Music road sound field effect evaluation device and method

Abstract

The application discloses a music road surface sound field effect evaluation device, which comprises a collecting device and a computer electrically connected with the collecting device; the collecting device comprises a microphone set and a multi-channel data collection card; the computer comprises an analysis module and an evaluation module; the analysis module performs principal component analysis and generates a music signal relation graph and an ideal sound signal relation graph; the evaluation module compares the music signal relation graph with the ideal sound signal relation graph, and generates a four-dimensional evaluation result of frequency, loudness, tone and noise by comprehensively outputting a noise evaluation unit, an objective music evaluation unit and a subjective tone evaluation unit; the evaluation module calculates a comprehensive evaluation index; in addition, the application also discloses an evaluation method based on the device, solves the defect that the dynamic change of a sound field in a continuous playing process of a music piece cannot be captured in the prior art, can restore the dynamic rhythm of the music piece, makes the evaluation result more accurate, and realizes more comprehensive and more objective comprehensive evaluation on music road surface design and construction effect.

Description

Technical Field

[0001] This invention relates to the fields of traffic engineering and acoustic testing technology, specifically to a device and method for evaluating the sound field effect of musical road surfaces. Background Technology

[0002] Musical pavements are constructed by regularly arranged grooves or protrusions on the road surface. When a car travels at a certain speed, the tires vibrate under the rhythmic excitation of the pavement, producing sound. The varying depths and spacing of the grooves and protrusions alter the amplitude and frequency of the tire vibrations, creating pleasant sounds and forming a complete melody. Evaluating the sound-producing effect of musical pavements is crucial for their application as an acoustic landscape or warning surface. The sound-producing effect of musical pavements is influenced not only by design but also by construction quality, making it difficult to assess during the design phase, especially for typical application scenarios such as road curves and tunnels. Therefore, a post-evaluation method for the sound-producing effect of musical pavements is urgently needed.

[0003] The key to evaluating the sound quality of a musical instrument lies in extracting the characteristics of its frequency, loudness, and timbre that evolve over time. These evolutionary characteristics include overall harmony, local accuracy, noise levels, and the degree of matching with the target music. In patent 202210450809.9 – A method for evaluating the timbre quality of a pipa based on vibration acoustics – the evaluation system uses an average of expert timbre scores to obtain a subjective score, and also simulates and analyzes the vibrations of the instrument body and strings. However, this evaluation focuses on individual standard syllables and cannot consider the sound field changes of a musical piece. Furthermore, breaking the music down into individual syllables fails to assess the overall characteristics of the piece. In patent 202410648908.7, "A Method, Device, and Car Speaker for In-Vehicle Sound Field Zoning Control," the interior space of an experimental vehicle is zoned and data is collected based on a speaker distribution cluster and multiple seat receivers. This yields sound field distribution data for multiple regions, and the control parameters for each zone are optimized, thereby improving the accuracy of in-vehicle sound field zoning control. However, this patent focuses on the spatial distribution and control of the sound field, and does not address the time-varying characteristics of musical frequencies, loudness, etc.

[0004] Therefore, there is an urgent need to propose a device and method for evaluating the sound field effect of musical roads in order to solve the above problems. Summary of the Invention

[0005] One of the objectives of this invention is to provide a musical road surface sound field effect evaluation device to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a music road surface sound field effect evaluation device, comprising a data acquisition device and a computer electrically connected to the data acquisition device; The acquisition device includes a microphone array and a multi-channel data acquisition card. One end of the multi-channel data acquisition card is electrically connected to the microphone array, and the other end of the multi-channel data acquisition card is electrically connected to a computer. The multi-channel data acquisition card is used to transmit the actual sound signal of the music road surface acquired by the microphone array to the computer. The computer includes an analysis module and an evaluation module. The analysis module is used for: On the one hand, the actual sound signal of the music road surface is preprocessed and synchronously corrected through multiple channels, and principal component analysis is performed to generate music signal relationship diagram and ideal sound signal relationship diagram: the analysis module separates the actual sound signal of the music road surface into noise signal and music signal through principal component analysis; On the other hand, in the analysis module, the music signal is transformed using the Hilbert-Huang transform to generate a music signal relationship graph X2; the analysis module also transforms the target musical score of the music signal using the Hilbert-Huang transform to generate an ideal sound signal relationship graph X1; both the music signal relationship graph X2 and the ideal sound signal relationship graph X1 are based on sound pressure levels. Or an energy-time-frequency relationship diagram; The evaluation module is used for subjective and objective evaluations: On the one hand, time alignment and frequency band matching are performed on the ideal sound signal relationship diagram X1 and the music signal relationship diagram X2, and frequency accuracy index and loudness comfort index are output based on the preset similarity / error metric. On the other hand, the outputs of the integrated noise evaluation unit, objective music evaluation unit, and subjective timbre evaluation unit are used to generate frequency accuracy labels. Loudness and comfort labels Average score of subjective rating Noise evaluation indicators The four-dimensional evaluation results; The evaluation module calculates a comprehensive evaluation index based on subjective and objective evaluations. Comprehensive evaluation indicators The calculation method is as follows: (1) in, The sound pressure level of the noise signal. To objectively evaluate the frequency accuracy labels obtained from the music signal relationship diagram X2. The loudness comfort label was obtained from the music signal relationship diagram X2 for objective evaluation. The average score in the subjective ratings. Noise evaluation index; comprehensive evaluation index Used to reflect the sound field effect of a musical road surface.

[0007] The average score in the subjective rating is calculated based on the subjective timbre evaluation unit, which is derived from the subjective ratings of drivers and passengers on harmony, melody, humanism, and rhythm obtained from the survey.

