Vehicle sound evaluation data determination method and device, equipment and storage medium
By obtaining the first objective score and manual score of the car sound data, eliminating the target evaluation parameters, and using the car sound quality evaluation model to calculate the parameter score and contribution degree, the problem of poor generalization ability of the automobile accelerated sound quality evaluation model in the existing technology is solved, and efficient car sound quality evaluation and design optimization are achieved.
Patent Information
- Application Number
- CN202510345157.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-21
- Publication Date
- 2025-07-11
AI Technical Summary
The existing automobile accelerating sound quality evaluation model has poor generalization capabilities and low prediction accuracy, which cannot accurately reflect the subjective perception of drivers and passengers, affecting the optimization and design efficiency of automobile accelerating sound quality.
By obtaining the first objective score and manual score of the car sound data, the objective score of the target evaluation parameters is eliminated, and the parameter score and contribution degree are calculated using the pre-trained car sound quality evaluation model to establish a mapping relationship between objective evaluation and subjective evaluation of psychoacoustics.
It significantly improves the efficiency and accuracy of vehicle sound quality evaluation, provides convenient and efficient evaluation tools for automotive acoustic design, and optimizes vehicle sound design.
Smart Images
Figure CN120297784A_ABST
Abstract
Description
Technical Field
[0001] Embodiments of the present application relate to the field of signal processing, and in particular, to a method, apparatus, device, and storage medium for determining vehicle sound evaluation data. Background Art
[0002] High-quality vehicle acceleration sounds can provide good acoustic feedback to the driver and passengers, thereby enhancing the driving experience and riding comfort. Therefore, in the field of automotive sound design, designing an acceleration sound that meets user expectations for the in-vehicle acceleration condition is a crucial task. Currently, in related technologies, psychoacoustic objective parameters are usually used to objectively evaluate the noise sound quality under automotive acceleration conditions. However, although psychoacoustic parameters have been widely used in sound quality evaluation, in the field of automotive acceleration sound quality prediction, existing objective evaluation models still have significant deficiencies. Specifically, these models have poor generalization ability and low prediction accuracy, and cannot accurately reflect the subjective perception of the driver and passengers towards the acceleration sound. These problems severely restrict the optimization and design efficiency of automotive acceleration sound quality.
[0003] Therefore, how to provide a method that can accurately evaluate the automotive acceleration sound quality has become an urgent problem to be solved in this field. Summary of the Invention
[0004] Embodiments of the present application provide a method, apparatus, electronic device, and computer-readable storage medium for determining vehicle sound evaluation data, aiming to improve the technical problem of how to provide a method that can accurately evaluate the automotive acceleration sound quality.
[0005] To solve the above problems, embodiments of the present application disclose a method for determining vehicle sound evaluation data, the method comprising:
[0006] Obtain vehicle sound data, a first objective score corresponding to the vehicle sound data, and an artificial score corresponding to the vehicle sound data; the first objective score includes objective scores for preset different evaluation parameters;
[0007] Input the first objective score into a pre-trained vehicle sound quality evaluation model for processing to obtain a target score corresponding to the vehicle sound data;
[0008] Exclude the objective score of the target evaluation parameter from the first objective score to obtain a second objective score; the target evaluation parameter is determined from the evaluation parameters;
[0009] Input the second objective score into the vehicle sound quality evaluation model for processing to obtain a parameter score corresponding to the vehicle sound data;
[0010] Calculate the contribution degree corresponding to the target evaluation parameter according to the parameter score, the manual score, and the target loss value, where the target loss value is calculated according to the target score and the manual score;
[0011] Store the target score and the contribution degree as the evaluation data corresponding to the vehicle sound data.
[0012] In an embodiment of the present application, vehicle sound data, a first objective score corresponding to the vehicle sound data, and a manual score corresponding to the vehicle sound data are obtained; the first objective score includes objective scores for different preset evaluation parameters; the first objective score is input into a pre-trained vehicle sound quality evaluation model for processing to obtain a target score corresponding to the vehicle sound data; the objective score of the target evaluation parameter in the first objective score is removed to obtain a second objective score; the target evaluation parameter is determined from the evaluation parameters; the second objective score is input into the vehicle sound quality evaluation model for processing to obtain a parameter score corresponding to the vehicle sound data; the contribution degree corresponding to the target evaluation parameter is calculated according to the parameter score, the manual score, and the target loss value, where the target loss value is calculated according to the target score and the manual score; the target score and the contribution degree are stored as the evaluation data corresponding to the vehicle sound data. In the embodiment of the present application, by inputting the first objective score and the second objective score into the vehicle sound quality evaluation model to obtain the target score and the parameter score, and based on the parameter score, the target score, and the manual score, the contribution degree of each evaluation parameter in the subjective evaluation of vehicle sound quality is calculated, thereby establishing a mapping relationship between the psychoacoustic objective evaluation parameters of vehicle sound quality and the subjective evaluation, clarifying the contribution degree of each evaluation parameter in the subjective evaluation, and thus significantly improving the efficiency of vehicle sound quality evaluation. In addition, the embodiment of the present application can predict the vehicle sound quality of a newly designed vehicle, provides a convenient and efficient evaluation tool for automotive acoustic design, and is of great significance for the optimization and guidance of vehicle sound design.
[0013] Optionally, the obtaining of the vehicle sound data, the first objective score corresponding to the vehicle sound data, and the manual score corresponding to the vehicle sound data includes:
[0014] Obtain the vehicle sound data to be processed and a preset overlap rate;
[0015] Overlap and segment the vehicle sound data to be processed according to the preset overlap rate to obtain vehicle sound data slices, and there is a corresponding loudness for the vehicle sound data slices;
[0016] Adjust the loudness of the vehicle sound data slices to obtain the vehicle sound data;
[0017] Calculate the first objective score and the manual score according to the vehicle sound data.
[0018] In the embodiments of the present application, the vehicle sound data to be processed is overlapped and segmented according to a preset overlap rate, and processed to obtain vehicle sound data. The first objective score and the manual score are calculated for the vehicle sound data, and are subsequently used in the training of the vehicle sound quality evaluation model and the determination of the target score and contribution degree.
[0019] Optionally, calculating the manual score according to the vehicle sound data further includes:
[0020] Obtain an initial manual score;
[0021] Filter out abnormal scores in the initial manual score to obtain a filtered manual score;
[0022] Calculate the average value of the filtered manual scores corresponding to the vehicle sound data, and use the average value as the manual score corresponding to the vehicle sound data.
[0023] In the embodiments of the present application, by preprocessing the initial manual score, the manual score corresponding to the vehicle sound data is obtained, thereby improving the accuracy of the manual score data, and also significantly improving the accuracy of the output data of the trained vehicle sound quality evaluation model and the accuracy of the contribution degree calculation.
[0024] Optionally, calculating the contribution degree corresponding to the target evaluation parameter according to the parameter score, the manual score and the target loss value includes:
[0025] Calculate the parameter loss value between the parameter score and the manual score;
[0026] Calculate the contribution degree corresponding to the target evaluation parameter according to the parameter loss value and the target loss value.
