A vehicle computer voice response method, device, equipment and product

Abstract

The present invention provides a vehicle machine voice response method, device, equipment and product, which include: obtaining a text question obtained by a user's voice input in a vehicle; inputting the text question into a preset question-answering model to obtain a first answer matching the text question; obtaining a first simplified answer based on the first answer; wherein the first simplified answer is the answer content refined and simplified based on the content of the first answer; and outputting the first simplified answer. After the user's voice input, the text question is obtained, the first answer is obtained based on the text question, and then the simplified content of the first answer is obtained and displayed to the user, so that the user can quickly obtain the core answer to the question and improve the user experience.

Description

Technical Field

[0001] The present invention relates to the field of vehicle-computer interaction technology, and in particular to a vehicle-computer voice response method, device, equipment and product. Background Art

[0002] At present, the car machine has the basic capabilities of the big model 1.0, such as multiple characters, small talk, encyclopedia questions, basic car question answers, etc. In the process of developing the big model of the car machine voice, combined with the high-frequency questions that users are concerned about after the pre-sale and release of the product, monitoring is carried out from the terminal and vertical media public opinion end, and standardized answering scripts are compiled. However, the existing car machine can only passively output large-volume answers, which cannot allow users to quickly understand the core answer to the question, causing inconvenience to users. Summary of the invention

[0003] The purpose of the present invention is to provide a vehicle computer voice response method, device, equipment and product, which is used to solve the problem in the prior art that the vehicle computer can only passively output a large amount of answers, and cannot allow users to quickly understand the core answer to the question, causing inconvenience to users.

[0004] In order to solve the above technical problems, an embodiment of the present invention provides a vehicle machine voice response method, which includes:

[0005] Obtaining text questions obtained by voice input by the user in the vehicle;

[0006] Input the text question into a preset question-answering model to obtain a first answer matching the text question;

[0007] Obtain a first condensed answer according to the first answer; wherein the first condensed answer is the answer content that is refined and simplified according to the content of the first answer;

[0008] The first condensed answer is output.

[0009] Optionally, in the method, after outputting the first simplified answer, the method further comprises:

[0010] In the case of obtaining a first instruction input by the user, outputting a full answer corresponding to the first simplified answer;

[0011] Among them, the full answer is the entire content of the first answer corresponding to the first concise answer.

[0012] Optionally, the method further comprises:

[0013] Obtaining original data; wherein the original data includes original questions and original answers;

[0014] Generalize the original problem to obtain a generalized problem;

[0015] The generalized question and the original answer are input into a deep learning model as training data to obtain the question-answering model.

[0016] Optionally, in the method, generalizing the original problem to obtain a generalized problem includes:

[0017] Obtaining generalization parameters of the original problem;

[0018] Regularization is performed according to the generalization parameter to obtain the generalization problem.

[0019] Optionally, the method, wherein the generalized question and the original answer are input into a deep learning model as training data to obtain the question-answering model, comprises:

[0020] Divide the training data into multiple parts evenly to obtain multiple sub-training data;

[0021] Training the plurality of sub-training data in the deep learning model simultaneously to obtain training results; wherein the training results are answers corresponding to the generalization questions in the training data;

[0022] Verify the training result and the original answer in the training data to obtain a verification result;

[0023] The deep learning model is adjusted according to the verification result to obtain the question-answering model.

[0024] Optionally, in the method, the step of outputting the first simplified answer comprises one or more of the following:

[0025] Converting the first simplified answer into a voice form and playing it in a player of the vehicle;

[0026] The first condensed answer is displayed in text form on a screen of the vehicle.

[0027] Optionally, in the method, the step of obtaining a text question obtained by voice input by a user in a vehicle comprises:

[0028] When the user wakes up the voice assistant of the vehicle, collecting a first question input by the user through voice;

[0029] The first question is converted into text data to obtain the text question.

