end recognition result and target cloud-end recognition result, are used to optimize the vehicle-end control model. Among them, the quantity threshold can be set according to the actual situation. For example, it can be 10 or 100, etc.

[0044] In this method, by clustering, text information of the same category is found, and then only when the quantity of this text information exceeds the quantity threshold, the target text information, the target vehicle-end recognition result, and the target cloud-end recognition result are used to optimize the vehicle-end control model, which can exclude the interference of some abnormal data, thereby making the vehicle-end control model more accurate.

[0045] During the process of the processing device optimizing the vehicle-end control model by using the target text information, the target vehicle-end recognition result, and the target cloud-end recognition result, it can control the output result corresponding to the target text information to be close to the target cloud-end recognition result and far from the target vehicle-end recognition result, so as to further improve the accuracy of the optimized vehicle-end control model and reduce the misrecognition situation of the vehicle-end control model.

[0046] In some embodiments, after the processor determines the target text information, the target text information can be first input into the cloud control model to obtain the target cloud-end recognition result corresponding to each target text information. Of course, the processing device can also directly obtain the target cloud-end recognition result corresponding to the target text information (because in the previous step, the target text information has been input to the cloud control model, and the cloud control model has output the target cloud-end recognition result of this target text information). Then, the processing device can determine whether the target cloud-end recognition results corresponding to each target text information are consistent. When the target cloud-end recognition results corresponding to each target text information are all consistent, it further indicates that these target text information belong to the same category, and then the vehicle-end control model is optimized by using these target text information, the corresponding target cloud-end recognition result, and the target vehicle-end recognition result.

[0047] In some embodiments, when there are inconsistencies in the target cloud-end recognition results corresponding to each target text information, the quantities corresponding to each target cloud-end recognition result are obtained; the text information corresponding to the recognition result with the largest quantity among the multiple target cloud-end recognition results is determined as the remaining text information; the vehicle-end control model is optimized by using the remaining text information, the remaining cloud-end recognition result corresponding to the remaining text information, and the remaining vehicle-end recognition result.

[0048] For example, taking the number of target text information as 15 as an example, multiple pieces of target text information are shown in Table 1 below.

[0049] Table 1:

[0050] Combined with Table 1, the target cloud recognition results include Result A, Result B, Result C, and Result D. Among them, the quantity of Result A is the largest. Therefore, the text information corresponding to Result A can be determined as the remaining text information, that is, Text 1 - Text 3, Text 5, Text 6, Text 8, Text 10 - Text 12, Text 14, and Text 15. Then, using these remaining text information, as well as the remaining cloud recognition results and remaining vehicle - end recognition results corresponding to these remaining text information, the vehicle - end control model is optimized. By selecting the text information corresponding to the recognition result with the largest quantity for subsequent optimization, abnormal data can be screened, so as to exclude the abnormal data, improve the accuracy of the samples used for optimization, and further improve the accuracy of the optimized vehicle - end control model.

[0051] In the above example, the number of target cloud recognition results is 3 or more (including 3). In some other embodiments, the number of categories of target cloud recognition results is 2, for example, the first recognition result and the second recognition result. The processing device obtains the first quantity of the inconsistent first recognition result and the second quantity of the second recognition result. When the difference between the first quantity and the second quantity is greater than the difference threshold, the text information corresponding to the recognition result with the smallest quantity is removed from the multiple pieces of target text information, obtaining multiple pieces of remaining text information, as well as the remaining cloud recognition results and remaining vehicle - end recognition results corresponding to the remaining text information. The difference threshold can be determined based on the maximum value of the first quantity and the second quantity, and then the vehicle - end control model is optimized using the remaining text information, remaining cloud recognition results, and remaining vehicle - end recognition results.

[0052] For example: Taking the number of target text information as 10 as an example, multiple pieces of target text information are shown in Table 2 below.

[0053] Table 2:

[0054] Combined with Table 2, taking the first recognition result as Result A and the second recognition result as Result B as an example, the first quantity of Result A is 9, and the second quantity of Result B is 1. Exemplarily, the difference threshold can be determined by the first quantity. Specifically, it can be a preset ratio of the first quantity, where the preset ratio can be 0.8. Based on this, the calculated difference threshold is 0.8×9 = 7.2. The difference between the first quantity and the second quantity is 8, and 8>7.2. Therefore, it is necessary to remove the text information corresponding to the recognition result with the smallest quantity from multiple target text information, that is, remove text 4. The remaining text information includes text 1 to text 3, and text 5 to text 10. Then, use the remaining text information, as well as the remaining cloud recognition results and the remaining vehicle-end recognition results, to optimize the vehicle-end control model.