[0027] In the embodiments of the present application, the second objective score obtained by removing the objective score of the target evaluation parameter from the first objective score is input into the vehicle sound quality evaluation model to obtain a parameter score, and the parameter loss value between the parameter score and the manual score is calculated. Thus, the contribution degree of each evaluation parameter in the subjective evaluation of the vehicle sound can be obtained, thereby improving the efficiency of the vehicle sound quality evaluation and the efficiency of the vehicle sound design.
[0028] Optionally, the vehicle sound quality evaluation model is trained in the following manner:
[0029] Obtain a vehicle sound quality evaluation model to be trained;
[0030] Generate a first objective score matrix corresponding to the first objective score according to the first objective score;
[0031] Input the first objective score matrix into the vehicle sound quality evaluation model to be trained for processing to obtain a predicted score;
[0032] Calculate the training loss value between the predicted score and the manual score;
[0033] Update the to-be-trained vehicle sound quality evaluation model according to the training loss value to obtain a trained vehicle sound quality evaluation model.
[0034] In the embodiment of the present application, the vehicle sound quality evaluation model is trained according to the first objective score and the manual score, so as to establish the mapping relationship between the psychoacoustic objective parameters and the subjective evaluation of the vehicle sound quality, and establish a vehicle sound quality evaluation model, thereby improving the vehicle sound quality evaluation efficiency.
[0035] Optionally, the to-be-trained vehicle sound quality evaluation model includes a first fully connected layer, a second fully connected layer, a bidirectional long short-term memory network module, and a self-attention module. The process of inputting the first objective score matrix into the to-be-trained vehicle sound quality evaluation model to obtain a predicted score includes:
[0036] Input the first objective score matrix into the bidirectional long short-term memory network module for processing to obtain a first output vector;
[0037] Flatten the first output vector to obtain a flattened first output vector, and the flattened first output vector has a corresponding feature scale;
[0038] Use the first fully connected layer to reduce the feature scale of the flattened first output vector to obtain a second output vector;
[0039] Input the second output vector into the self-attention module for processing to obtain a third output vector;
[0040] Input the third output vector into the second fully connected layer for score prediction to obtain the predicted score.
[0041] In the embodiment of the present application, by inputting the first objective score matrix for training into the to-be-trained vehicle sound quality evaluation model for processing, a predicted score is obtained, which is used to calculate the training loss value between the predicted score and the manual score and update the to-be-trained vehicle sound quality evaluation model subsequently.
[0042] Optionally, the step of updating the to-be-trained vehicle sound quality evaluation model according to the training loss value includes:
[0043] Update the to-be-trained vehicle sound quality evaluation model according to the training loss value to obtain an updated vehicle sound quality evaluation model;
[0044] Calculate the training loss value corresponding to the updated vehicle sound quality evaluation model;
[0045] Comparing the training loss values corresponding to the updated vehicle sound quality evaluation model to obtain a target vehicle sound quality evaluation model;
[0046] Obtain test data and test indicators;
[0047] Calculating the model index of the target vehicle sound quality evaluation model using the test data;
[0048] If the model index satisfies the test index, the target vehicle sound quality evaluation model is used as the trained vehicle sound quality evaluation model.
[0049] The embodiment of the present application selects a target vehicle sound quality evaluation model whose model indicators meet the test indicators as the trained vehicle sound quality evaluation model, which can improve the accuracy of data subsequently calculated by the vehicle sound quality evaluation model, thereby improving the efficiency and accuracy of vehicle sound quality evaluation.
[0050] The embodiment of the present application also discloses a device for determining vehicle sound evaluation data, including:
[0051] A vehicle sound data acquisition module, used to acquire vehicle sound data, a first objective score corresponding to the vehicle sound data, and a manual score corresponding to the vehicle sound data; the first objective score includes objective scores for different preset evaluation parameters;
[0052] A first model processing module, used for inputting the first objective score into a pre-trained vehicle sound quality evaluation model for processing to obtain a target score corresponding to the vehicle sound data;
[0053] A second objective score acquisition module is used to remove the objective score of the target evaluation parameter from the first objective score to obtain a second objective score; the target evaluation parameter is determined from the evaluation parameter;
[0054] A second model processing module, used for inputting the second objective score into the vehicle sound quality evaluation model for processing to obtain a parameter score corresponding to the vehicle sound data;
[0055] A contribution calculation module, used to calculate the contribution corresponding to the target evaluation parameter according to the parameter score, the manual score and the target loss value, wherein the target loss value is calculated based on the target score and the manual score;
[0056] An evaluation data integration module is used to store the target score and the contribution degree as evaluation data corresponding to the vehicle sound data.
[0057] An embodiment of the present application also discloses an electronic device, including a processor and a memory. The memory is used to store a computer program. The processor is used to execute the program stored in the memory to implement one or more of the methods as described in the embodiments of the present application.
[0058] An embodiment of the present application also discloses a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements one or more of the methods as described in the embodiments of the present application. Description of the Drawings
[0059] Figure 1 is a flowchart of a method for determining vehicle sound evaluation data provided by an embodiment of the present application;
[0060] Figure 2 is a model training structure diagram of a method for determining vehicle sound evaluation data provided by an embodiment of the present application;
[0061] Figure 3 is a model training flowchart of a method for determining vehicle sound evaluation data provided by an embodiment of the present application;
[0062] Figure 4 is an implementation flowchart of a method for determining vehicle sound evaluation data provided by an embodiment of the present application;
[0063] Figure 5 is a structure diagram of a device for determining vehicle sound evaluation data provided by an embodiment of the present application;
[0064] Figure 6 is a structure diagram of an electronic device provided by an embodiment of the present application. Detailed Embodiments
[0065] In order to make the technical problems, technical solutions and beneficial effects solved by the present application more clear, the present application will be further described in detail below in conjunction with embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.
[0066] Term Explanation
[0067] In the present application, the psychoacoustic objective parameter is to quantify the subjective perception of sound by the human ear (such as loudness, sharpness, roughness, etc.) into a computable physical quantity through a mathematical model and a signal processing method. In the embodiments of the present application, the psychoacoustic objective parameter is the evaluation parameter.
[0068] The vehicle sound quality evaluation model is a model obtained through training in the embodiments of the present application for predicting the vehicle sound quality.
[0069] The Bidirectional Long Short-Term Memory Network is an improved Recurrent Neural Network (RNN) specifically designed to process sequential data. Based on the traditional LSTM, it introduces a bidirectional structure that can capture both the forward and backward dependencies of sequential data, thereby enhancing the model's ability to understand context information.