[0030] Optionally, in the method, after outputting the first simplified answer, the method further comprises:

[0031] The text question, the first answer and the first condensed answer are recorded.

[0032] In order to achieve the above object, an embodiment of the present invention further provides a vehicle machine voice response device, which includes:

[0033] A first acquisition module is used to acquire a text question obtained by a user's voice input in a vehicle;

[0034] A second acquisition module is used to input the text question into a preset question-answering model to obtain a first answer matching the text question;

[0035] A third acquisition module is used to acquire a first simplified answer according to the first answer; wherein the first simplified answer is the answer content that is refined and simplified according to the content of the first answer;

[0036] The first output module is used to output the first simplified answer.

[0037] In order to achieve the above-mentioned purpose, an embodiment of the present invention also provides an electronic device, including: a processor, a memory, and a program or instruction stored in the memory and executable on the processor; wherein, when the processor executes the program or instruction, the vehicle voice response method as described above is implemented.

[0038] In order to achieve the above-mentioned purpose, an embodiment of the present invention further provides a readable storage medium on which a program or instruction is stored, wherein the program or instruction, when executed by a processor, implements the steps in the vehicle machine voice response method as described above.

[0039] In order to achieve the above-mentioned purpose, an embodiment of the present invention further provides a computer program product, which includes computer instructions, and when the computer instructions are executed by a processor, the steps of the vehicle computer voice response method as described above are implemented.

[0040] The beneficial effects of the above technical solution of the present invention are as follows:

[0041] In the above scheme, a text question is obtained after the user's voice input, a first answer is obtained based on the text question, and then a concise content of the first answer is obtained and displayed to the user, so that the user can quickly obtain the core answer to the question and improve the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 A schematic diagram of a vehicle computer voice response method according to an embodiment of the present invention;

[0043] Figure 2 A schematic diagram of distributed training of the vehicle machine voice response method according to an embodiment of the present invention;

[0044] Figure 3A schematic diagram of data parallelism of a vehicle machine voice response method according to an embodiment of the present invention;

[0045] Figure 4 This is a flow chart of the vehicle computer voice response method according to an embodiment of the present invention;

[0046] Figure 5 The figure is a schematic diagram of a vehicle computer voice response device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0047] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0048] The present invention aims to solve the problem that in the prior art, a vehicle computer can only passively output a large amount of answers, which cannot allow users to quickly understand the core answers to questions, causing inconvenience to users. The present invention provides a vehicle computer voice response method, device, equipment and product.

[0049] like Figure 1 As shown, an embodiment of the present invention provides a vehicle machine voice response method, which includes:

[0050] S10, obtaining a text question obtained by voice input by the user in the vehicle;

[0051] S20, inputting the text question into a preset question-answering model to obtain a first answer matching the text question;

[0052] S30, obtaining a first condensed answer according to the first answer; wherein the first condensed answer is an answer content that is refined and simplified according to the content of the first answer;

[0053] S40: outputting the first simplified answer.

[0054] In this embodiment, the text question is obtained after the user's voice input, the first answer is obtained based on the text question, and then the concise content of the first answer is obtained and displayed to the user, so that the user can quickly obtain the core answer to the question and improve the user experience.

[0055] It should be noted that, in step S30, the process of obtaining the first simplified answer according to the first answer is as follows:

[0056] 1. Data preprocessing:

[0057]

[0058]

[0059] context="""In order to ensure the safety of the frameless door design, S5 adopts an integrated hot-formed door ring design. The strength of the entire door ring is 1500mpa, and the deformation of the side collision is smaller than that of Volvo. At present, only our S5 and the latest generation of Tesla model3 use this technology. The crossbeam in the middle of the skylight is connected to the integrated door to form a safe and stable cage-type body. The torsional rigidity of the body reaches 51897N·m / deg, which can more comprehensively protect the safety of the passengers. In side collision, top pressure and other working conditions, it can effectively transmit the collision load to ensure the integrity of the passenger compartment and provide all-round safety protection for the passengers in the car. At the same time, it also gives the driver a wider rear view and more headroom for the rear passengers. In addition, the only rear floodlight in its class is set on the crossbeam: the light length is 376mm, the color temperature is about 6500k, which can be used to find things in dim light, etc."""