[0008] Furthermore, the microphone set (1) includes multiple microphones (11). One microphone (11) is fixed at the center of the outer side of one front wheel and one rear wheel of the experimental vehicle. The distance between the microphone (11) and the surface of the front wheel and the surface of the rear wheel is 10cm. Four microphones (11) are fixed at the position of the headrests of the four seats of the experimental vehicle near the ears. The seats include the driver's seat, the passenger seat, the left rear seat and the right rear seat.

[0009] Furthermore, the analysis module performs principal component analysis as follows: A time-frequency matrix is ​​constructed using the Hilbert-Huang transform; the time-frequency matrix is ​​orthogonally decomposed; and the top K principal components are selected for reconstruction based on the cumulative contribution rate threshold. Music and noise signals are then distinguished by the target main frequency band energy ratio or spectral peak consistency, and the signal-to-noise ratio of the music path is calculated based on the music and noise signals. .

[0010] Furthermore, subjective evaluation includes noise evaluation. The noise evaluation process involves the noise evaluation unit calculating the sound pressure level of the noise signal. The noise evaluation index is compared with a preset threshold and then output as a noise evaluation index. The preset threshold is a configurable parameter, which can be 70dB or a threshold range that matches the application scenario.

[0011] Furthermore, the objective music evaluation unit employs a convolutional neural network. The input to the convolutional neural network is a music signal relationship graph X2, and the output is a frequency accuracy label. and loudness comfort label .

[0012] Furthermore, the structure of the convolutional neural network includes: one input layer, three convolutional blocks, one global average pooling layer, one fully connected layer, and two dual-output layers. The input layer is used to input the music signal relationship graph X2; the convolutional blocks contain convolutional layers, normalization layers, and pooling layers, used to extract features from the music signal relationship graph X2; the global average pooling layer is used to reduce the impact of temporal axis differences caused by factors such as time shift and speed fluctuations on feature extraction, thereby improving the robustness of frequency accuracy and loudness comfort evaluation; the fully connected layer is used to fuse the extracted features of the music signal relationship graph X2; and the dual-output layers are used to process the frequency accuracy labels. Loudness and comfort labels Output the results.

[0013] Furthermore, the convolutional block includes a first convolutional block, a second convolutional block, and a third convolutional block. The first convolutional block is used to extract low-level texture features of the music signal relationship graph X2, the second convolutional block is used to obtain mid-level features of the music signal relationship graph X2, and the third convolutional block is used to identify high-level composite features of the music signal relationship graph X2.

[0014] The convolutional layers include a first convolutional layer, a second convolutional layer, and a third convolutional layer. The padding of each convolutional layer is set to 2, and the ReLU activation function is used. A normalization layer is added after each convolutional layer. Dropout layers are set in the second and third convolutional layers, and the dropout ratio is set to 0.3.

[0015] Furthermore, subjective evaluation also includes timbre evaluation: The timbre evaluation process is as follows: Based on the subjective ratings of drivers and passengers regarding harmony, melody, humanism, and rhythm obtained from the survey, the subjective timbre evaluation unit calculates the average score of the subjective ratings. .

[0016] The second objective of this invention is to provide a method for evaluating the sound field effect of a road surface based on the aforementioned sound field effect evaluation device. The steps of the evaluation method are as follows: S1.1 Constructing a Pure Music Signal: Based on the relationship between the time signature and tempo v of the target score, calculate the duration, frequency, and duration of each note to reproduce the notes in the score. Add irregular harmonic signals to the original note signal to change the timbre, and add harmonic signals from different instruments to simulate the playing of different instruments, thereby obtaining... A set of pure music sound signals; S1.2 Building the Convolutional Neural Network Model: The convolutional neural network model is implemented using PyTorch and trained in PyTorch, based on the results obtained in step S1.1. A set of pure music sound signals are used to generate a sequence by gradually modifying the frequency characteristics of the pure music sound signals. , ,…, Based on a pure music signal, the loudness characteristics are gradually modified to generate a sequence. , ,…, The loudness characteristic varies from -10dB to +10dB, thus obtaining 3 A set of pure music sound signal training data; S1.3, 3 The Hilbert-Huang transform was performed on the training data of pure music sound signals to obtain the ideal sound signal relationship graph X1 of group 3a. The ideal sound signal relationship graph X1 was standardized and then input into the input layer of the convolutional neural network for model training to obtain the sample relationship graph. S1.4. The sample relationship graph is input from the input layer into the convolutional block for model training. The convolutional neural network uses two independent loss functions to calculate the loss for sample frequency and sample loudness respectively, and then sums them with weights to obtain the total sample loss. The mean squared error is used to calculate the loss for sample frequency and sample loudness. The weighted formula for the total loss is: (2) in It is the sample frequency label of the i-th sample. It is the predicted frequency label of the i-th sample; It is the sample loudness label of the i-th sample. It is the predicted loudness label of the i-th sample; S1.5, The output layer outputs a two-dimensional vector containing sample frequency labels and sample loudness labels; S2.1. Collecting actual sound signals: When the experimental vehicle passes over the road surface paved with music, it generates actual sound signals of the music road surface. The actual sound signals of the music road surface inside and outside the vehicle are collected by the microphone and stored in the experimental computer through the data acquisition card. S2.2. Six microphones (11 in total) are located at the front wheels, rear wheels, and four headrests of the vehicle. Time-domain sound pressure signals are synchronously acquired via a multi-channel numerical acquisition card. The sampling frequency of the sound signals... f s It is 10kHz.

