Sound quality optimization method and device based on transmission path analysis

Abstract

The invention discloses a sound quality optimization method and device based on transmission path analysis, and the method comprises the steps: collecting the noise data of a detected object, calculating the features of the noise data, and evaluating the noise comfort value of the noise data as a noise data set; constructing a prediction model based on an XGBoost network structure, and optimizing the prediction model by using the noise data set to obtain an XGBoost sound quality prediction model; the noise contribution of each path to a target point is calculated through a transmission path analysis method to serve as path noise, and the noise comfort value of the path noise is predicted through an XGBoost sound quality prediction model to serve as sound quality contribution; and carrying out Shap value analysis on the first M paths with the highest sound quality contribution to obtain optimization characteristics and optimization frequency so as to realize sound quality optimization. According to the method, the XGBoost sound quality prediction model is introduced to establish the sound quality transmission path analysis method, and the main path influencing the sound quality and the characteristics and frequency of the main path can be quickly found out in the transmission path analysis so as to guide the sound quality optimization work.

Description

Technical Field

[0001] The present invention relates to the technical field of noise analysis, and in particular to a sound quality optimization method and device based on transfer path analysis. Background Art

[0002] As people become more aware of noise characteristics, the auditory system will form a specific perceptual connection between the physical characteristics of noise and the functional quality of the sound source, generating specific psychological perceptual needs. Traditional acoustic research, which mainly focuses on the optimization of objective parameters, is gradually unable to meet the "people-oriented" research trend, so more and more research work on sound quality has been generated.

[0003] In traditional vibration and noise reduction optimization work, transfer path analysis (TAP) is the most commonly used method. Transfer path analysis effectively controls noise by identifying and quantifying the paths of noise and vibration from the source to the receiving point, and optimizing the main paths with higher contributions. However, since transfer path analysis is aimed at the optimization of sound pressure level, it is difficult to directly reflect the optimization results of sound quality through transfer path analysis. Therefore, combining sound quality with transfer path analysis methods can effectively identify the main paths that affect the sound quality optimization results and provide guidance for subsequent sound quality optimization work. Simply calculating the sound quality contribution of each transfer path can indeed identify the main path to guide the optimization work, but when determining the main optimization frequency range, it is still necessary to repeatedly modify the transfer function and calculate the path contribution. This process is cumbersome and blind. Summary of the invention

[0004] The purpose of the present invention is to solve the problem that in the prior art sound quality optimization process, the transfer function needs to be repeatedly modified and the path contribution needs to be calculated.

[0005] The technical solution adopted by the present invention to solve the technical problem is: to provide a sound quality optimization method based on transfer path analysis, comprising the following steps:

[0006] Collecting noise data of the object under test, calculating the characteristics of the noise data and evaluating the noise comfort value of the noise data, and taking the characteristics of the noise data and the noise comfort value as a noise data set; the characteristics of the noise data include local frequency domain characteristics and global sound quality characteristics;

[0007] A prediction model is constructed based on the network structure of XGBoost, and the prediction model is optimized using the noise data set to obtain an XGBoost sound quality prediction model; the XGBoost sound quality prediction model takes the characteristics of the noise data as input and outputs a noise comfort value;

[0008] The noise contribution of each path to the target point is calculated as the path noise through the transfer path analysis method, and the noise comfort value of the path noise is predicted as the sound quality contribution using the XGBoost sound quality prediction model.

[0009] The Shap value analysis is performed on the top M paths with the highest contribution to the sound quality to obtain the optimized features and optimized frequencies, and the sound quality is optimized based on the optimized features and optimized frequencies.

[0010] Preferably, the collecting of noise data of the measured object under different working conditions is specifically: using a microphone to measure the target point noise, near-field noise and far-field noise of the measured object under different working conditions, cutting the data into 5-10s, and the sampling frequency is 44.1kHz.