[0060] question="Why is the sky dome divided in the middle?"

[0061] answer=answer_question(question,context)

[0062] print(answer)

[0063] 2. Build the Flask application:

[0064]

[0065] 3. Test API:

[0066] curl -X POST\

[0067] -H"Content-Type:application / json"\

[0068] -d'{"question":"Why is the middle of the skylight divided?","context":"In order to ensure the safety of the frameless door design, the S5 adopts an integrated hot-formed door ring design. The strength of the entire door ring is 1500mpa, and the deformation of the side collision is smaller than that of Volvo. At present, the only cars that use this technology are our S5 and the latest generation of Tesla model3. The crossbeam in the middle of the skylight is connected to the integrated door to form a safe and stable cage-type body. The torsional rigidity of the body reaches 51897N·m / deg, which can more comprehensively protect the safety of the occupants. In side collisions, top pressure and other working conditions, it can effectively transmit the collision load, ensure the integrity of the passenger compartment, and provide all-round safety protection for the occupants in the car. At the same time, it also gives the driver a wider rear view and more headroom for the rear passengers. In addition, the only rear floodlight in its class is set on the crossbeam: the luminous length is 376mm, the color temperature is about 6500k, which can be used to find things in dim light, etc."}'

[0069] Reply: {"answer":"The crossbeam in the middle of the S5 skylight is connected to the integrated door, forming a safe and stable cage-type body. The torsional rigidity of the body reaches 51897N·m / deg, which can more comprehensively protect the safety of the passengers, effectively transmit the collision load in side collision, top pressure and other working conditions, and provide all-round safety protection for the passengers in the car. At the same time, it also gives the driver a wider rear view and the rear passengers more headroom."}.

[0070] Optionally, the method, wherein after step S40, the method further includes:

[0071] In the case of obtaining a first instruction input by the user, outputting a full answer corresponding to the first simplified answer;

[0072] Among them, the full answer is the entire content of the first answer corresponding to the first concise answer.

[0073] In this embodiment, after obtaining the first answer, in order to facilitate the user to quickly obtain the core content of the answer, the first condensed answer of the first answer is output to the user. When the user needs the full answer, the user only needs to input the first instruction to obtain the full answer. The first instruction can be a voice input of "View more" or a virtual button of "View more" set on the vehicle screen displaying the first condensed answer. The user jumps to the full answer after voice input or clicking the virtual button. The first instruction can also be set to other forms, not limited to the above methods.

[0074] Optionally, the method further comprises:

[0075] Obtaining original data; wherein the original data includes original questions and original answers;

[0076] Generalize the original problem to obtain a generalized problem;

[0077] The generalized question and the original answer are input into a deep learning model as training data to obtain the question-answering model.

[0078] In this embodiment, the question-answering model is trained and acquired through generalized questions and answers. After generalized questions, the situation where one original question corresponds to one original answer can be converted into multiple generalized questions corresponding to one original answer, thereby corresponding the user's various different questioning methods to the same standard answer, thereby improving user convenience. The original questions are for users' high-frequency concerns, and standard replies are compiled to break down technical points in detail. While answering user questions, users are informed of how to better use the vehicle. It has a great effect on improving user reputation and public opinion. The original answers are standard answers that have been manually verified. They are all formulated through review and are authoritative and comprehensive. After getting the answers, users can also answer questions and resolve doubts for potential customers with similar questions on social platforms to improve brand favorability.

[0079] Optionally, in the method, generalizing the original problem to obtain a generalized problem includes:

[0080] Obtaining generalization parameters of the original problem;

[0081] Regularization is performed according to the generalization parameter to obtain the generalization problem.