[0017] S3.1 Conduct a subjective evaluation. The steps for the subjective evaluation are as follows: S3.1.1 The actual sound signal undergoes principal component analysis by the analysis module to obtain the music signal and noise signal. Then, the signal-to-noise ratio of the music road surface is calculated based on the music signal and noise signal. The calculation method is as follows: (3) in, The signal-to-noise ratio of the music track. The average power of the music signal; The average power of the noise signal; S3.1.2 In the noise evaluation unit, based on the signal-to-noise ratio of the music path... Obtain noise evaluation indicators The calculation method is as follows: (4) when When =0, the noise signal does not meet the sound environment quality standards. = At that time, the noise signal met the environmental sound quality standards; S3.1.3 From the perspective of timbre, the driver collects subjective scores based on the listener's subjective feelings. The evaluation indicators for subjective scores include harmony, melody, humanism, and rhythm. The scoring range for each evaluation indicator is 0-1. The subjective timbre evaluation unit calculates the average score based on the subjective scores. ; S3.2 The objective evaluation steps are as follows: In the objective music evaluation unit, the music signal relationship graph X2 is input into the convolutional neural network for evaluation, and the convolutional neural network outputs the accurate frequency label. Loudness and comfort labels Two-dimensional evaluation labels; S3.3 Calculate the comprehensive evaluation index Comprehensive evaluation indicators The comprehensive evaluation index is calculated based on a combination of subjective and objective evaluations. Used to reflect the sound field effect of a musical road surface.

[0018] Furthermore, the output layer uses the Sigmoid activation function to ensure that each item in the output is in the range of 0-1; The formula for calculating the frequency accuracy label is: (5) in, For accurate frequency labeling, The error rate of musical notes on the road surface; The loudness comfort label is calculated as follows: When the loudness of a music signal is greater than or equal to 70dB, the loudness comfort label of the music signal is recorded as 0; When the loudness of the music signal is less than 70dB, the calculation formula is used: (6) in, For loudness comfort label, For music signal loudness labels, For sample loudness; If the output satisfies: 0.8 < <1, 0.4< If the output is less than 1, it means the construction of the music-themed pavement fully meets the design standards; if the output satisfies: 0 < <0.8, 0.4< If the output is less than 1, it indicates that there are many incorrect notes in the musical pavement construction; if the output satisfies: 0.8 < <1, 0< A value of <0.4 indicates that while the musical pavement meets design standards, the human experience is poor.

[0019] Compared with the prior art, the beneficial effects of the present invention are: 1) This application uses a microphone array to acquire the actual sound signal of the music road surface and transmit it to a computer to extract and evaluate the complete music piece. This solves the problem that the existing technology can only evaluate independent standard syllables and cannot capture the dynamic changes of the sound field during the continuous performance of the music piece.

[0020] 2) The analysis module of this application generates sound pressure level from the music signal through Hilbert-Huang transform. Alternatively, the music signal relationship diagram X2 (energy-time-frequency) can be used. The target musical score of the music signal is then transformed using the Hilbert-Huang transform to generate the ideal sound signal relationship diagram X1. The evaluation module compares the ideal sound signal relationship diagram X1 with the music signal relationship diagram X2. This solves the problem that existing technologies only focus on the spatial zoning and control of the sound field and ignore the characteristics of the changes in the frequency and loudness of the music over time. This ensures that the subsequent evaluation can restore the dynamic rhythm of the music and the evaluation results are more accurate.

[0021] 3) This application ultimately generates four-dimensional evaluation results of frequency, loudness, timbre and noise, constructs a complete quantitative evaluation system, and provides a standardized basis for sound quality evaluation; the introduction of signal-to-noise ratio as an evaluation index can more scientifically reflect the clarity of music signals and achieve a more comprehensive and objective evaluation of the design and construction effect of music-themed road surfaces. Attached Figure Description

[0022] Figure 1 This is a diagram of the actual acoustic signal acquisition system of the present invention; Figure 2 This is a flowchart of the method of the present invention; Figure 3 This is a diagram of the convolutional neural network structure of the present invention; Figure 4 This is the ideal acoustic signal relationship diagram of the present invention; Figure 5 This is the music signal relationship diagram of the present invention; Figure 6 This is the intended representation of timbre evaluation in this invention; In the picture: 1- Microphone set; 2- Multi-channel data acquisition card; 3- Computer; 11-Microphone. Detailed Implementation

[0023] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Example 1 Please see Figure 1-3 This embodiment provides a music road surface sound field effect evaluation device, including a data acquisition device and a computer electrically connected to the data acquisition device; The data acquisition device includes a microphone set 1 and a multi-channel data acquisition card. One end of the multi-channel data acquisition card is electrically connected to the microphone set 1, and the other end is electrically connected to the computer 3. The multi-channel data acquisition card is used to transmit the actual sound signal of the music road surface acquired by the microphone set 1 to the computer 3. Furthermore, the microphone set 1 includes multiple microphones 11. One microphone 11 is fixed to the outer center of one front wheel and one rear wheel of the experimental vehicle by clamping or using a strong magnet. The distance between the microphone 11 and the surface of the front wheel and the surface of the rear wheel is 10cm. Four microphones 11 are fixed to the headrests of the four seats of the experimental vehicle near the ears by adhesive or other means. The seats include the driver's seat, the front passenger seat, the left rear seat, and the right rear seat. Preferably, the microphone 11 consists of a microphone, a preamplifier, and a windscreen. Specifically, the microphone 11 uses an AWA14423 microphone and an AWA14603 preamplifier. The microphone's sampling frequency range is 10-20kHz, and its nominal sensitivity is less than 50mV / Pa. The microphone and preamplifier are connected by a threaded connection and housed within the windscreen. The microphone 11 is connected to the input channel of a multi-channel data acquisition card via a cable. Furthermore, the microphone 11 converts the sound pressure signal into a voltage signal, which is then converted into a binary signal that can be read by the computer 3 via the multi-channel data acquisition card 2 and stored in the computer 3.