[0011] Preferably, the calculating the characteristics of the noise data comprises the following steps:

[0012] Use the Grey Wolf Optimization Algorithm to optimize the parameters in VMD;

[0013] The noise signal is decomposed into multiple time-frequency localized modal components using VMD with optimized parameters, the first five eigenmodal components are taken and the remaining signals are combined as the sixth eigenmodal components;

[0014] Perform fast Fourier transform (FFT) on the extracted intrinsic mode components and extract local frequency domain features, including kurtosis, waveform factor and energy;

[0015] Calculate traditional sound quality parameters for each noise signal, including loudness, sharpness, roughness, volatility, pitch, and speech intelligibility.

[0016] Preferably, the use of the grey wolf optimization algorithm to optimize the parameters in VMD is specifically as follows: in each iteration of the grey wolf algorithm, the penalty factor α and the modal decomposition number K of VMD are updated according to the social level stratification and search process of GWO.

[0017] Preferably, the noise comfort value of the noise data is evaluated, comprising the following steps:

[0018] The acoustic comfort is divided into 1 to 10 levels, and the noise samples are scored according to the acoustic comfort level to obtain the noise comfort value;

[0019] The mean and standard deviation of the noise comfort values ​​were calculated, and data points falling outside the mean plus or minus three times the standard deviation were eliminated.

[0020] Preferably, the optimization of the prediction model using the noise data set comprises the following steps:

[0021] Perform correlation analysis on the noise data set, select the main features to form the data set d2, and normalize the data in the data set d2 to obtain the data set D;

[0022] Based on the dataset D, the prediction model is trained by gradient boosting method and regularization method.

[0023] Preferably, the step of calculating the noise contribution of each path to the target point as the path noise by using a transfer path analysis method comprises the following steps:

[0024] Define the transfer path analysis model of the object under test;

[0025] Collect vibration signals and noise signals from noise sources, and obtain noise data based on the transfer path analysis model;

[0026] Measure the transfer function of the transfer path and use the transfer function and the spectrum of the noise data to calculate the contribution of each path to the noise or vibration of the target point;

[0027] Perform inverse Fourier transform on the contribution of each path to the noise or vibration at the receiving point to obtain the time domain contribution noise of each path as the path noise.

[0028] Preferably, the XGBoost sound quality prediction model predicts the noise comfort value of the path noise as the sound quality contribution, comprising the following steps:

[0029] Calculate the characteristics of each path noise, where the characteristics are the main characteristics screened out by correlation analysis;

[0030] The XGBoost sound quality prediction model predicts the noise comfort value as the sound quality contribution of the path based on the characteristics of the path noise.

[0031] Preferably, the Shap value analysis for the first M paths with the highest contribution to the sound quality comprises the following steps:

[0032] Select the first M paths with the highest contribution to sound quality as the main paths;

[0033] For each path in the main path, the contribution of each feature in the path to the prediction result is calculated, the top N features with the highest contribution are selected as the optimized features, and the central frequency corresponding to the IMF of the optimized features is used as the optimized frequency.

[0034] The present invention also provides a sound quality optimization device based on transfer path analysis, comprising:

[0035] A data set acquisition module collects noise data of the object under test under different working conditions, calculates the characteristics of the noise data and evaluates the noise comfort value of the noise data, and uses the characteristics of the noise data and the noise comfort value as the noise data set; the characteristics include local frequency domain characteristics and global sound quality characteristics;

[0036] A sound quality prediction model acquisition module builds a prediction model based on the XGBoost network structure, optimizes the prediction model using a noise data set, and obtains an XGBoost sound quality prediction model; the XGBoost sound quality prediction model takes the characteristics of the noise data as input and outputs a noise comfort value;

[0037] The sound quality contribution acquisition module calculates the noise contribution of each path as the path noise through the transfer path analysis method, and uses the XGBoost sound quality prediction model to predict the noise comfort value of the path noise as the sound quality contribution;

[0038] The sound quality optimization module performs Shap value analysis on the top M paths with the highest contribution to sound quality, obtains optimization features and optimization frequencies, and realizes sound quality optimization based on the optimization features and optimization frequencies.

[0039] The present invention has the following beneficial effects:

[0040] (1) The present invention uses the GWO-VMD algorithm to extract the frequency domain characteristics of noise, which can accurately locate the optimized main frequency range;

[0041] (2) The present invention combines the XGBoost sound quality prediction model with the transfer path analysis method, which can quickly find the main paths that affect the sound quality in the transfer path analysis, and can point out the key frequencies for optimization based on importance analysis to guide subsequent sound quality optimization work.