[0070] A method for determining vehicle sound evaluation data provided by an embodiment of the present application includes:
[0071] Obtain vehicle sound data, a first objective score corresponding to the vehicle sound data, and a manual score corresponding to the vehicle sound data; the first objective score includes objective scores for different preset evaluation parameters;
[0072] Input the first objective score into a pre-trained vehicle sound quality evaluation model for processing to obtain a target score corresponding to the vehicle sound data;
[0073] Eliminate the objective scores of the target evaluation parameters from the first objective score to obtain a second objective score; the target evaluation parameters are determined from the evaluation parameters;
[0074] Input the second objective score into the vehicle sound quality evaluation model for processing to obtain a parameter score corresponding to the vehicle sound data;
[0075] Calculate the contribution degree corresponding to the target evaluation parameter according to the parameter score, the manual score, and a target loss value, where the target loss value is calculated based on the target score and the manual score;
[0076] Store the target score and the contribution degree as the evaluation data corresponding to the vehicle sound data.
[0077] In an embodiment of the present application, by inputting the first objective score and the second objective score into the vehicle sound quality evaluation model to obtain the target score and the parameter score, and based on the parameter score, the target score, and the manual score, calculate the contribution degree of each evaluation parameter in the subjective evaluation of vehicle sound quality, thereby establishing a mapping relationship between the psychoacoustic objective evaluation parameters and the subjective evaluation of vehicle sound quality, clarifying the contribution degree of each evaluation parameter in the subjective evaluation, and thus significantly improving the efficiency of vehicle sound quality evaluation. In addition, an embodiment of the present application can predict the vehicle sound quality of a newly designed vehicle, providing a convenient and efficient evaluation tool for automotive acoustic design, which is of great significance for the optimization and guidance of vehicle sound design.
[0078] Embodiment 1
[0079] An embodiment of the present application provides a method for determining vehicle sound evaluation data. Please refer to Figure 1, including the following steps:
[0080] S110: Obtain vehicle sound data, a first objective score corresponding to the vehicle sound data, and a manual score corresponding to the vehicle sound data; the first objective score includes objective scores for different preset evaluation parameters.
[0081] S120: Input the first objective score into a pre-trained vehicle sound quality evaluation model for processing to obtain a target score corresponding to the vehicle sound data.
[0082] S130: Eliminate the objective scores of the target evaluation parameter in the first objective score to obtain a second objective score; the target evaluation parameter is determined from the evaluation parameters.
[0083] S140: Input the second objective score into the vehicle sound quality evaluation model for processing to obtain a parameter score corresponding to the vehicle sound data.
[0084] S150: Calculate the contribution degree corresponding to the target evaluation parameter according to the parameter score, the manual score, and a target loss value, where the target loss value is calculated according to the target score and the manual score.
[0085] S160: Store the target score and the contribution degree as evaluation data corresponding to the vehicle sound data.
[0086] In the embodiments of the present application, the vehicle sound can be the vehicle acceleration sound or other types of vehicle sounds.
[0087] Before step S110 in the embodiments of the present application, the vehicle sound quality evaluation model can be trained first according to the vehicle sound data in the training data and its corresponding first objective score and manual score, so as to obtain a pre-trained vehicle sound quality evaluation model. Specifically, input the first objective score into the vehicle sound quality evaluation model for processing to obtain a predicted score, and calculate the training loss value between the predicted score and the manual score, and update the vehicle sound quality evaluation model using the training loss value until the vehicle sound quality evaluation model meets the preset requirements, so as to obtain a vehicle sound quality evaluation model that reflects the mapping relationship between the psychoacoustic objective parameters and subjective evaluations of vehicle sound quality, that is, the mapping relationship between the objective score and the manual score. And the vehicle sound quality evaluation model of the present application can predict the quality of newly designed vehicle sounds, providing a convenient and efficient evaluation tool for automotive acoustic design.
[0088] In step S110, vehicle sound data is obtained, and a first objective score and a manual score corresponding to the vehicle sound data are obtained. The first objective score includes objective scores for different preset evaluation parameters. Psychoacoustic objective parameters (i.e., evaluation parameters) comprehensively consider both the auditory characteristics of the human ear and the physical characteristics of sound, and can systematically describe the differences in subjective feelings generated by people under the influence of different sounds. In one embodiment, the objective scores of different evaluation parameters for each piece of vehicle sound data in the database can be calculated, including pitch, clarity index, roughness, loudness, fluctuation, sharpness, etc. The above list of evaluation parameters is only an example, and the embodiments of the present application do not impose any restrictions on specific evaluation parameters.
[0089] In step S120, the first objective score is input into a pre-trained vehicle sound quality evaluation model for processing, and a target score corresponding to the vehicle sound data predicted by the vehicle sound quality evaluation model is obtained. The target score describes the quality of the vehicle sound data predicted by the vehicle sound quality evaluation model.
[0090] In steps S130 and S140, a target evaluation parameter is determined from the evaluation parameters, and the objective score of the target evaluation parameter in the first objective score is removed to obtain a second objective score. The second objective score is input into the vehicle sound quality evaluation model for processing, and a parameter score predicted by the vehicle sound quality evaluation model based on the second objective score is obtained.
[0091] Specifically, in this embodiment, the objective scores of the evaluation parameters included in the objective score input into the vehicle sound quality evaluation model can be removed one by one, and the parameter scores in several cases can be calculated. For example, if the first objective score has objective scores of M evaluation parameters, then in the first case, the objective score of evaluation parameter P1 can be removed, and the input to the vehicle sound quality evaluation model is [P2, P3,..., P M , in the second case, the objective score of evaluation parameter P2 can be removed, and the input to the vehicle sound quality evaluation model is [P1, P3,..., P M , and so on, to obtain the parameter scores corresponding to M evaluation parameters respectively.
[0092] In step S150, a target loss value corresponding to the target score is calculated based on the target score and the manual score, and further, the contribution degrees corresponding to different evaluation parameters are calculated based on the parameter scores, manual scores, and target loss values respectively corresponding to different evaluation parameters, so as to obtain the contribution degrees of each objective evaluation parameter to the subjective perception evaluation of people.
[0093] In step S160, the evaluation data corresponding to the vehicle sound data is the target score and contribution degree obtained in the above steps.
[0094] In an embodiment of the present application, vehicle sound data, a first objective score corresponding to the vehicle sound data, and a manual score corresponding to the vehicle sound data are obtained; the first objective score includes objective scores for different preset evaluation parameters; the first objective score is input into a pre-trained vehicle sound quality evaluation model for processing to obtain a target score corresponding to the vehicle sound data; the objective score of the target evaluation parameter in the first objective score is removed to obtain a second objective score; the target evaluation parameter is determined from the evaluation parameters; the second objective score is input into the vehicle sound quality evaluation model for processing to obtain a parameter score corresponding to the vehicle sound data; the contribution degree corresponding to the target evaluation parameter is calculated according to the parameter score, the manual score, and a target loss value, and the target loss value is calculated according to the target score and the manual score; the target score and the contribution degree are stored as evaluation data corresponding to the vehicle sound data. In the embodiment of the present application, by inputting the first objective score and the second objective score into the vehicle sound quality evaluation model to obtain the target score and the parameter score, and based on the parameter score, the target score, and the manual score, the contribution degree of each evaluation parameter in the subjective evaluation of vehicle sound quality is calculated, thereby establishing a mapping relationship between the objective evaluation parameters of vehicle sound quality in psychoacoustics and the subjective evaluation, clarifying the contribution degree of each evaluation parameter in the subjective evaluation, and thus significantly improving the efficiency of vehicle sound quality evaluation. In addition, the embodiment of the present application can predict the vehicle sound quality of a newly designed vehicle, provides a convenient and efficient evaluation tool for automotive acoustic design, and is of great significance for the optimization and guidance of vehicle sound design.