[0025] Furthermore, the computer 3 includes an analysis module and an evaluation module. The analysis module is used for: On the one hand, multi-channel preprocessing and synchronization correction are performed on the actual sound signal of the music road surface, and the sound pressure level is obtained by performing Hilbert-Huang Transform (HHT) on the synchronized actual sound signal. Alternatively, an energy-time-frequency relationship matrix can be generated, and a time-frequency matrix X(t,f) can be constructed. Principal component analysis is performed on the time-frequency matrix X(t,f), and the top K principal components are selected for reconstruction based on the cumulative contribution rate threshold. The music signal and noise signal are then distinguished by the energy proportion of the target main frequency band or the consistency of the spectral peaks. The signal-to-noise ratio is calculated based on the music signal and noise signal. or sound pressure level The noise-related quantities are also considered. On the other hand, the music signal is transformed using the Hilbert-Huang transform to generate a music signal relationship graph X2, and the ideal sound signal corresponding to the target musical score is transformed using the Hilbert-Huang transform to generate an ideal sound signal relationship graph X1. Both the ideal sound signal relationship graph X1 and the music signal relationship graph X2 represent sound pressure levels. Or an energy-time-frequency relationship diagram.

[0026] Specifically, the analysis module converts the time-domain acoustic signal acquired by microphone set 1 into a signal with both time and frequency domains using the Hilbert-Huang transform. The actual acoustic signal exhibits strong periodicity and structure, and also contains noise signals with a wide frequency spectrum. The analysis module separates the actual acoustic signal from the music track into noise and music signals using principal component analysis. Specifically, the principal component analysis process involves: orthogonally decomposing the time-frequency matrix X(t,f) obtained from the Hilbert-Huang transform to obtain a principal component sequence; selecting the top K principal components for reconstruction based on a cumulative contribution rate not lower than a preset threshold (preferably 80%); and determining whether the reconstructed signal is a music signal or a noise signal based on the energy proportion of the corresponding main frequency band or the consistency of the spectral peaks in the target musical score; and calculating the signal-to-noise ratio based on the music and noise signals. It is used to characterize the clarity of musical components in a musical road surface sound environment.

[0027] The time-frequency matrix was constructed using Hilbert-Huang transform. After orthogonal decomposition, the two principal components with the largest variance were retained to reconstruct the music signal and noise signal respectively. The signal-to-noise ratio of the music road surface was then calculated based on the music signal and noise signal. .

[0028] Furthermore, in the analysis module, the music signal is transformed using the Hilbert-Huang transform to generate a music signal relationship graph X2; the analysis module also transforms the target musical score of the music signal using the Hilbert-Huang transform to generate an ideal sound signal relationship graph X1; both the music signal relationship graph X2 and the ideal sound signal relationship graph X1 are based on sound pressure levels. Alternatively, the time-frequency relationship diagram of the energy. Furthermore, the evaluation module compares the ideal sound signal relationship diagram X1 input from the analysis module with the music signal relationship diagram X2. Cross-correlation peak localization is performed on the ideal sound signal relationship diagram X1 and the music signal relationship diagram X2 for time alignment and frequency band matching, achieving time axis alignment under different vehicle speeds or beat deviations; frequency band matching resamples the time-frequency energy matrix according to a preset frequency resolution and truncates the same frequency band range.

[0029] Furthermore, the evaluation module is used to: on the one hand, perform time alignment and frequency band matching on the ideal sound signal relationship diagram X1 and the music signal relationship diagram X2, wherein the time alignment is obtained by cross-correlation peak localization to obtain the time offset. The time axis of the music signal relationship graph X2 is shifted and compensated. Frequency band matching includes resampling X1 and X2 to the same frequency resolution and truncating the same frequency band. After alignment and matching, frequency accuracy and loudness comfort indices are output based on normalized mean square error (NMSE) or structural similarity (SSIM) as similarity / error measures. On the other hand, frequency accuracy labels are generated by integrating the outputs of the noise evaluation unit, objective music evaluation unit, and subjective timbre evaluation unit. Loudness and comfort labels Average score of subjective rating Noise evaluation indicators The four-dimensional evaluation results.

[0030] The evaluation module calculates a comprehensive evaluation index based on subjective and objective evaluations. Comprehensive evaluation indicators The calculation method is as follows: (1) in, The sound pressure level of the noise signal. To objectively evaluate the frequency accuracy labels obtained from the music signal relationship diagram X2. The loudness comfort label was obtained from the music signal relationship diagram X2 for objective evaluation. This represents the average score of the subjective ratings in the subjective evaluation. Noise evaluation index; comprehensive evaluation index Used to reflect the sound field effect of a musical road surface.

[0031] Furthermore, the evaluation module performs both subjective and objective evaluations. The subjective evaluation includes a noise assessment to reflect the impact of noise on music recognition and auditory experience. The noise assessment process involves the noise assessment unit measuring the sound pressure level of the noise signal. Compared to 70dB, which is the ambient noise quality standard (GB3096-2008), this embodiment uses the daytime limit of 70dB as the standard for noise signal evaluation on both sides of a main traffic artery, and bases the evaluation on the signal-to-noise ratio of the noise level relative to the ambient noise level of the road surface. Relationship judgment and output noise evaluation index This allows us to determine whether the noise signal meets the environmental sound quality standards.

[0032] Specifically, in this embodiment, the noise evaluation unit is used to calculate the A-weighted equivalent sound pressure level L of the noise signal. AThe signal-to-noise ratio (SNR) of noise signal A is calculated and compared with a preset threshold. The threshold is a configurable parameter, preferably one that matches the ambient sound quality standard (e.g., 70 dB during the day). To improve evaluation resolution, the noise evaluation index is preferably mapped to a continuous variable noise evaluation index I within the range of 0–1 using a piecewise function or a sigmoid function. n Noise evaluation index I n The closer to 1, the smaller the noise impact; the closer to 0, the larger the noise impact. It can also output a binary judgment result of "pass / fail" as additional information.