[0042] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments, but the present invention is not limited to the embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 A method step diagram of an embodiment of the present invention;

[0044] Figure 2 is a detailed flow chart of an embodiment of the present invention;

[0045] Figure 3 FIG. 4 is a diagram showing the structure of a device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0046] See also Figure 1 and Figure 2 The method step diagram and detailed flow chart of the embodiment of the present invention are shown, which include the following steps:

[0047] S101, collecting noise data of the object under test under different working conditions, calculating the characteristics of the noise data and evaluating the noise comfort value of the noise data, and taking the characteristics of the noise data and the noise comfort value as a noise data set; the characteristics include local frequency domain characteristics and global sound quality characteristics;

[0048] S102, constructing a prediction model based on the network structure of XGBoost, optimizing the prediction model using the noise data set, and obtaining an XGBoost sound quality prediction model; the XGBoost sound quality prediction model takes the characteristics of the noise data as input and outputs a noise comfort value;

[0049] S103, calculating the noise contribution of each path as path noise by using a transfer path analysis method, and predicting the noise comfort value of the path noise as the sound quality contribution by using an XGBoost sound quality prediction model;

[0050] S104, performing Shap value analysis on the top M paths with the highest contribution to the sound quality, obtaining optimized features and optimized frequencies, and achieving sound quality optimization based on the optimized features and optimized frequencies.

[0051] Specifically, S101 uses a microphone to measure the target point noise, near field noise and far field noise of the measured object under different working conditions, cuts the data into 5-10s, with a sampling frequency of 44.1kHz, and saves it in a suitable format for subsequent calculation and sound playback.

[0052] Specifically, the step S101 calculates the local frequency domain features and the global sound quality features of the collected noise, including the following:

[0053] (2.1) Use the GWO algorithm to optimize the parameters in VMD. In each iteration, according to the social hierarchy stratification and search process of GWO, the penalty factor α and the modal decomposition number K of VMD are updated to obtain a better VMD decomposition effect;

[0054] (2.2) Using the optimized parameters to perform VMD, the signal is decomposed into multiple time-frequency localized modal components, the first five modal components are taken, and the remaining signals are combined as the sixth modal components;

[0055] (2.3) Perform fast Fourier transform (FFT) on the extracted IMF components to extract frequency domain features;

[0056] (2.4) Extract frequency domain features such as kurtosis, waveform factor, and energy of each eigenmode component;

[0057] (2.5) Calculate the traditional sound quality parameters (i.e., psychoacoustic parameters) of each original signal, including loudness, sharpness, roughness, fluctuation, pitch, speech clarity, etc.

[0058] Specifically, the step S101 uses a grading method to evaluate the noise comfort values ​​of all collected noise signals as a subjective noise evaluation, including the following:

[0059] (3.1) Acoustic comfort is divided into 1 to 10 levels, and professionals are organized to complete the acoustic comfort evaluation of all samples by replaying the noise samples and scoring them according to the acoustic comfort level;

[0060] (3.2) Conduct statistical analysis on the evaluation results and use Spearman correlation analysis to test consistency to ensure the reliability of the evaluation results. Test whether there are significant differences between evaluators and whether the evaluation results are statistically significant.

[0061] Specifically, S102 constructs an XGBoost sound quality prediction model and uses the Shap model to analyze the overall feature importance, which specifically includes the following:

[0062] (4.1) The sound quality parameters obtained in step 2) and the corresponding acoustic comfort evaluation results obtained in step 3) are used to form a data set d1. The main characteristic parameters are screened out through parameter correlation analysis to form a data set d2. The data is normalized to obtain a data set D.

[0063] (4.2) The data D is divided into a training set and a test set according to a certain ratio. The extreme gradient boosting (XGBoost) regression algorithm is used to fit the data in the training set, and the model prediction ability is evaluated on the test set to obtain the sound quality prediction model.