[0095] In an alternative embodiment of the present application, step S110 includes:
[0096] Obtain the vehicle sound data to be processed and a preset overlap rate;
[0097] Slice the vehicle sound data to be processed according to the preset overlap rate to obtain vehicle sound data slices, and there is a corresponding loudness for the vehicle sound data slices;
[0098] Adjust the loudness of the vehicle sound data slices to obtain the vehicle sound data;
[0099] Calculate the first objective score and the manual score according to the vehicle sound data.
[0100] In this embodiment, the vehicle sound data can be preprocessed first, and the vehicle sound data can be the sound of a vehicle accelerating.
[0101] Specifically, first, collect in-vehicle noise audio samples under the vehicle acceleration condition as the vehicle sound data to be processed. In one example, each sample can be longer than 10 seconds (s), and ensure that the samples cover a variety of vehicle models. Then, slice the vehicle sound data to be processed according to a preset overlap rate. For example, if a vehicle sound data to be processed is overlapped and sliced into 5-second vehicle sound data slices at an overlap rate of 20%, it can be determined that the start time point of each vehicle sound data slice is 1 second earlier than the previous one. For example, the first segment is from 0 s to 5 s, the second segment is from 4 s to 9 s, and so on, repeating the slicing until the audio ends. After that, adjust the loudness of the vehicle sound data slices by adjusting the signal gain. In one example, it can be adjusted so that the loudness of the vehicle sound data slices all reaches 15 sone (a psychoacoustic unit used to describe the loudness perceived by the human ear).
[0102] After completing the collection and preprocessing of the vehicle sound data, the first objective score and the human score can be calculated for the vehicle sound data. In one embodiment, the calculated first objective score can include the objective scores of evaluation parameters such as pitch, clarity index, roughness, loudness, fluctuation degree, sharpness, etc. Since the non-stationary noise within a short period can be approximately regarded as stationary noise, the first objective score of the vehicle sound data can be calculated using the traditional psychoacoustic objective parameter calculation method. The first objective score can be divided into time frames with a time length of 100 milliseconds (ms), and the overlap rate of each frame is 50%, and the objective scores of different evaluation parameters are calculated.
[0103] The embodiments of the present application preprocess the vehicle sound data and calculate the corresponding first objective score and human score for use in subsequent vehicle sound quality evaluation model training and contribution degree calculation.
[0104] In an alternative embodiment of the present application, calculating the human score according to the vehicle sound data further includes:
[0105] Obtain the initial human score;
[0106] Filter the abnormal scores in the initial human score to obtain the filtered human score;
[0107] Calculate the average value of the filtered human score corresponding to the vehicle sound data, and use the average value as the human score corresponding to the vehicle sound data.
[0108] In this embodiment, the human score can be obtained after processing the initial human score for the vehicle sound data. In one embodiment, when obtaining the initial human score, a subjective evaluation method can be adopted, inviting multiple test subjects to use a 10-level scoring method to score the sound quality of each vehicle sound data to obtain the initial human score.
[0109] In one embodiment, after obtaining the initial manual score, the quartile method can be used to eliminate abnormal data. Specifically, the quartile includes three key parameters, namely the values ranked 25% (Q1), 50% (Q2), and 75% (Q3) in the dataset. First, calculate the Q1, Q2, and Q3 values of the initial manual score corresponding to each vehicle sound data. Then, calculate the interquartile range IQR according to formula (1):
[0110] IQR = Q3 - Q1 (1)
[0111] where Q1 is the score ranked 25% in the initial manual score corresponding to the vehicle sound data, Q3 is the score ranked 75% in the initial manual score corresponding to the vehicle sound data, and IQR is the interquartile range.
[0112] Finally, filter out the initial manual scores outside the range of [Q1 - 1.5×IQR, Q3 + 1.5×IQR] as abnormal scores to obtain the filtered manual scores, and take the average of the filtered manual scores to obtain the manual score corresponding to the final vehicle sound data. In one embodiment, the establishment of an automotive acceleration noise database can be further completed accordingly, where the data is vehicle sound data of equal length and the same loudness, and the label is the manual score corresponding to the vehicle sound data.
[0113] In the embodiment of the present application, the initial manual score is obtained through a subjective evaluation experiment and processed to obtain the manual score corresponding to the vehicle sound data, improving the accuracy of the manual score corresponding to the vehicle sound data, which is thus used in subsequent vehicle sound quality evaluation model training and contribution degree calculation.
[0114] In an alternative embodiment of the present application, step S150 includes:
[0115] Calculate the parameter loss value between the parameter score and the manual score;;
[0116] Calculate the contribution degree corresponding to the target evaluation parameter according to the parameter loss value and the target loss value.
[0117] In this embodiment, after obtaining the parameter score, the parameter loss value between the parameter score and the manual score can be calculated for each case where an evaluation parameter is missing. According to the parameter loss value and the target loss value, calculate the degree of decrease in the loss value, thereby obtaining the contribution degree of each evaluation parameter. Specifically, the contribution degree can be calculated using formula (2):
[0118]
[0119] where C m is the contribution degree of the m-th evaluation parameter, M is the number of types of evaluation parameters, L mis the parameter loss value between the parameter score calculated for the case of missing the m-th evaluation parameter and the manual score, and L0 is the target loss value.
[0120] Since the vehicle sound quality evaluation model itself reflects the mapping relationship between the psychoacoustic objective parameters of vehicle sound quality and subjective evaluation, the embodiments of the present application can calculate the degree of decrease in the loss value based on the parameter loss value and the target loss value calculated between the output of the vehicle sound quality evaluation model and the manual score, so as to obtain the contribution degree of each evaluation parameter in the subjective perception of people, thereby significantly improving the vehicle sound quality evaluation efficiency.
[0121] In an optional embodiment of the present application, the vehicle sound quality evaluation model is trained in the following manner:
[0122] Obtain the vehicle sound quality evaluation model to be trained;
[0123] Generate a first objective score matrix corresponding to the first objective score;
[0124] Input the first objective score matrix into the vehicle sound quality evaluation model to be trained for processing to obtain a predicted score;
[0125] Calculate the training loss value between the predicted score and the manual score;
[0126] Update the vehicle sound quality evaluation model to be trained according to the training loss value to obtain a trained vehicle sound quality evaluation model.
