Abstract：In order to study the relatively ideal subjective and objective evaluation models of car interior sound quality, the diver binaural noise samples were collected at the speed of 60 km?h-1 while different music was played inside the car. The samples were pretreated with edit, equal loudness and frequency segmentation filtering. Subjective evaluation test of interior sound quality and objective parameter calculation were conducted based on paired comparison method. Three subjective and objective evaluation models of multiple linear regression model, BP neural network model and radial basis function (RBF) model were established. The prediction effects of three models were compared. The results show that every frequency segment has large error in multiple linear regression model with poor stability, which proves that the evaluation of interior sound quality is nonlinear process. The error of overall sample is the lowest in BP neural network model except frequency from 160 to 1 280 Hz, which is lower than that of multiple linear regression model but with a quite gap from ideal error. The error can be controlled within 20％ in the RBF model with good stability. The RBF model has fine prediction performance for interior sound quality under the test condition.
2017, Vol. 38 