[0064] Specifically, in S103, the noise contribution of each path is calculated as the path noise by using the transfer path analysis method, including the following:

[0065] (5.1) Define the transfer path analysis model of the object under test;

[0066] (5.2) Collect vibration and noise signals from the noise source and measure the transfer function of the transmission path;

[0067] (5.3) Using the transfer function and the spectrum of the noise source, calculate the contribution of each path to the noise or vibration at the receiving point;

[0068] (5.4) Perform an inverse Fourier transform on the contribution of each path to obtain the time domain contribution of each path (TPA path contribution).

[0069] Specifically, in S103, the XGBoost sound quality prediction model is used to predict the noise comfort value of each path as the sound quality contribution, and several paths with higher contribution are used as main paths. In another embodiment, the XGBoost sound quality prediction model predicts the sound quality contribution of noise based on the right ear synthetic noise.

[0070] Specifically, in S104, a local SHAP value analysis is performed on the main paths with higher sound quality contribution. For each path, the specific contribution of each feature of the path to the prediction result is calculated, including positive contribution (increasing the prediction value) and negative contribution (reducing the prediction value), to determine the main features that affect the overall sound quality, and determine the main frequency range for sound quality optimization based on the center frequency corresponding to the IMF.

[0071] See also Figure 3 FIG. 1 is a diagram showing a structure of a device according to an embodiment of the present invention, comprising:

[0072] The data set acquisition module 301 acquires noise data of the object under test under different working conditions, calculates the characteristics of the noise data and evaluates the noise comfort value of the noise data, and uses the characteristics of the noise data and the noise comfort value as the noise data set; the characteristics include local frequency domain characteristics and global sound quality characteristics;

[0073] The sound quality prediction model acquisition module 302 builds a prediction model based on the XGBoost network structure, optimizes the prediction model using the noise data set, and obtains an XGBoost sound quality prediction model; the XGBoost sound quality prediction model takes the characteristics of the noise data as input and outputs a noise comfort value;

[0074] The sound quality contribution acquisition module 303 calculates the noise contribution of each path as the path noise by using the transfer path analysis method, and uses the XGBoost sound quality prediction model to predict the noise comfort value of the path noise as the sound quality contribution;

[0075] The sound quality optimization module 304 performs Shap value analysis on the first M paths with the highest sound quality contribution to obtain optimization features and optimization frequencies, and realizes sound quality optimization based on the optimization features and optimization frequencies.

[0076] It can be seen that the present invention introduces the XGBoost sound quality prediction model to establish a sound quality transfer path analysis method that can reflect the main optimization frequency. The main path affecting the sound quality can be quickly found in the transfer path analysis, and the key frequency for optimization can be pointed out based on the importance analysis. It has important research value for guiding the subsequent sound quality optimization work.

[0077] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A sound quality optimization method based on transfer path analysis, characterized in that: The steps include: Collecting noise data of the object under test, calculating the characteristics of the noise data and evaluating the noise comfort value of the noise data, and taking the characteristics of the noise data and the noise comfort value as a noise data set; the characteristics of the noise data include local frequency domain characteristics and global sound quality characteristics; A prediction model is constructed based on the network structure of XGBoost, and the prediction model is optimized using the noise data set to obtain an XGBoost sound quality prediction model; the XGBoost sound quality prediction model takes the characteristics of the noise data as input and outputs a noise comfort value; The noise contribution of each path to the target point is calculated as the path noise through the transfer path analysis method, and the noise comfort value of the path noise is predicted as the sound quality contribution using the XGBoost sound quality prediction model. The Shap value analysis is performed on the top M paths with the highest contribution to the sound quality to obtain the optimized features and optimized frequencies, and the sound quality is optimized based on the optimized features and optimized frequencies.

2. The sound quality optimization method based on transfer path analysis according to claim 1, characterized in that: The noise data of the measured object under different working conditions are collected, specifically: using a microphone to measure the target point noise, near field noise and far field noise of the measured object under different working conditions, cutting the data into 5-10s, and the sampling frequency is 44.1kHz.