[0055] Among them, by setting a threshold, abnormal data is screened, so that the abnormal data is excluded, improving the accuracy of the samples used for optimization, and further improving the accuracy of the optimized vehicle-end control model.

[0056] In some embodiments, after the processing device optimizes the vehicle-end control model, it can also obtain test samples and use the test samples to test the accuracy of the optimized vehicle-end control model. When the accuracy is greater than the accuracy threshold, the optimized vehicle-end control model is deployed to the vehicle-end.

[0057] In this method, after the processing device completes the optimization of the vehicle-end control model, it also verifies the accuracy of the optimized vehicle-end control model. Only when the accuracy meets the requirements, it is deployed to the vehicle-end. In this way, it can further ensure that the optimized vehicle-end control model has a certain accuracy.

[0058] Furthermore, the accuracy threshold can be determined based on the accuracy of the vehicle-end control model before optimization. For example, the minimum value of the accuracy threshold can be the accuracy of the vehicle-end control model before optimization, and the accuracy threshold can also be a preset multiple of the accuracy of the vehicle-end control model before optimization, where the preset multiple can be 1.1. Based on this, it can be ensured that the accuracy of the optimized vehicle-end control model is greater than or equal to the accuracy of the vehicle-end control model before optimization.

[0059] In some embodiments, the processing device can optimize the vehicle-end control model according to a preset period, where the preset period can be 1 week or 1 month, and the present application does not limit this. The processing device optimizes the vehicle-end control model according to a preset period, which can ensure that as the usage time of the user becomes longer, the vehicle-end control model can be continuously optimized, and thus the accuracy of the vehicle-end control model is also continuously improved.

[0060] Based on the above description, an embodiment of the present application provides an optimization method for a vehicle-end control model. The method includes: obtaining the voice information of a user, then converting it into text information, and respectively inputting it into the vehicle-end control model and the cloud control model to obtain a vehicle-end recognition result and a cloud recognition result, and then comparing whether the vehicle-end recognition result and the cloud recognition result are consistent. If they are inconsistent, it means that there is a problem with the vehicle-end recognition result and the vehicle-end control model needs to be updated. Then, the vehicle-end control model is optimized using the text information, the vehicle-end recognition result, and the cloud recognition result, so as to realize the automatic optimization of the vehicle-end control model. Compared with the traditional optimization scheme, the optimization scheme of the present application does not require manual collection of end-side data and participation in the model optimization process. Therefore, this method can improve the optimization efficiency.

[0061] As described above in conjunction with Figure 1 the optimization method for the vehicle-end control model provided by the embodiment of the present application has been introduced in detail. Next, the devices and equipment provided by the embodiment of the present application will be introduced in conjunction with the accompanying drawings.

[0062] As Figure 2 shown, this figure is a schematic diagram of an optimization device for a vehicle-end control model provided by an embodiment of the present application. The device includes: An acquisition module 201, configured to acquire the voice information of a user; An identification module 202, configured to identify the voice information to obtain text information; input the text information into the vehicle-end control model to obtain a vehicle-end recognition result, and input the text information into the cloud control model to obtain a cloud recognition result; A judgment module 203, configured to judge whether the vehicle-end recognition result is consistent with the cloud recognition result to obtain a judgment result; An optimization module 204, configured to, if the judgment result indicates that the vehicle-end recognition result is inconsistent with the cloud recognition result, optimize the vehicle-end control model according to the text information, the cloud recognition result, and the vehicle-end recognition result.

[0063] In some possible implementation manners, the judgment module 203 is specifically configured to, if the vehicle-end recognition result is a vehicle control instruction and the cloud recognition result is a rejection recognition instruction, or, the vehicle-end recognition result is a rejection recognition instruction and the cloud recognition result is a vehicle control instruction, obtain a judgment result that the vehicle-end recognition result is inconsistent with the cloud recognition result; if the vehicle-end recognition result is a vehicle control instruction and the cloud recognition result is a vehicle control instruction, or, the vehicle-end recognition result is a rejection recognition instruction and the cloud recognition result is a rejection recognition instruction, obtain a judgment result that the vehicle-end recognition result is consistent with the cloud recognition result.

[0064] In some possible implementation manners, the optimization module 204 is specifically configured to add the text information, the corresponding cloud recognition result and vehicle-end recognition result of the text information into an optimization sample data set; cluster the text information in the optimization sample data set based on semantic similarity to obtain multiple categories; in the case that there is a target category in the multiple categories whose sample quantity is greater than a quantity threshold, optimize the vehicle-end control model by using the target text information in the target category, the corresponding target vehicle-end recognition result and target cloud recognition result of the target text information.