[0127] In this embodiment, it is necessary to train the vehicle sound quality evaluation model to be trained to obtain a trained vehicle sound quality evaluation model. Specifically, first generate a first objective score matrix corresponding to the first objective score. In one embodiment, if the first objective score includes the objective scores of M evaluation parameters, the objective scores of the M evaluation parameters can be stacked row by row to obtain a matrix with a dimension of M×T n where T n is the reciprocal of the vehicle sound data sampling rate.
[0128] After inputting the first objective score matrix into the vehicle sound quality evaluation model to be trained for processing, a predicted score for the vehicle sound data output by the model is obtained, the training loss value between the predicted score and the manual score is calculated, and the vehicle sound quality evaluation model to be trained is updated according to the training loss value, so as to obtain a trained vehicle sound quality evaluation model. In one embodiment, the mean square error function can be selected as the loss function, and the model parameters are iteratively updated based on the backpropagation of gradient descent. The mean square error loss function can be expressed as the following formula (3). The mean square error function can also be selected as the loss function when calculating the parameter loss value and the target loss value in the embodiments of the present application.
[0129]
[0130] where x i and y i respectively represent the predicted score and the corresponding human score of the i-th vehicle sound data, N is the number of samples, and MSE is the training loss value.
[0131] In the embodiment of the present application, the vehicle sound quality evaluation model is trained according to the first objective score and the human score, so as to establish the mapping relationship between the psychoacoustic objective parameters and the subjective evaluation of the vehicle sound quality, and a vehicle sound quality evaluation model is established to improve the vehicle sound quality evaluation efficiency.
[0132] In an alternative embodiment of the present application, the vehicle sound quality evaluation model to be trained includes a first fully connected layer, a second fully connected layer, a bidirectional long short-term memory network module, and a self-attention module. Inputting the first objective score matrix into the vehicle sound quality evaluation model to be trained for processing to obtain a predicted score includes:
[0133] Inputting the first objective score matrix into the bidirectional long short-term memory network module for processing to obtain a first output vector;
[0134] Flatten the first output vector to obtain a flattened first output vector, and there is a corresponding feature scale for the flattened first output vector;
[0135] Use the first fully connected layer to reduce the feature scale of the flattened first output vector to obtain a second output vector;
[0136] Input the second output vector into the self-attention module for processing to obtain a third output vector;
[0137] Input the third output vector into the second fully connected layer for score prediction to obtain the predicted score.
[0138] LSTM is a special type of recurrent neural network suitable for processing and predicting sequential data. LSTM contains memory units and gating mechanisms (input gate, forget gate, output gate), and this kind of structure is beneficial to capturing the long-term and short-term interdependencies of signals, so it is suitable for time series processing. The unidirectional LSTM network predicts the output of the next moment based on the sequential signals of the previous moments. However, in the subjective evaluation task of vehicle sound quality, the subjective feeling at the current moment is not only related to the previous state, but may also be related to the future state. Therefore, the embodiment of the present application proposes to apply a bidirectional LSTM network to perform deep feature extraction on the forward and reverse sequential information.
[0139] Specifically, the first objective scoring matrix can be denoted as P, and the dimension of P is assumed to be M×T n . First, input the matrix P into the bidirectional long short-term memory network module for processing to obtain the first output vector, and flatten the obtained first output vector into a 1D vector OUT1 to obtain the flattened first output vector, as shown in formula (4):
[0140] OUT1 = Flatten(BiLSTM(P)) (4)
[0141] Where Flatten() is the flattening operation, BiLSTM() is the processing of the bidirectional long short-term memory network module, P is the first objective scoring matrix, and OUT1 is the flattened first output vector.
[0142] Then, input the flattened first output vector OUT1 into the first fully connected layer to reduce its feature scale, and obtain a 1D vector OUT2, that is, the second output vector, as shown in formula (5).
[0143] OUT2 = h1(w1 * OUT1 + b1) (5)
[0144] Where h1 is the activation function, w1 and b1 are the weights and biases of the first fully connected layer, OUT1 is the flattened first output vector, and OUT2 is the second output vector.
[0145] After that, input OUT2 into the self-attention module for processing to obtain the third output vector. Specifically, in the self-attention module, multiply OUT2 by its own transpose to obtain a matrix, and then perform normalization processing on the matrix to obtain the self-attention weight OUT3, as shown in formula (6).
[0146] OUT3 = Norm(OUT2 × OUT2 T ) (6)
[0147] Where Norm() represents the normalization operation, OUT2 is the second output vector, OUT2 T is the transpose matrix of the second output vector, and OUT3 is the self-attention weight.
[0148] In the self-attention module, further multiply OUT3 by OUT2 to obtain the third output vector OUT4, as shown in formula (7).
[0149] OUT4 = OUT3 × OUT2 (7)
[0150] Where OUT4 is the third output vector, OUT2 is the second output vector, and OUT3 is the self-attention weight.
[0151] Finally, input OUT4 into the second fully connected layer to predict the sound quality score, and the predicted score is OUT, as shown in formula (8).
[0152] OUT =h2(w2*OUT4+b2) (8)
[0153] Among them, h2 is the activation function, w2 and b2 are the weights and biases of the second fully connected layer, OUT is the predicted score, and OUT4 is the third output vector.
[0154] In the embodiment of the present application, by inputting the first objective scoring matrix for training into the vehicle sound quality evaluation model to be trained for processing, a predicted score is obtained, which is used to calculate the training loss value between the predicted score and the manual score and update the vehicle sound quality evaluation model to be trained in the subsequent calculation.
[0155] In an alternative embodiment of the present application, updating the vehicle sound quality evaluation model to be trained according to the training loss value includes:
[0156] Update the vehicle sound quality evaluation model to be trained according to the training loss value to obtain an updated vehicle sound quality evaluation model;
[0157] Calculate the training loss value corresponding to the updated vehicle sound quality evaluation model;
[0158] Compare the training loss value corresponding to the updated vehicle sound quality evaluation model to obtain a target vehicle sound quality evaluation model;
[0159] Obtain test data and test metrics;
[0160] Calculate the model metrics of the target vehicle sound quality evaluation model using the test data;
[0161] If the model metrics meet the test metrics, then use the target vehicle sound quality evaluation model as the trained vehicle sound quality evaluation model.
[0162] In this embodiment, after using the training loss value to update the vehicle sound quality evaluation model to be trained to obtain an updated vehicle sound quality evaluation model, the model can be further compared and tested. Compare the training loss value corresponding to the updated vehicle sound quality evaluation model, retain the model parameters with the smallest training loss value during the training process to obtain a target vehicle sound quality evaluation model. Calculate the model metrics of the target vehicle sound quality evaluation model according to the vehicle sound data and its corresponding first objective score and manual score in the test data. If the model metrics meet the test metrics, then use the target vehicle sound quality evaluation model as the trained vehicle sound quality evaluation model.
