Research On Sound Quality Evaluation Model Of Vehicle Interior Noise Based On BP Neural Network

With the improvement of people's living standards and the rapid development of the automotive industry,people pay more and more attention to the quality of automobile products,but the vehicle interior sound quality is one of the important factors of the automobile products quality.In recent years,scholars have done a lot of research on the sound quality of the interior noise,and have made great achievements.However,due to the uncertainty of the subjective evaluation results in the evaluation of the sound quality,the limitations of the objective psychoacoustic parameters,and the complex nonlinear relationship between them,the sound quality evaluation model currently established still has many problems.It is necessary to make constant attempts to find a prediction method that can accurately reflect people's subjective feelings of noise.Therefore,we also need to make continuous exploration to find a prediction method that can accurately and comprehensively reflect people's subjective perception of noise.In psychoacoustic parameters,only the loudness has a standard calculation model,the other parameters do not have a unified calculation standard,and even some parameters can not be calculated by the specific formula.Moreover,most of the loudness calculation model is based on Zwicker model.The model does not consider the masking effect of human ear in time domain,and there is a large error in calculating the unsteady state noise.Because most of the noise in the vehicle interior belongs to the nonsteady noise,the latest loudness calculation standard(ISO 532-2)will be used in the calculation of the loudness of this paper.A large number of research results show that loudness,sharpness and roughness have the greatest impact on the objective evaluation results of sound quality.Therefore,we only combine the three main psychoacoustic parameters and the subjective evaluation results to establish the sound quality evaluation model for the interior noise of the vehicle,and try to use the optimization algorithm to improve the prediction accuracy of the model.Details are as follows:(1)According to the GB/T18697-2002 internal noise measurement standard,the noise sampling experiment is carried out,and the best noise signal is selected to set up the database of noise sample.The calculation programs of loudness,sharpness and roughness is written by Matlab software,and the accuracy of the calculation results is verified.(2)The annoyance was taken as the evaluation index of the sound quality,and subjective evaluation experiment was carried out by the adaptive grouping and paired comparison method.Based on statistical analysis software SPSS,the subjective evaluation results of 25 evaluators were analyzed,and the evaluation results with correlation coefficients less than 0.7 were excluded.By calculating the pearson correlation coefficient between loudness,sharpness,roughness and subjective evaluation results,it is shown that the correlation between them is high,and the three psychoacoustic parameters have a greater impact on the evaluation of the sound quality.(3)The psychoacoustic parameters and the subjective evaluation results are normalized to be used as the input and output of the BP neural network sound quality evaluation model,and establish the noise quality evaluation model of the vehicle interior,and verify the validity of the model.In order to improve the prediction accuracy of the model,the improved fish swarm algorithm is applied to optimize the initial weights and thresholds of the BP neural network,and the trained network is applied to the simulation prediction.The results show that the prediction accuracy of the optimized BP neural network model has been significantly improved.