[0065] In some possible implementation manners, the optimization module 204 is specifically configured to input each target text information in the target category into the cloud control model to obtain a corresponding target cloud recognition result for each target text information; in the case that the target cloud recognition results corresponding to each target text information are all the same, optimize the vehicle-end control model by using the target text information, the target vehicle-end recognition result and the target cloud recognition result.

[0066] In some possible implementation manners, the optimization module 204 is further configured to, in the case that there are inconsistent target cloud recognition results corresponding to each target text information, obtain the quantities corresponding to the respective target cloud recognition results; determine the text information corresponding to the recognition result with the largest quantity among the multiple target cloud recognition results as the remaining text information; optimize the vehicle-end control model by using the remaining text information, the corresponding remaining cloud recognition result and remaining vehicle-end recognition result of the remaining text information.

[0067] In some possible implementation manners, the apparatus further includes a test module and a deployment module; The acquisition module 201 is further configured to acquire test samples; The test module is configured to test the accuracy rate of the optimized vehicle-end control model by using the test samples; The deployment module is configured to deploy the optimized vehicle-end control model to the vehicle-end in the case that the accuracy rate is greater than an accuracy rate threshold.

[0068] In some possible implementation manners, the optimization module 204 is specifically configured to optimize the vehicle-end control model according to a preset period.

[0069] The optimization apparatus for the vehicle-end control model according to the embodiments of the present application may correspond to execute the methods described in the embodiments of the present application, and the above other operations and / or functions of each module / unit of the optimization apparatus for the vehicle-end control model respectively implement Figure 1 the corresponding processes of the respective methods in the illustrated embodiments. For the sake of brevity, details are not described herein again.

[0070] Embodiments of the present application also provide a computing device. As Figure 3 shown, this figure is a schematic diagram of a computing device provided by an embodiment of the present application. The computing device 300 includes a bus 301, a processor 302, a communication interface 303, and a memory 304. The processor 302, the memory 304, and the communication interface 303 communicate with each other through the bus 301.

[0071] The bus 301 can be a peripheral component interconnect (PCI) bus, an extended industry standard architecture (EISA) bus, or the like. The bus can be divided into an address bus, a data bus, a control bus, etc. For the sake of simplicity of representation, Figure 3 only a thick line is used to represent it in the figure, but it does not mean that there is only one bus or one type of bus.

[0072] The processor 302 can be any one or more of a central processing unit (CPU), a graphics processing unit (GPU), a microprocessor (MP), or a digital signal processor (DSP).

[0073] The communication interface 303 is used for external communication.

[0074] The memory 304 can include volatile memory, such as random access memory (RAM). The memory 304 can also include non-volatile memory, such as read-only memory (ROM), flash memory, a hard disk drive (HDD), or a solid state drive (SSD).

[0075] The memory 304 stores executable code, and the processor 302 executes the executable code to perform the foregoing optimization method of the vehicle-side control model.

[0076] Specifically, in the case of implementing Figure 2 the embodiments shown, and Figure 2 when each module or unit of the optimization device of the vehicle-side control model described in the embodiments is implemented by software, execute Figure 2The software or program code required for the functions of each module / unit in can be partially or entirely stored in the memory 304. The processor 302 executes the program code corresponding to each unit stored in the memory 304 to execute the optimization method of the vehicle-end control model described above.

[0077] An embodiment of this application also provides a computer-readable storage medium. The computer-readable storage medium can be any available medium that a computing device can store or a data storage device such as a data center that contains one or more available media. The available medium can be a magnetic medium (such as a floppy disk, hard disk, magnetic tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid-state drive), etc. The computer-readable storage medium includes instructions that direct the computing device to execute the optimization method of the vehicle-end control model described above.

[0078] An embodiment of this application also provides a computer program product that includes one or more computer instructions. When the computer instructions are loaded and executed on a computing device, they entirely or partially generate the processes or functions according to the embodiments of this application.

[0079] The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from a website, computer, or data center to another website, computer, or data center by wire (such as coaxial cable, optical fiber, digital subscriber line) or wirelessly (such as infrared, wireless, microwave, etc.).

[0080] When the computer program product is executed by a computer, the computer executes any of the methods of the optimization method of the vehicle-end control model described above. The computer program product can be a software installation package. In the case where any of the methods of the optimization method of the vehicle-end control model described above are needed, the computer program product can be downloaded and executed on the computer.