[0082] In this embodiment, the original data provides diversified training data for the question-answering model, ensuring that the question-answering model can be exposed to a variety of different scenarios and data types. After obtaining the generalization parameter θ of the original question, regularization technology (L1, L2 regularization) is used to prevent the model from overfitting, thereby improving the generalization ability. The formula used is as follows:

[0083]

[0084] Among them, formula ① is L1 regularization, formula ② is L2 regularization, MSE(θ) reflects the difference between the estimator θ and the estimated quantity. In the generalization process, J(θ) is minimized, and θ is adjusted by adjusting the parameters. i Approaching 0. After adjustment, the same type of questions can get the same answer, such as: "The rear windshield is of poor material, the rear windshield is not classy, ​​what material is the rear windshield made of? What material is the rear glass made of? Why is the rear glass of poor material?", etc. This expands the range of possible questions that users may ask, increases the volume of questions that can be identified, and improves the user experience.

[0085] Optionally, the method, wherein the generalized question and the original answer are input into a deep learning model as training data to obtain the question-answering model, comprises:

[0086] Divide the training data into multiple parts evenly to obtain multiple sub-training data;

[0087] Training the plurality of sub-training data in the deep learning model simultaneously to obtain training results; wherein the training results are answers corresponding to the generalization questions in the training data;

[0088] Verify the training result and the original answer in the training data to obtain a verification result;

[0089] The deep learning model is adjusted according to the verification result to obtain the question-answering model.

[0090] In this embodiment, the generalized question and the original answer are stored in the question-answering knowledge base, input into the large model platform for training and testing, and the model is trained using deep learning (distributed training). Distributed training uses collective communication training, such as Figure 2 As shown in the figure, each training Worker (worker), including Worker0, Worker1, and Worker2, communicates through AllReduce (collective communication primitive) to connect and match text data. Each node not only forms a deep connection with the upstream and downstream, but also grasps the latest global gradient information. Figure 3As shown, data partition0 and datapartition1 are used to partition the data, and the training data is evenly divided into multiple parts to obtain multiple sub-training data; layer0, layer1, ..., layer n-1, layer n are multiple layers of the generalization problem, Synchronization synchronously obtains the answer stored in the knowledge base, synchronizes the gradient through Synchronized gradients, and SGD (Stochastic Gradient Descent), the optimization algorithm optimizes the parameters of the obtained answer, so as to obtain the training result; in this process, through gradient aggregation (grad0, grad1, ..., grad n-1, grad n), AllReduce multi-stream communication is used for communication to ensure the consistency of the global parameter gradient value, and the optimized parameters are also equal; after obtaining the training result, the training result is returned to the multiple layers of the generalization problem for verification, and the deep learning model is adjusted according to the verification result to obtain the question-answering model. After training, according to the optimized parameter results, the problem test is carried out, and the generalization problem is input, and the corresponding original answer in the knowledge base can be generated.

[0091] Optionally, in the method, step S40 includes one or more of the following:

[0092] Converting the first simplified answer into a voice form and playing it in a player of the vehicle;

[0093] The first condensed answer is displayed in text form on a screen of the vehicle.

[0094] In this embodiment, the first condensed answer is converted into voice form and played in the player of the vehicle. The first condensed answer can be set to be within 150 words that can be played within one minute. By reducing the number of words in the answer, the playing time of the answer is reduced. Using the first condensed answer can enhance the user experience. The first condensed answer is displayed on the screen of the vehicle in text form. Due to the limitations of the vehicle screen size and text size, the answer is limited to 7 lines and 300 words. The complete first condensed answer can be displayed on only one page, and the user does not need to swipe the screen to understand the core content, thereby enhancing the user experience.

[0095] Optionally, in the method, step S10 includes:

[0096] When the user wakes up the voice assistant of the vehicle, collecting a first question input by the user through voice;

[0097] The first question is converted into text data to obtain the text question.

