Predication Of Door Closing Sound Quality Based On EEMD And SVM

With the vigorous development of automotive technology,people have not just to the requirement of automobile driving and walking function,to comfort and driving experience at the same time put forward higher requirements.In order to improve the vehicle comfort,reduce the interior noise has become a major as the focus of attention and research.The noise of the traditional research usually in reducing A levels as the main target,noise not only has the branch of the size,however,there is also the stand or fall of quality.Sound quality research is in the study of traditional noise joined the people's subjective factors,aimed at improving the noise to people's subjective feeling,therefore is of great significance for improving the vehicle riding comfort.Door closing sound is one of the most common noise occurs in the car using scene,the quality of door closing sound largely determines the level and the comfort of a car.Based on 15 cars' door closing as the research object,use the HEAD acoustics company's digital dummy HEAD model to collect every car close voice signal,and process the collected signals by methods such as loudness equation,screening,eventually get 24 valid samples of door closing sound.Choose grade evaluation method to make subjective evaluation test,then get voice sample performance value of subjective evaluation through calculation and analysis of the evaluation result.ArtemiS software is used to calculate the traditional psychoacoustic parameters,get the sound samples sharp degree,objective parameters such as loudness,roughness.The results of subjective evaluation and objective parameter correlation analysis proved that sharpness and loudness had higher correlation between subjective evaluation and objective parameter.There are some disadvantages for using traditional psychoacoustic parameters to reflect the characteristics of unstable noise.So introduce average empirical mode decomposition(EEMD)feature extraction method of noise signal,get contains the local characteristics of original signal in different time scales signals-intrinsic mode function(IMF).Based on the IMF components,respectively introduce the concept of sample entropy and energy eigenvector for door closing sound quality structuring,finally get the two groups of sound quality signal feature vectors respectively based on the sample entropy and energy.To get in front of the three objective parameter can reflect the quality of door closing sound,the introduction of support vector machine(SVM)with the result of subjective evaluation and objective parameters for mathematical modeling.Through the cross validation method and the grid search method to optimize the kernel function parameter.Then process eigenvectors by normalization and dimensionality reduction method,and took the feature vector as input value,the subjective evaluation performance value as output value to train the support vector regression machine.Finally got three subjective and objective forecast model.After contrasting and analyzing the forecast precision of three models,the results showed that the prediction model based on the sample entropy had the best forecast effect.The conclusion make a firm foundation for the further improvement of sound quality.