[0081] The descriptions of the processes or structures corresponding to the above respective drawings each have their own emphases. For parts not detailed in a certain process or structure, reference can be made to the relevant descriptions of other processes or structures.

[0082] As described above, the above are only specific implementation manners of this application, but the protection scope of this application is not limited thereto. Any changes or substitutions within the technical scope disclosed in this application should be covered by the protection scope of this application.

Claims

1. An optimization method for a car-end control model, characterized in that, The method includes: Obtaining the voice information of the user; Identifying the voice information to obtain text information; Inputting the text information into the vehicle-side control model to obtain a vehicle-side recognition result, and inputting the text information into the cloud-side control model to obtain a cloud-side recognition result; Judging whether the vehicle-side recognition result is consistent with the cloud-side recognition result to obtain a judgment result; If the judgment result indicates that the vehicle-side recognition result is inconsistent with the cloud-side recognition result, optimizing the vehicle-side control model according to the text information, the cloud-side recognition result, and the vehicle-side recognition result.

2. The method according to claim 1, wherein The judging whether the vehicle-side recognition result is consistent with the cloud-side recognition result to obtain a judgment result includes: If the vehicle-side recognition result is a vehicle control instruction and the cloud-side recognition result is a rejection recognition instruction, or the vehicle-side recognition result is a rejection recognition instruction and the cloud-side recognition result is a vehicle control instruction, obtaining a judgment result that the vehicle-side recognition result is inconsistent with the cloud-side recognition result; If the vehicle-side recognition result is a vehicle control instruction and the cloud-side recognition result is a vehicle control instruction, or the vehicle-side recognition result is a rejection recognition instruction and the cloud-side recognition result is a rejection recognition instruction, obtaining a judgment result that the vehicle-side recognition result is consistent with the cloud-side recognition result.

3. The method according to claim 1, wherein The optimizing the vehicle-side control model according to the text information, the cloud-side recognition result, and the vehicle-side recognition result includes: Adding the text information, the corresponding cloud-side recognition result, and the vehicle-side recognition result to the optimization sample data set; Clustering the text information in the optimization sample data set based on semantic similarity to obtain multiple categories; In the case that there is a target category with the number of samples greater than the number threshold in the multiple categories, optimizing the vehicle-side control model by using the target text information in the target category, the corresponding target vehicle-side recognition result, and the target cloud-side recognition result.

4. The method according to claim 3, wherein The optimizing the vehicle-side control model by using the target text information in the target category, the corresponding target vehicle-side recognition result, and the target cloud-side recognition result includes: Inputting each target text information in the target category into the cloud-side control model to obtain a corresponding target cloud-side recognition result for each target text information; In the case that the corresponding target cloud-side recognition results for each target text information are all consistent, optimizing the vehicle-side control model by using the target text information, the target vehicle-side recognition result, and the target cloud-side recognition result.

5. The method according to claim 4, characterized in that, The method further includes: In the case that there are inconsistencies in the corresponding target cloud-side recognition results for each target text information, obtaining the quantities corresponding to the respective target cloud-side recognition results; Determining the text information corresponding to the recognition result with the largest quantity among the multiple target cloud-side recognition results as the remaining text information; Optimizing the vehicle-side control model by using the remaining text information, the corresponding remaining cloud-side recognition result, and the remaining vehicle-side recognition result.

6. The method according to claim 1, wherein After optimizing the vehicle-side control model, the method further includes: Obtain a test sample; Use the test sample to test the accuracy of the optimized vehicle-end control model; When the accuracy is greater than the accuracy threshold, deploy the optimized vehicle-end control model to the vehicle end.

7. The method according to any one of claims 1-6, characterized in that, The optimization of the vehicle-end control model includes: Optimize the vehicle-end control model according to a preset period.

8. An optimization device for a car-end control model, characterized in that, The device includes: An acquisition module for acquiring the user's voice information; An identification module for identifying the voice information to obtain text information; inputting the text information into the vehicle-end control model to obtain a vehicle-end identification result, and inputting the text information into the cloud control model to obtain a cloud identification result; A judgment module for judging whether the vehicle-end identification result is consistent with the cloud identification result to obtain a judgment result; An optimization module for, if the judgment result indicates that the vehicle-end identification result is inconsistent with the cloud identification result, optimizing the vehicle-end control model according to the text information, the cloud identification result, and the vehicle-end identification result.

9. A computing device, characterized in that, Includes a memory and a processor; Wherein, one or more computer programs are stored in the memory, and the one or more computer programs include instructions; when the instructions are executed by the processor, the computing device executes the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store a computer program, and the computer program is used to execute the method according to any one of claims 1 to 7.

Images