[0098] In this embodiment, if Figure 4 As shown, in step S10, the question input and text processing are performed. The user can perform voice input on the vehicle computer, and after waking up the vehicle computer AI voice assistant by voice / button / vehicle computer, describe the corresponding question, i.e., the first question. At the same time, the corresponding picture, driving log, and vehicle information can be uploaded after obtaining the user's consent, and the recognized voice signal is converted into text data to obtain the text question. In step S20, the question-answering model is performed and the answer is returned, i.e., the text question is input into the question-answering model to obtain the first answer.

[0099] Optionally, the method, wherein after step S40, the method further includes:

[0100] The text question, the first answer and the first condensed answer are recorded.

[0101] In this embodiment, the text question, the first answer and the first condensed answer are stored in the secondary menu of the vehicle manual, and the text question, the first answer and the first condensed answer are matched so that they can be viewed at any time through the manual.

[0102] like Figure 5 As shown, in order to achieve the above-mentioned purpose, an embodiment of the present invention further provides a vehicle machine voice response device, which includes:

[0103] The first acquisition module 501 is used to acquire a text question obtained by a user's voice input in a vehicle;

[0104] A second acquisition module 502 is used to input the text question into a preset question-answering model to obtain a first answer matching the text question;

[0105] The third acquisition module 503 is used to acquire a first simplified answer according to the first answer; wherein the first simplified answer is the answer content after being refined and simplified according to the content of the first answer;

[0106] The first output module 504 is used to output the first simplified answer.

[0107] Optionally, the device further comprises:

[0108] A second output module is used to output a full answer corresponding to the first simplified answer when obtaining a first instruction input by a user;

[0109] Among them, the full answer is the entire content of the first answer corresponding to the first concise answer.

[0110] Optionally, the device further comprises:

[0111] A fourth acquisition module is used to acquire original data; wherein the original data includes original questions and original answers;

[0112] A fifth acquisition module, used for generalizing the original problem to obtain a generalized problem;

[0113] The sixth acquisition module is used to input the generalized question and the original answer as training data into the deep learning model to obtain the question-answering model.

[0114] Optionally, in the device, the fifth acquisition module includes:

[0115] A first acquisition unit, used to acquire a generalization parameter of the original problem;

[0116] The second acquisition unit is used to perform regularization according to the generalization parameter to acquire the generalization problem.

[0117] Optionally, in the device, the sixth acquisition module includes:

[0118] A third acquisition unit, used for evenly dividing the training data into multiple parts to acquire multiple sub-training data;

[0119] A fourth acquisition unit is used to train the plurality of sub-training data in the deep learning model at the same time to obtain a training result; wherein the training result is an answer corresponding to the generalization question in the training data;

[0120] A fifth acquiring unit, configured to perform verification according to the training result and the original answer in the training data to acquire a verification result;

[0121] The sixth acquisition unit is used to adjust the deep learning model according to the verification result to obtain the question-answering model.

[0122] Optionally, in the device, the first output module 504 includes one or more of the following:

[0123] A first processing unit, configured to convert the first simplified answer into a voice form and play it in a player of the vehicle;

[0124] The second processing unit is used to display the first simplified answer in text form on the screen of the vehicle.

[0125] Optionally, in the device, the first acquisition module 501 includes:

[0126] A third processing unit is used to collect a first question input by a user through voice when the user wakes up the voice assistant of the vehicle;

[0127] A seventh acquisition unit is used to convert the first question into text data to acquire the text question.

[0128] Optionally, the device further comprises:

[0129] The fourth processing unit is used to record the text question, the first answer and the first simplified answer.

[0130] It should be noted here that the above-mentioned device provided in the embodiment of the present invention can implement all the method steps implemented in the above-mentioned method embodiment, and can achieve the same technical effect. The parts and beneficial effects that are the same as the method embodiment in this embodiment will not be described in detail here.