3. The sound quality optimization method based on transfer path analysis according to claim 1, characterized in that: The calculation of the characteristics of the noise data comprises the following steps: Use the Grey Wolf Optimization Algorithm to optimize the parameters in VMD; The noise signal is decomposed into multiple time-frequency localized modal components using VMD with optimized parameters, the first five eigenmodal components are taken and the remaining signals are combined as the sixth eigenmodal components; Perform fast Fourier transform (FFT) on the extracted intrinsic mode components and extract local frequency domain features, including kurtosis, waveform factor and energy; Calculate traditional sound quality parameters for each noise signal, including loudness, sharpness, roughness, volatility, pitch, and speech intelligibility.

4. The sound quality optimization method based on transfer path analysis according to claim 3, characterized in that: The gray wolf optimization algorithm is used to optimize the parameters in VMD, specifically: in each iteration of the gray wolf algorithm, the penalty factor α and the modal decomposition number K of VMD are updated according to the social level stratification and search process of GWO.

5. The sound quality optimization method based on transfer path analysis according to claim 1, characterized in that: The step of evaluating the noise comfort value of noise data comprises the following steps: The acoustic comfort is divided into 1 to 10 levels, and the noise samples are scored according to the acoustic comfort level to obtain the noise comfort value; The mean and standard deviation of the noise comfort values ​​were calculated, and data points falling outside the mean plus or minus three times the standard deviation were eliminated.

6. The sound quality optimization method based on transfer path analysis according to claim 1, characterized in that: The method of optimizing the prediction model by using the noise data set comprises the following steps: Perform correlation analysis on the noise data set, select the main features to form the data set d2, and normalize the data in the data set d2 to obtain the data set D; Based on the dataset D, the prediction model is trained by gradient boosting method and regularization method.

7. The sound quality optimization method based on transfer path analysis according to claim 1, characterized in that: The method of calculating the noise contribution of each path to the target point as path noise by using the transfer path analysis method comprises the following steps: Define the transfer path analysis model of the object under test; Collect vibration signals and noise signals from noise sources, and obtain noise data based on the transfer path analysis model; Measure the transfer function of the transfer path and use the transfer function and the spectrum of the noise data to calculate the contribution of each path to the noise or vibration of the target point; Perform inverse Fourier transform on the contribution of each path to the noise or vibration at the receiving point to obtain the time domain contribution noise of each path as the path noise.

8. The sound quality optimization method based on transfer path analysis according to claim 6, characterized in that: The XGBoost sound quality prediction model predicts the noise comfort value of the path noise as the sound quality contribution, including the following steps: Calculate the characteristics of each path noise, where the characteristics are the main characteristics screened out by correlation analysis; The XGBoost sound quality prediction model predicts the noise comfort value as the sound quality contribution of the path based on the characteristics of the path noise.

9. The sound quality optimization method based on transfer path analysis according to claim 1, characterized in that: The Shap value analysis for the first M paths with the highest contribution to the sound quality includes the following steps: Select the first M paths with the highest contribution to sound quality as the main paths; For each path in the main path, the contribution of each feature in the path to the prediction result is calculated, the top N features with the highest contribution are selected as the optimized features, and the central frequency corresponding to the IMF of the optimized features is used as the optimized frequency.

10. A sound quality optimization device based on transfer path analysis, characterized in that: include: A data set acquisition module collects noise data of the object under test under different working conditions, calculates the characteristics of the noise data and evaluates the noise comfort value of the noise data, and uses the characteristics of the noise data and the noise comfort value as the noise data set; the characteristics include local frequency domain characteristics and global sound quality characteristics; A sound quality prediction model acquisition module builds a prediction model based on the XGBoost network structure, optimizes the prediction model using a noise data set, and obtains an XGBoost sound quality prediction model; the XGBoost sound quality prediction model takes the characteristics of the noise data as input and outputs a noise comfort value; The sound quality contribution acquisition module calculates the noise contribution of each path as the path noise through the transfer path analysis method, and uses the XGBoost sound quality prediction model to predict the noise comfort value of the path noise as the sound quality contribution; The sound quality optimization module performs Shap value analysis on the top M paths with the highest contribution to sound quality, obtains optimization features and optimization frequencies, and realizes sound quality optimization based on the optimization features and optimization frequencies.

Images