[0163] In one embodiment, the Pearson correlation coefficient can be selected as the test index for evaluating the target vehicle sound quality evaluation model, and the Pearson correlation coefficient ρ X,Y can measure the linear relationship between two random variables, can characterize the prediction effect of the model, and can be expressed as formula (9).
[0164]
[0165] wherein, X and Y respectively represent the predicted value vector and the true value vector, cov(X, Y) represents the covariance between the two, and σ X and σ Y represent the standard deviations of the predicted value vector and the true value vector respectively.
[0166] In the embodiment of the present application, the target vehicle sound quality evaluation model whose model index meets the test index is selected as the trained vehicle sound quality evaluation model, which can improve the accuracy of the data calculated by the vehicle sound quality evaluation model subsequently, thereby improving the efficiency and accuracy of vehicle sound quality evaluation.
[0167] The following will combine Figure 2 with Figure 3 to further explain the training process of the vehicle sound quality evaluation model in the embodiment of the present application.
[0168] Referring to Figure 2 , it is the model training structure diagram of the method for determining vehicle sound evaluation data provided by an embodiment of the present application.
[0169] As Figure 2 shown, the specific structure and steps passed through when the first objective scoring matrix (i.e., the time-varying psychoacoustic parameter matrix) is input into the vehicle sound quality evaluation model to be trained for processing can be expressed as follows:[[]]
[0170] Step 201: Input the time-varying psychoacoustic parameter matrix into the BiLSTM module for processing to obtain the first output vector.
[0171] Step 202: Flatten the first output vector to obtain the flattened first output vector.
[0172] Step 203: Use the first fully connected layer to reduce the feature scale of the flattened first output vector to obtain the second output vector.
[0173] Step 204: Input the second output vector into the self-attention module for processing such as vector multiplication, normalization, and matrix multiplication to obtain the third output vector.
[0174] Step 205: Input the third output vector into the second fully connected layer for scoring prediction to obtain the predicted score output.
[0175] Referring to Figure 3, which is the model training flowchart of the method for determining vehicle sound evaluation data provided by an embodiment of the present application.
[0176] Figure 3 The specific training process of the vehicle sound quality evaluation model is shown, which may specifically include the following steps.
[0177] Step 301: Divide the data set to obtain training data and test data.
[0178] Step 302: Build a bidirectional LSTM model combined with the self-attention mechanism.
[0179] Step 303: Initialize the model parameters to obtain the vehicle sound quality evaluation model to be trained.
[0180] Step 304: Use the training data to train the vehicle sound quality evaluation model to be trained.
[0181] Step 305: After inputting the vehicle sound quality evaluation model to be trained, perform adaptive feature extraction based on the LSTM and self-attention mechanism modules.
[0182] Step 306: Output the prediction of the sound quality score based on the fully connected layer.
[0183] Step 307: Determine whether the set number of iterations is reached. If not, jump to Step 304; if so, jump to Step 308.
[0184] Step 308: Use the test data to test the vehicle sound quality evaluation model.
[0185] Step 309: Output the loss function value and evaluation index.
[0186] In an embodiment of the present application, vehicle sound data, a first objective score corresponding to the vehicle sound data, and a manual score corresponding to the vehicle sound data are obtained; the first objective score includes objective scores for different preset evaluation parameters; the first objective score is input into a pre-trained vehicle sound quality evaluation model for processing to obtain a target score corresponding to the vehicle sound data; the objective scores of the target evaluation parameters in the first objective score are removed to obtain a second objective score; the target evaluation parameter is determined from the evaluation parameters; the second objective score is input into the vehicle sound quality evaluation model for processing to obtain a parameter score corresponding to the vehicle sound data; the contribution degree corresponding to the target evaluation parameter is calculated according to the parameter score, the manual score, and the target loss value, and the target loss value is calculated according to the target score and the manual score; the target score and the contribution degree are stored as evaluation data corresponding to the vehicle sound data. In the embodiment of the present application, by inputting the first objective score and the second objective score into the vehicle sound quality evaluation model to obtain the target score and the parameter score, and based on the parameter score, the target score, and the manual score, the contribution degree of each evaluation parameter in the subjective evaluation of vehicle sound quality is calculated, thereby establishing a mapping relationship between the psychoacoustic objective evaluation parameters of vehicle sound quality and the subjective evaluation, clarifying the contribution degree of each evaluation parameter in the subjective evaluation, and thus significantly improving the efficiency of vehicle sound quality evaluation. In addition, the embodiment of the present application can predict the vehicle sound quality of a newly designed vehicle, provides a convenient and efficient evaluation tool for automotive acoustic design, and has important significance for the optimization and guidance of vehicle sound design.
[0187] Embodiment 2
[0188] Refer to Figure 4 , which is a flowchart of an implementation of a method for determining vehicle sound evaluation data provided by an embodiment of the present application.
[0189] In order to enable those skilled in the art to have a more comprehensive understanding of the technical solution of the embodiment of the present application, according to Figure 4 a complete description of a method for determining vehicle sound evaluation data shown in the embodiment of the present application is provided, including the following steps.
[0190] Step 401, establishing an accelerated audio database: data collection, data augmentation, and data preprocessing.
[0191] Step 402, calculating time-varying psychoacoustic objective parameters (i.e., the first objective score), including pitch, articulation index, roughness, loudness, fluctuation, sharpness, etc.
[0192] Step 403, establishing a bidirectional LSTM model combined with a self-attention mechanism.
[0193] Step 404, setting model parameter evaluation indicators, performing model training, and retaining the optimal parameters during the training process, so as to obtain a trained vehicle sound quality evaluation model.
[0194] Step 405: Gradually reduce the types of input features and input them into the optimal model, and record the loss value for each case.
[0195] Step 406: Sort the psychoacoustic objective parameters according to the loss value, and calculate the contribution degree of each feature according to the degree of loss value decrease.
[0196] In the embodiment of the present application, vehicle sound data, a first objective score corresponding to the vehicle sound data, and an artificial score corresponding to the vehicle sound data are obtained; the first objective score includes objective scores for different preset evaluation parameters; the first objective score is input into a pre-trained vehicle sound quality evaluation model for processing to obtain a target score corresponding to the vehicle sound data; the objective score of the target evaluation parameter in the first objective score is removed to obtain a second objective score; the target evaluation parameter is determined from the evaluation parameters; the second objective score is input into the vehicle sound quality evaluation model for processing to obtain a parameter score corresponding to the vehicle sound data; the contribution degree corresponding to the target evaluation parameter is calculated according to the parameter score, the artificial score, and the target loss value, and the target loss value is calculated according to the target score and the artificial score; the target score and the contribution degree are stored as evaluation data corresponding to the vehicle sound data. In the embodiment of the present application, by inputting the first objective score and the second objective score into the vehicle sound quality evaluation model to obtain the target score and the parameter score, and based on the parameter score, the target score, and the artificial score, the contribution degree of each evaluation parameter in the subjective evaluation of vehicle sound quality is calculated, thereby establishing a mapping relationship between the psychoacoustic objective evaluation parameters of vehicle sound quality and the subjective evaluation, clarifying the contribution degree of each evaluation parameter in the subjective evaluation, and thus significantly improving the efficiency of vehicle sound quality evaluation. In addition, the embodiment of the present application can predict the vehicle sound quality of a newly designed vehicle, provides a convenient and efficient evaluation tool for automotive acoustic design, and is of great significance for the optimization and guidance of vehicle sound design.