[0033] Furthermore, subjective evaluation also includes timbre evaluation: The timbre evaluation process is as follows: Based on the subjective ratings of drivers and passengers regarding harmony, melody, humanism, and rhythm obtained from the survey, the subjective timbre evaluation unit calculates the average score of the subjective ratings. It should be noted that the subjective rating range for each indicator is 0-1. The definitions of the aforementioned indicators are as follows: 1. Harmony: This refers to whether the placement of music on the musical street is appropriate and whether it has any impact on the surrounding environment and residents.

[0034] 2. Melodic quality, that is, whether the melody of the musical passage is accurate and smooth.

[0035] 3. Humanistic aspect: Musical pavements are often placed on scenic roads. Does the humanistic value of musical pavements combine with the characteristics of the scenic area and local features?

[0036] 4. Rhythm, that is, whether the rhythm of the music on the musical road will affect the driver's mood and interfere with the driver's driving behavior.

[0037] Furthermore, the objective music evaluation unit employs a convolutional neural network. The convolutional neural network structure includes: one input layer, three convolutional blocks, one global average pooling layer, one fully connected layer, and two dual-output layers. The input layer is used to input the music signal relationship graph X2; the convolutional blocks contain convolutional layers, normalization layers, and pooling layers, used to extract features from the music signal relationship graph X2; the global average pooling layer is used to eliminate the influence of the temporal features of the music signal relationship graph X2; the fully connected layer is used to fuse the extracted features of the music signal relationship graph X2; and the dual-output layers are used to evaluate the frequency accuracy labels. Loudness and comfort labels Output the signal. The convolutional neural network inputs a music signal (X2), and outputs frequency accuracy labels. and loudness comfort label .

[0038] Furthermore, the convolutional block includes a first convolutional block, a second convolutional block, and a third convolutional block. The first convolutional block is used to extract low-level texture features of the music signal relationship graph X2, such as the fundamental frequency components of the signal and basic physical properties such as short-term energy fluctuations. The second convolutional block is used to obtain mid-level features of the music signal relationship graph X2, which are combinations of low-level features, such as the periodic patterns of noise and the energy distribution of specific frequency bands. The third convolutional block is used to identify high-level composite features of the music signal relationship graph X2, such as noise type, noise intensity level, and noise category.

[0039] Furthermore, the convolutional layers include a first convolutional layer, a second convolutional layer, and a third convolutional layer. The number of kernels K in each convolutional layer is set to 32, 64, and 128, respectively. The kernel size of the first convolutional layer is 5×5, and the stride is set to 2. The kernel size of the second and third convolutional layers is 3×3, and the stride is set to 1. The padding of each convolutional layer is set to 2, and the ReLU activation function is used to enable the convolutional neural network to learn complex features. Preferably, a normalization layer is added after each convolutional layer to accelerate training and reduce internal covariate bias. Dropout layers are set in the second and third convolutional layers, and the dropout ratio is set to 0.3. The dropout layer randomly discards a certain proportion of neurons, thereby improving the robustness of the convolutional neural network model. More preferably, the fully connected layer and the convolutional block are connected through a feature flattening layer. The feature flattening layer converts the three-dimensional feature tensor into a one-dimensional feature vector for use by the fully connected layer. The fully connected layer is a two-branch fully connected layer, with each branch sharing the time-frequency features extracted by the convolutional layer. The fully connected layer uses the ReLU activation function to obtain the final output, and the dropout ratio of the fully connected layer is set to 0.5. This design effectively suppresses overfitting and significantly enhances generalization stability and robustness.

[0040] Furthermore, the convolutional neural network evaluates the actual sound signal of the musical road surface from two aspects: frequency and loudness, and outputs a loudness comfort label. Loudness comfort label and music signal loudness .

[0041] The second objective of this invention is to provide a method for evaluating the sound field effect of a road surface based on the aforementioned sound field effect evaluation device. The steps of the evaluation method are as follows: S1.1 Constructing a Pure Music Signal: Based on the relationship between the time signature and tempo v of the target musical score, the duration of each note is calculated, along with the frequencies and durations of other notes, to reproduce the notes in the target score. For example, this embodiment provides a simplified musical score with a time signature of 4 / 2, a key of C, a base frequency of 262Hz, and a tempo v of 110 beats / min. According to the twelve-tone equal temperament, the ratio of adjacent halftone frequencies is... From this, the frequency of each note can be calculated. Irregular harmonic signals are added to the original note signal to change the timbre. Harmonic signals from instruments such as the piano, violin, and pipa are added to simulate the playing of different instruments, thereby obtaining... A set of pure musical sound signals. Specifically, this embodiment acquires 100 pure musical sound signals and trains a model on them to ensure that, while keeping the notes constant, notes with different harmonic signals can obtain the same evaluation label. It should be noted that this invention only identifies and evaluates the accuracy of notes in the musical spectrum. In terms of music, high and low frequencies only exhibit an overtone relationship at the original note frequencies. Therefore, this invention does not consider the high and low frequency characteristics in the simplified musical notation during the pure music generation process.

[0042] S1.2 Building the Convolutional Neural Network Model: Please refer to further details. Figure 3 The convolutional neural network model is implemented using the PyTorch framework and trained within this framework. Based on the 100 sets of pure music sound signals obtained in step S1.1, the frequency features are progressively modified to generate a sequence ( , ,…, Based on a pure music signal, the loudness characteristics are gradually modified to generate a sequence. , ,…, To ensure the comfort of the human experience, the loudness characteristic range is from -10dB to +10dB. In this embodiment, a total of 300 sets of pure music sound signal training data were obtained.

