Research On Sound Quality Evaluation Of Automobile Closing Door Based On Wavelet Packet Transform

With the improvement of the basic performance of automobile such as power performance,people have higher requirements for other qualities of automobile such as automobile sound quality,so automobile sound quality has become a research hotspot.Among them,the research on sound quality of stationary acoustic signal started earlier,the research and evaluation process in this field are relatively perfect.However,there are many kinds of automobile acoustic signals,and most of them are non-stationary.Psychoacoustic parameters suitable for stationary acoustic signals are difficult to accurately describe nonstationary acoustic signals and reflect people’s subjective feelings towards sound.Therefore,it has become a research focus to construct objective evaluation parameters and evaluation model of non-stationary acoustic signals by using time-frequency analysis methods.The sound of automobile door closing is a non-stationary acoustic signal which is frequently contacted.It has theoretical significance and engineering value to study the sound quality of automobile door closing by time-frequency analysis method.This paper takes the sound of automobile door closing as the research object,uses the time-frequency analysis method to process the sound of automobile door closing,and extracts its signal characteristics based on the energy distribution in the frequency domain,and builds an sound quality evaluation model based on the acoustic signal characteristics extracted from the time-frequency analysis for sound quality evaluation.The main research work is as follows:Firstly,the sound samples of vehicle door closing were collected,and the subjective evaluation and traditional objective parameters were analyzed.The door closing sound samples of 25 different brands of passenger cars were collected at the speed of(1.0m/s±0.02m/s).And the consistency of the sound samples before and after processing was tested.The subjective evaluation adopts pairwise comparison method,and carries on the misjudgment analysis to the evaluation data of 24 evaluators,and obtains the final subjective evaluation value after the statistics.Considering the non-stationary characteristics of door closing sound,the characteristics of objective parameters of psychoacoustic are compared between stationary and non-stationary sound signals.The results show that the amplitude of objective parameters of psychoacoustic of non-stationary sound signals fluctuates more due to frequency attenuation than that of stationary sound signals.Therefore,it is difficult to evaluate the sound of door closing objectively by using the objective parameters of traditional psychoacoustics,and it is necessary to seek the energy characteristics of sound signal.Then,time-frequency analysis was performed on the acoustic samples that meet the requirements of consistency to study their energy characteristics.Wavelet packet transform is used to process the sound samples and the eigenvectors of the sound samples are obtained.Based on the properties of wavelet function,db35 wavelet was selected as the basis function to carry out wavelet packet analysis of door closing sound samples.According to the division of the critical frequency band,a wavelet packet filter bank is designed,and the door acoustic signal is decomposed into 24 components by the wavelet packet.The energy concept is introduced to construct the energy characteristic vector of the 24 components.GA-BP neural network was used to construct evaluation models of door closing sound quality with energy characteristic vector and traditional psychoacoustic objective parameters as variables,and the advantages and disadvantages of different variable models were compared.Genetic algorithm was used to optimize the starting weight and threshold of BP neural network.The evaluation model was constructed with the energy feature vectors extracted by wavelet packet transform and the psychoacoustic parameters after peaking or averaging processing as input and the subjective evaluation results as output.The fitting coefficients of training samples and test samples were 0.9953 and 0.9993,and the mean square errors were 0.1062 and 0.4715,respectively,for the acoustic quality evaluation model with energy characteristic vector as input.The fitting coefficients of training samples and test samples were 0.8563 and 0.9806,and the mean square errors were 2.8325 and 2.8886,respectively.The results show that the acoustic quality evaluation model based on wavelet packet transform with energy characteristic vector as input is superior to the acoustic quality evaluation model based on traditional psychoacoustic objective parameters.The correlation analysis shows that the correlation between the acoustic energy eigenvector and subjective evaluation value based on wavelet packet transform is higher than that between the traditional psychoacoustic parameters and subjective evaluation value.Four new sound samples under the same working condition were used to verify the stability of the sound quality evaluation model based on wavelet packet transform.The prediction and evaluation results of the four new sound samples were consistent with the actual subjective evaluation results,and the mean square error was only 0.4349.It shows that the evaluation model based on wavelet packet transform is more suitable for the evaluation of non-stationary acoustic signals such as door closing sound.The research results show that the time-frequency analysis method is more suitable to study the characteristics of non-stationary acoustic signals.