[0197] The embodiment of the present application also provides a device 50 for determining vehicle sound evaluation data. Please refer to Figure 5 , including:
[0198] A vehicle sound data acquisition module 510, configured to acquire vehicle sound data, a first objective score corresponding to the vehicle sound data, and an artificial score corresponding to the vehicle sound data; the first objective score includes objective scores for different preset evaluation parameters;
[0199] A first model processing module 520, configured to input the first objective score into a pre-trained vehicle sound quality evaluation model for processing to obtain a target score corresponding to the vehicle sound data;
[0200] A second objective score acquisition module 530, configured to remove the objective score of the target evaluation parameter in the first objective score to obtain a second objective score; the target evaluation parameter is determined from the evaluation parameters;
[0201] The model second processing module 540 is configured to input the second objective score into the vehicle sound quality evaluation model for processing to obtain a parametric score corresponding to the vehicle sound data;
[0202] The contribution degree calculation module 550 is configured to calculate the contribution degree corresponding to the target evaluation parameter according to the parametric score, the manual score, and the target loss value, where the target loss value is calculated according to the target score and the manual score;
[0203] The evaluation data integration module 560 is configured to store the target score and the contribution degree as evaluation data corresponding to the vehicle sound data.
[0204] Optionally, the vehicle sound data acquisition module 510 includes:
[0205] The to-be-processed vehicle sound acquisition sub-module is configured to acquire to-be-processed vehicle sound data and a preset overlap rate;
[0206] The vehicle sound splitting sub-module is configured to overlap and split the to-be-processed vehicle sound data according to the preset overlap rate to obtain vehicle sound data slices, and there is a corresponding loudness for the vehicle sound data slices;
[0207] The loudness adjustment sub-module is configured to adjust the loudness of the vehicle sound data slices to obtain the vehicle sound data;
[0208] The score calculation sub-module is configured to calculate the first objective score and the manual score according to the vehicle sound data.
[0209] Optionally, the score calculation sub-module is further configured to:
[0210] Obtain an initial manual score;
[0211] Filter abnormal scores in the initial manual score to obtain a filtered manual score;
[0212] Calculate the average value of the filtered manual scores corresponding to the vehicle sound data, and use the average value as the manual score corresponding to the vehicle sound data.
[0213] Optionally, the contribution degree calculation module 550 includes:
[0214] The parametric loss value calculation sub-module is configured to calculate the parametric loss value between the parametric score and the manual score;
[0215] The contribution degree determination sub-module is configured to calculate the contribution degree corresponding to the target evaluation parameter according to the parametric loss value and the target loss value.
[0216] Optionally, the apparatus 50 further includes:
[0217] A to-be-trained model acquisition module, configured to acquire a vehicle sound quality evaluation model to be trained;
[0218] A scoring matrix generation module, configured to generate a first objective scoring matrix corresponding to the first objective score according to the first objective score;
[0219] A predicted score acquisition module, configured to input the first objective scoring matrix into the vehicle sound quality evaluation model to be trained for processing to obtain a predicted score;
[0220] A training loss calculation module, configured to calculate a training loss value between the predicted score and the manual score;
[0221] A model update module, configured to update the vehicle sound quality evaluation model to be trained according to the training loss value to obtain a trained vehicle sound quality evaluation model.
[0222] Optionally, the vehicle sound quality evaluation model to be trained includes a first fully connected layer, a second fully connected layer, a bidirectional long short-term memory network module, and a self-attention module. The predicted score acquisition module includes:
[0223] A bidirectional long short-term memory network sub-module, configured to input the first objective scoring matrix into the bidirectional long short-term memory network module for processing to obtain a first output vector;
[0224] A vector flattening sub-module, configured to flatten the first output vector to obtain a flattened first output vector, and there is a corresponding feature scale for the flattened first output vector;
[0225] A feature scale processing sub-module, configured to use the first fully connected layer to reduce the feature scale of the flattened first output vector to obtain a second output vector;
[0226] A self-attention processing sub-module, configured to input the second output vector into the self-attention module for processing to obtain a third output vector;
[0227] A score prediction sub-module, configured to input the third output vector into the second fully connected layer for score prediction to obtain the predicted score.
[0228] Optionally, the model update module includes:
[0229] A to-be-trained model update sub-module, configured to update the vehicle sound quality evaluation model to be trained according to the training loss value to obtain an updated vehicle sound quality evaluation model;
[0230] A training loss determination sub-module, configured to calculate a training loss value corresponding to the updated vehicle sound quality evaluation model;
[0231] A loss value comparison sub-module, configured to compare the training loss value corresponding to the updated vehicle sound quality evaluation model to obtain a target vehicle sound quality evaluation model;
[0232] An index acquisition sub-module, configured to acquire test data and test indexes;
[0233] A model index calculation sub-module, configured to calculate the model indexes of the target vehicle sound quality evaluation model by using the test data;
[0234] A model evaluation sub-module, configured to, if the model indexes meet the test indexes, use the target vehicle sound quality evaluation model as the trained vehicle sound quality evaluation model.
[0235] In an embodiment of the present application, vehicle sound data, a first objective score corresponding to the vehicle sound data, and an artificial score corresponding to the vehicle sound data are acquired; the first objective score includes objective scores for different preset evaluation parameters; the first objective score is input into a pre-trained vehicle sound quality evaluation model for processing to obtain a target score corresponding to the vehicle sound data; the objective score of the target evaluation parameter in the first objective score is removed to obtain a second objective score; the target evaluation parameter is determined from the evaluation parameters; the second objective score is input into the vehicle sound quality evaluation model for processing to obtain a parameter score corresponding to the vehicle sound data; a contribution degree corresponding to the target evaluation parameter is calculated according to the parameter score, the artificial score, and a target loss value, and the target loss value is calculated according to the target score and the artificial score; the target score and the contribution degree are stored as evaluation data corresponding to the vehicle sound data. In the embodiment of the present application, by inputting the first objective score and the second objective score into the vehicle sound quality evaluation model to obtain the target score and the parameter score, and based on the parameter score, the target score, and the artificial score, the contribution degree of each evaluation parameter in the subjective evaluation of the vehicle sound quality is calculated, so as to establish a mapping relationship between the psychoacoustic objective evaluation parameter of the vehicle sound quality and the subjective evaluation, clarify the contribution degree of each evaluation parameter in the subjective evaluation, and thus significantly improve the vehicle sound quality evaluation efficiency. In addition, the embodiment of the present application can predict the newly designed vehicle sound quality, provides a convenient and efficient evaluation tool for automotive acoustic design, and has important significance for the optimization and guidance of vehicle sound design.