[0043] S1.3. The 300 sets of pure music sound signal training data were subjected to Hilbert-Huang transform to obtain 300 sets of ideal sound signal relationship graphs X1. The statistical size was standardized to 128×256 pixels. The ideal sound signal relationship graphs X1 were then standardized and then input into the input layer of the convolutional neural network for model training to obtain sample relationship graphs of frequency, time and loudness. S1.4. The sample relationship graph is input from the CNN input layer into the convolutional block for model training. The training learning rate is 0.001, which decays with the training progress, halving every 20 rounds. The maximum number of training rounds is 50, and the number of training samples per iteration is 64, to balance convergence speed, training stability, and generalization ability. In this embodiment, the convolutional neural network uses two independent loss functions to calculate the loss of sample frequency and sample loudness respectively, and then sums them with weights to obtain the total sample loss. The mean squared error is used to calculate the loss of sample frequency and sample loudness. The weighted calculation formula for the total loss is: (2) in It is the sample frequency label of the i-th sample. It is the predicted frequency label of the i-th sample; It is the sample loudness label of the i-th sample. It is the predicted loudness label of the i-th sample. Preferably, the Adam optimizer is used to update the network weights during training to further improve the convergence efficiency and stability of model training.

[0044] S1.5. The output layer outputs a two-dimensional vector containing sample frequency labels and sample loudness labels; the output layer uses the Sigmoid activation function to ensure that the output is within the range of 0-1. Specifically, the output label of the output layer of the convolutional neural network model is a 2×1 vector containing sample frequency labels and sample loudness labels. Further, the predicted output label of the pure music sound signal with timbre changes obtained through step S1.1 is (1, 1), indicating that both the sample frequency label and the sample loudness label match the basic pure music signal. The predicted frequency label of the sample with frequency changes is (x, 1), indicating that the frequency matching degree with the pure music sound signal is constantly changing. The predicted loudness label of the sample with loudness changes is (1, y), indicating that the loudness matching degree with the pure music sound signal is constantly changing. Wherein (x, y) ranges from [0, 1].

[0045] S2. Collecting Actual Sound Signals: The experimental vehicle used in this embodiment is a new energy vehicle with a length of 458cm, a width of 179cm, a height of 152cm, and a total weight of approximately 1.8 tons. The experimental vehicle travels at a preset speed through a section of road paved with musical surfaces, generating actual sound signals from the musical surfaces. The time-domain sound pressure signals from the front and rear wheel surfaces and the four headrests near the ears of the test vehicle are simultaneously collected by microphone array 1, and the data is stored in the experimental computer 3 via a multi-channel data acquisition card. Preferably, the sound signal sampling frequency... f s The frequency was 10kHz, and vehicle speed and environmental information were recorded as evaluation references.

[0046] S3.1 Conduct a subjective evaluation. The steps for the subjective evaluation are as follows: S3.1.1 The actual sound signal is processed by the analysis module to perform principal component analysis, resulting in music and noise signals. Principal component analysis projects high-dimensional time-frequency features onto a low-dimensional space using orthogonal transformation. Typically, the direction with the largest variance contains the music signal with the highest energy proportion, while the direction with the smaller variance retains the noise signal and other information features. In this example, the music and noise signals are reconstructed, retaining only two principal components: the music signal with the largest energy proportion and the noise signal with the lowest energy proportion. The signal-to-noise ratio of the music road surface is calculated based on the music and noise signals. The calculation method is as follows: (3) in, The signal-to-noise ratio of the music track. The average power of the music signal; This represents the average power of the noise signal.

[0047] S3.1.2 In the noise evaluation unit, based on the signal-to-noise ratio of the music path... Obtain noise evaluation indicators The calculation method is as follows: (4) when When =0, the noise signal does not meet the sound environment quality standards. = When the noise signal meets the sound environment quality standards, more attention is paid to the clarity of the music signal, and the higher the signal-to-noise ratio, the higher the evaluation index.

[0048] S3.1.3 From the perspective of timbre, the driver collects subjective scores based on the listener's subjective feelings. The evaluation indicators for subjective scores include harmony, melody, humanism, and rhythm. The scoring range for each evaluation indicator is 0-1. The subjective timbre evaluation unit calculates the average score based on the subjective scores. Please refer to further information. Figure 6 This embodiment refers to the timbre evaluation table shown in the figure, which is based on music evaluation standards.

[0049] S3.2 The objective evaluation steps are as follows: In the objective music evaluation unit, the music signal relationship graph X2 is input into the convolutional neural network for evaluation, and the convolutional neural network outputs the accurate frequency label. Loudness and comfort labels Two-dimensional evaluation labels: The output layer uses the Sigmoid activation function to ensure that each item in the output is in the range of 0-1; The formula for calculating the frequency accuracy label is: (5) in, For accurate frequency labeling, The error rate of musical notes on the road surface; The loudness comfort label is calculated as follows: When the loudness of a music signal is greater than or equal to 70dB, the loudness comfort label of the music signal is recorded as 0; When the loudness of the music signal is less than 70dB, the calculation formula is used: (6) in, For loudness comfort label, For music signal loudness labels, For sample loudness; If the output satisfies 0.8 < <1, 0.4< If the output is less than 1, it means the construction of the musical pavement fully meets the design standards; if the output satisfies 0 < <0.8, 0.4< If the output is less than 1, it indicates that there are many incorrect notes in the musical pavement construction; if the output satisfies 0.8 < <1, 0< A value of <0.4 indicates that while the musical pavement meets design standards, the human experience is poor.

[0050] S3.2 Calculate the comprehensive evaluation index that combines subjective and objective evaluations for musical road surfaces. Comprehensive evaluation indicators Used to reflect the sound field effect of a musical road surface.

[0051] Comprehensive evaluation indicators According to noise sound pressure level Whether it exceeds 70dB, different weights are assigned to different evaluation items. Sound pressure level of the noise signal. At a sound pressure level >70dB, the noise signal has the greatest impact on the sound field environment. The human experience of the music-driven environment is not the dominant factor. At this point, the overall evaluation focuses more on the objective frequency accuracy and loudness comfort of the music-driven environment; the sound pressure level of the noise signal... When the noise level is ≤70dB, the impact of the noise signal is relatively small, and noise evaluation indicators should be increased. The weighting reflects the attention to detail in low-noise scenarios.