[0131] In order to achieve the above-mentioned purpose, an embodiment of the present invention also provides an electronic device, including: a processor, a memory, and a program or instruction stored in the memory and executable on the processor; wherein, when the processor executes the program or instruction, the vehicle voice response method as described above is implemented.

[0132] In order to achieve the above-mentioned purpose, an embodiment of the present invention further provides a readable storage medium on which a program or instruction is stored, wherein the program or instruction, when executed by a processor, implements the steps in the vehicle machine voice response method as described above.

[0133] In order to achieve the above-mentioned purpose, an embodiment of the present invention further provides a computer program product, which includes computer instructions, and when the computer instructions are executed by a processor, the steps of the vehicle computer voice response method as described above are implemented.

[0134] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A vehicle computer voice response method, characterized in that: include: Obtaining text questions obtained by voice input by the user in the vehicle; Input the text question into a preset question-answering model to obtain a first answer matching the text question; Obtain a first condensed answer according to the first answer; wherein the first condensed answer is the answer content that is refined and simplified according to the content of the first answer; The first condensed answer is output.

2. The method according to claim 1, characterized in that: After outputting the first simplified answer, the method further includes: In the case of obtaining a first instruction input by the user, outputting a full answer corresponding to the first simplified answer; Among them, the full answer is the entire content of the first answer corresponding to the first concise answer.

3. The method according to claim 1, characterized in that The method further comprises: Obtaining original data; wherein the original data includes original questions and original answers; Generalize the original problem to obtain a generalized problem; The generalized question and the original answer are input into a deep learning model as training data to obtain the question-answering model.

4. The method according to claim 3, characterized in that The generalizing the original problem to obtain a generalized problem includes: Obtaining generalization parameters of the original problem; Regularization is performed according to the generalization parameter to obtain the generalization problem.

5. The method according to claim 3, characterized in that: Inputting the generalized question and the original answer as training data into a deep learning model to obtain the question-answering model, including: Divide the training data into multiple parts evenly to obtain multiple sub-training data; Training the plurality of sub-training data in the deep learning model simultaneously to obtain training results; wherein the training results are answers corresponding to the generalization questions in the training data; Verify the training result and the original answer in the training data to obtain a verification result; The deep learning model is adjusted according to the verification result to obtain the question-answering model.

6. The method according to claim 1, characterized in that The outputting of the first condensed answer includes one or more of the following: Converting the first simplified answer into a voice form and playing it in a player of the vehicle; The first condensed answer is displayed in text form on a screen of the vehicle.

7. The method according to claim 1, characterized in that The step of obtaining a text question obtained by voice input by a user in a vehicle includes: When the user wakes up the voice assistant of the vehicle, collecting a first question input by the user through voice; The first question is converted into text data to obtain the text question.

8. The method according to claim 1, characterized in that After outputting the first simplified answer, the method further includes: The text question, the first answer and the first condensed answer are recorded.

9. A vehicle voice response device, characterized in that: include: A first acquisition module is used to acquire a text question obtained by a user's voice input in a vehicle; A second acquisition module is used to input the text question into a preset question-answering model to obtain a first answer matching the text question; A third acquisition module is used to acquire a first simplified answer according to the first answer; wherein the first simplified answer is the answer content that is refined and simplified according to the content of the first answer; The first output module is used to output the first simplified answer.

10. An electronic device comprising: A processor, a memory, and a program or instruction stored in the memory and executable on the processor; characterized in that when the processor executes the program or instruction, the vehicle machine voice response method as described in any one of claims 1-8 is implemented.

11. A readable storage medium having a program or instruction stored thereon, characterized in that: When the program or instruction is executed by the processor, the steps in the vehicle computer voice response method as described in any one of claims 1-8 are implemented.

12. A computer program product, characterized in that It includes computer instructions, which, when executed by a processor, implement the steps of the vehicle computer voice response method as described in any one of claims 1-8.

Images