[0236] The embodiment of the present application further provides an electronic device 60, please refer to Figure 6 , including a processor 610 and a memory 620, wherein the memory 610 is configured to store a computer program; the processor 620 is configured to execute the program stored on the memory 610 to implement the method for determining vehicle sound evaluation data introduced in any embodiment of the present application.
[0237] An embodiment of the present application further provides a computer-readable storage medium, in which a computer program is stored, and when the computer program is executed by a processor, the method for determining vehicle sound evaluation data introduced in any embodiment of the present application is implemented.
[0238] In the present application, "a plurality" means two or more.
[0239] In the present application, unless otherwise clearly defined, the terms "installed", "connected", and "coupled" should be understood in a broad sense. For example, it may be a fixed connection, a detachable connection, or an integral connection; it may be a mechanical connection or an electrical connection; it may be directly connected or indirectly connected through an intermediate medium, and it may be the communication inside two components. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to specific situations.
[0240] The terms "first", "second", "third", "fourth", etc. (if any) in the present application are used to distinguish similar objects and do not necessarily describe a specific order or sequence.
[0241] The term "and / or" in the present application is only a description of the association relationship between associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " in the present application generally represents an "or" relationship between the associated objects before and after.
[0242] If there is no special description, all steps of the present application can be carried out in sequence or randomly. For example, the method includes steps A and B, indicating that the method may include steps A and B carried out in sequence, or steps B and A carried out in sequence. For example, it is mentioned that the method may further include step C, indicating that step C can be added to the method in any order. For example, the method may include steps A, B, and C, or steps A, C, and B, or steps C, A, and B, etc.
[0243] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present application should be included in the protection scope of the present application.
Claims
1. A method for determining vehicle sound evaluation data, characterized in that, Including: Obtain vehicle sound data, a first objective score corresponding to the vehicle sound data, and a manual score corresponding to the vehicle sound data; The first objective score includes objective scores for different preset evaluation parameters; Input the first objective score into a pre-trained vehicle sound quality evaluation model for processing to obtain a target score corresponding to the vehicle sound data; Exclude the objective scores of the target evaluation parameters in the first objective score to obtain a second objective score; The target evaluation parameter is determined from the evaluation parameters; Input the second objective score into the vehicle sound quality evaluation model for processing to obtain a parameter score corresponding to the vehicle sound data; Calculate the contribution degree corresponding to the target evaluation parameter according to the parameter score, the manual score, and a target loss value, where the target loss value is calculated based on the target score and the manual score; Store the target score and the contribution degree as evaluation data corresponding to the vehicle sound data.
2. The method according to claim 1, characterized in that, The obtaining of the vehicle sound data, the first objective score corresponding to the vehicle sound data, and the manual score corresponding to the vehicle sound data includes: Obtain vehicle sound data to be processed and a preset overlap rate; Overlap and segment the vehicle sound data to be processed according to the preset overlap rate to obtain vehicle sound data slices, and there is a corresponding loudness for the vehicle sound data slices; Adjust the loudness of the vehicle sound data slices to obtain the vehicle sound data; Calculate the first objective score and the manual score according to the vehicle sound data.
3. The method according to claim 2, wherein The calculating of the manual score according to the vehicle sound data further includes: Obtain an initial manual score; Filter abnormal scores in the initial manual score to obtain a filtered manual score; Calculate the average value of the filtered manual scores corresponding to the vehicle sound data, and use the average value as the manual score corresponding to the vehicle sound data.
4. The method according to claim 1, wherein The calculating of the contribution degree corresponding to the target evaluation parameter according to the parameter score, the manual score, and the target loss value includes: Calculate a parameter loss value between the parameter score and the manual score; Calculate the contribution degree corresponding to the target evaluation parameter according to the parameter loss value and the target loss value.
5. The method according to claim 1, wherein The vehicle sound quality evaluation model is trained in the following manner: Obtain a vehicle sound quality evaluation model to be trained; Generate a first objective score matrix corresponding to the first objective score according to the first objective score; Input the first objective score matrix into the vehicle sound quality evaluation model to be trained for processing to obtain a predicted score; Calculate a training loss value between the predicted score and the manual score; Update the vehicle sound quality evaluation model to be trained according to the training loss value to obtain a trained vehicle sound quality evaluation model.
6. The method according to claim 5, characterized in that, The vehicle sound quality evaluation model to be trained includes a first fully connected layer, a second fully connected layer, a bidirectional long short-term memory network module, and a self-attention module. The inputting of the first objective score matrix into the vehicle sound quality evaluation model to be trained for processing to obtain a predicted score includes: Input the first objective score matrix into the bidirectional long short-term memory network module for processing to obtain a first output vector; Flatten the first output vector to obtain a flattened first output vector, and there is a corresponding feature scale for the flattened first output vector; Use the first fully connected layer to reduce the feature scale of the flattened first output vector to obtain a second output vector; Input the second output vector into the self-attention module for processing to obtain a third output vector; Input the third output vector into the second fully connected layer for score prediction to obtain the predicted score.
7. The method according to claim 5, characterized in that, The updating the vehicle sound quality evaluation model to be trained according to the training loss value includes: Update the vehicle sound quality evaluation model to be trained according to the training loss value to obtain an updated vehicle sound quality evaluation model; Calculate the training loss value corresponding to the updated vehicle sound quality evaluation model; Compare the training loss value corresponding to the updated vehicle sound quality evaluation model to obtain a target vehicle sound quality evaluation model; Obtain test data and test metrics; Use the test data to calculate the model metrics of the target vehicle sound quality evaluation model; If the model metrics meet the test metrics, use the target vehicle sound quality evaluation model as the trained vehicle sound quality evaluation model.
8. An apparatus for determining vehicle sound evaluation data, characterized in that, Includes: A vehicle sound data acquisition module for acquiring vehicle sound data, the first objective score corresponding to the vehicle sound data, and the manual score corresponding to the vehicle sound data; The first objective score includes objective scores for different preset evaluation parameters; A model first processing module for inputting the first objective score into a pre-trained vehicle sound quality evaluation model for processing to obtain a target score corresponding to the vehicle sound data; A second objective score acquisition module for removing the objective score of the target evaluation parameter from the first objective score to obtain a second objective score; the target evaluation parameter is determined from the evaluation parameters; A model second processing module for inputting the second objective score into the vehicle sound quality evaluation model for processing to obtain a parameter score corresponding to the vehicle sound data; A contribution calculation module for calculating the contribution corresponding to the target evaluation parameter according to the parameter score, the manual score, and a target loss value, where the target loss value is calculated according to the target score and the manual score; An evaluation data integration module for storing the target score and the contribution as evaluation data corresponding to the vehicle sound data.
9. An electronic device, characterized in that, Includes a processor and a memory, where The memory is used to store a computer program; The processor is used to execute the program stored on the memory to implement the method according to any one of claims 1-7.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, the method according to any one of claims 1-7 is implemented.