[0052] This dynamic adjustment of evaluation item weights makes the evaluation results more closely reflect the actual listening experience under different noise environments. Frequency accuracy is the core of the music road surface evaluation, and it has the highest weight; noise evaluation indicators... = Because it is directly related to human experience, its evaluation metrics have a secondary weight. Even if the notes on the musical path are accurate, the signal-to-noise ratio... Too low, comprehensive evaluation indicators This will also significantly reduce [the noise level], and this design is more in line with the actual auditory experience. This invention can more scientifically reflect the clarity of music signals, enabling a more comprehensive and objective evaluation of the design and construction effects of music-themed road surfaces.

[0053] It is worth noting that those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0054] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0055] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0056] These computer program instructions can also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0057] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

[0058] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus.

[0059] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A device for evaluating the sound field effect of a musical road surface, characterized in that, Includes a data acquisition device and a computer (3) electrically connected to the data acquisition device; The acquisition device includes a microphone set (1) and a multi-channel data acquisition card (2). One end of the multi-channel data acquisition card (2) is electrically connected to the microphone set (1), and the other end of the multi-channel data acquisition card (2) is electrically connected to the computer (3). The multi-channel data acquisition card (2) is used to transmit the actual sound signal of the music road surface acquired by the microphone set (1) to the computer (3). The computer (3) includes an analysis module and an evaluation module, the analysis module being used for: On the one hand, the actual sound signal of the music road surface is preprocessed and synchronously corrected through multiple channels, and principal component analysis is performed to generate a music signal relationship diagram and an ideal sound signal relationship diagram: the analysis module separates the actual sound signal of the music road surface into noise signal and music signal through principal component analysis; On the other hand, in the analysis module, the music signal is transformed into a music signal relationship diagram X2 through the Hilbert-Huang transform; the analysis module transforms the target musical score of the music signal into an ideal sound signal relationship diagram X1 through the Hilbert-Huang transform; both the music signal relationship diagram X2 and the ideal sound signal relationship diagram X1 are sound pressure levels. Or an energy-time-frequency relationship diagram; The evaluation module is used for subjective and objective evaluations: On the one hand, time alignment and frequency band matching are performed on the ideal sound signal relationship diagram X1 and the music signal relationship diagram X2, and frequency accuracy index and loudness comfort index are output based on preset similarity / error metric. On the other hand, the outputs of the integrated noise evaluation unit, objective music evaluation unit, and subjective timbre evaluation unit are used to generate frequency accuracy labels. Loudness and comfort labels Average score of subjective rating Noise evaluation indicators The four-dimensional evaluation results; The evaluation module calculates a comprehensive evaluation index based on the subjective evaluation and the objective evaluation. The comprehensive evaluation index The calculation method is as follows: (1) in, The sound pressure level of the noise signal. The frequency accuracy label is obtained from the music signal relationship diagram X2 in the objective evaluation. The loudness comfort label is obtained from the music signal relationship diagram X2 in the objective evaluation. The average score of the subjective ratings in the subjective evaluation. The noise evaluation index; the comprehensive evaluation index Used to reflect the sound field effect of the musical road surface; The average score is calculated by the subjective timbre evaluation unit based on the subjective scores of drivers and passengers on harmony, melody, humanism, and rhythm obtained from the survey.

2. The music road surface sound field effect evaluation device according to claim 1, characterized in that, The microphone set (1) includes multiple microphones (11). One microphone (11) is fixed at the center of the outer side of one front wheel and one rear wheel of the experimental vehicle. The distance between the microphone (11) and the surface of the front wheel and the surface of the rear wheel is 10cm. Four microphones (11) are fixed at the position of the headrests of the four seats of the experimental vehicle near the ears. The seats include the driver's seat, the passenger seat, the left rear seat and the right rear seat.

3. The music road surface sound field effect evaluation device according to claim 1, characterized in that, The process of principal component analysis performed by the analysis module is as follows: a time-frequency matrix is ​​constructed through Hilbert-Huang transform, the time-frequency matrix is ​​orthogonally decomposed, and the top K principal components are selected for reconstruction based on the cumulative contribution rate threshold; The music signal and noise signal are distinguished by combining the energy proportion of the target main frequency band or the consistency of spectral peaks, and the signal-to-noise ratio of the music road surface is calculated based on the music signal and the noise signal. .

4. The music road surface sound field effect evaluation device according to claim 1, characterized in that, The subjective evaluation includes noise evaluation. The noise evaluation process is as follows: the noise evaluation unit is used to calculate the sound pressure level of the noise signal. The noise evaluation index is compared with a preset threshold and then output as a noise evaluation index. The preset threshold is a configurable parameter, which is 70dB or a threshold range that matches the application scenario.

5. The music road surface sound field effect evaluation device according to claim 1, characterized in that, The objective music evaluation unit employs a convolutional neural network. The convolutional neural network takes the music signal relationship graph X2 as input and outputs a frequency accuracy label. and loudness comfort label .

6. The music road surface sound field effect evaluation device according to claim 5, characterized in that, The convolutional neural network structure includes: one input layer, three convolutional blocks, one global average pooling layer, one fully connected layer, and two dual-output layers. The input layer is used to input the music signal relationship graph X2. The convolutional blocks include convolutional layers, normalization layers, and pooling layers, used to extract features from the music signal relationship graph X2. The global average pooling layer is used to eliminate the influence of the temporal features of the music signal relationship graph X2. The fully connected layers are used to fuse the extracted features of the music signal relationship graph X2. The dual-output layers are used to process the frequency accuracy labels. and the loudness comfort label Output the results.

7. The music road surface sound field effect evaluation device according to claim 6, characterized in that, The convolutional block includes a first convolutional block, a second convolutional block, and a third convolutional block. The first convolutional block is used to extract low-level texture features of the music signal relationship graph X2, the second convolutional block is used to obtain mid-level features of the music signal relationship graph X2, and the third convolutional block is used to identify high-level composite features of the music signal relationship graph X2. The convolutional layer includes a first convolutional layer, a second convolutional layer, and a third convolutional layer. The padding of each convolutional layer is set to 2, and the ReLU activation function is used. A normalization layer is added after each convolutional layer. The second convolutional layer and the third convolutional layer are equipped with dropout layers, and the dropout ratio is set to 0.

3.

8. The music road surface sound field effect evaluation device according to claim 1, characterized in that, The subjective evaluation also includes timbre evaluation: the timbre evaluation process is as follows: the subjective timbre evaluation unit calculates the average score of the subjective scores based on the subjective ratings of drivers and passengers on harmony, melody, humanism, and rhythm obtained from the survey. The timbre evaluation includes at least one or more of the evaluation indicators. .

9. A method for evaluating the sound field effect of a musical road surface based on the musical road surface sound field effect evaluation device according to any one of claims 1-8, characterized in that, The steps of the evaluation method are as follows: S1.1 Constructing a Pure Music Signal: Based on the relationship between the time signature and tempo v of the target score, calculate the duration, frequency, and duration of each note to reproduce the notes in the score. Add irregular harmonic signals to the original note signal to change the timbre, and add harmonic signals from different instruments to simulate the playing of different instruments, thereby obtaining... A set of pure music sound signals; S1.2 Building the Convolutional Neural Network Model: The convolutional neural network model is implemented using PyTorch and trained in PyTorch, based on the results obtained in step S1.

1. The pure music sound signal described in the group is used to gradually modify the frequency characteristics of the pure music sound signal to generate a sequence ( , ,…, Based on the pure music signal, the loudness characteristics are gradually modified to generate a sequence ( , ,…, The loudness characteristic varies from -10dB to +10dB, thus obtaining 3 A set of pure music sound signal training data; S1.3, 3 The pure music sound signal training data of the group is subjected to Hilbert-Huang transform to obtain 3a groups of ideal sound signal relationship diagrams X1. The ideal sound signal relationship diagram X1 is standardized and then input into the input layer of the convolutional neural network for model training to obtain the sample relationship diagram. S1.

4. The sample relationship graph is input from the input layer into the convolutional block for model training. The convolutional neural network uses two independent loss functions to calculate the loss for sample frequency and sample loudness respectively, and then sums them with weights to obtain the total sample loss. The loss for sample frequency and sample loudness is calculated using mean squared error. The weighted calculation formula for the total loss is: (2) in It is the sample frequency label of the i-th sample. It is the predicted frequency label of the i-th sample; It is the sample loudness label of the i-th sample. It is the predicted loudness label of the i-th sample; S1.5, The output layer outputs a two-dimensional vector containing sample frequency labels and sample loudness labels; S2. Collect actual sound signals: The experimental vehicle passes over the road surface paved with music, generating actual sound signals of the music road surface. The actual sound signals of the music road surface inside and outside the vehicle are collected by the microphone and stored in the experimental computer through the data acquisition card. S3.1 Conduct a subjective evaluation. The steps for the subjective evaluation are as follows: The actual sound signal described in S3.1.1 undergoes principal component analysis by the analysis module to obtain a music signal and a noise signal. The signal-to-noise ratio of the music road surface is then calculated based on the music signal and the noise signal. The calculation method is as follows: (3) in, The signal-to-noise ratio of the music track. The average power of the music signal; The average power of the noise signal; S3.1.2 In the noise evaluation unit, based on the signal-to-noise ratio of the music path... Obtain noise evaluation indicators The calculation method is as follows: (4) when When =0, the noise signal does not meet the sound environment quality standard. = At that time, the noise signal conforms to the sound environment quality standard; S3.1.3 From the perspective of timbre, the driver collects subjective scores based on the listener's subjective feelings. The evaluation indicators for the subjective scores include harmony, melody, humanism, and rhythm. The scoring range for each evaluation indicator is 0-1. The subjective timbre evaluation unit calculates the average score of the subjective scores based on the subjective scores. ; S3.2 The objective evaluation steps are as follows: In the objective music evaluation unit, the music signal relationship graph X2 is input into the convolutional neural network for evaluation, and the convolutional neural network outputs frequency accuracy labels. Loudness and comfort labels Two-dimensional evaluation labels; S3.3 Calculate the comprehensive evaluation index The comprehensive evaluation index The comprehensive evaluation index is calculated based on a combination of the subjective evaluation and the objective evaluation. Used to reflect the sound field effect of a musical road surface.

10. The method for evaluating the sound field effect of a musical road surface according to claim 9, characterized in that, The process of evaluating the music signal relationship graph X2 in the convolutional neural network includes the following steps: The output layer uses the Sigmoid activation function to ensure that each item in the output is in the range of 0-1; The formula for calculating the frequency accuracy label is: (5) in, For the accurate label of the frequency. The error rate of musical notes on the road surface; The loudness comfort label is calculated as follows: When the loudness of the music signal is greater than or equal to 70dB, the loudness comfort label of the music signal is recorded as 0; When the loudness of the music signal is less than 70dB, the calculation formula is used: (6) in, For loudness comfort label, For music signal loudness labels, The loudness of the sample; If the output satisfies: 0.8 < <1, 0.4< If the output is less than 1, it means the construction of the music-themed pavement fully meets the design standards; if the output satisfies: 0 < <0.8, 0.4< If the output is less than 1, it indicates that there are many incorrect notes in the musical pavement construction; if the output satisfies: 0.8 < <1, 0< If the value is less than 0.4, it means that although the musical pavement meets the design standards, the human experience is poor.

Images