Development of a new sound quality metric for evaluation of the interior noise in a passenger car using the logarithmic Mahalanobis distance

Abstract

Previous methods for the evaluation of the sound quality in vehicle interiors focused on the linear regression analysis of subjective sound quality metrics using statistics and estimations of subjective sound quality values by neural networks. Recently, sound quality evaluation using subjective measures has focused on identifying sound quality metrics which can predict subjective responses. It has been used to study a variety of subjective measures such as the four parameters used by Zwicker, but it is difficult to identify highly correlated sound quality metrics with the jury test. The Mahalanobis distance is a useful method to reduce the number of dimensions and to develop measures based on the correlation between the various variables. In particular, the Mahalanobis distance can be used as a new sound quality metric because it can convert the sound quality that is represented by several measures to a single value. In this study, a new sound quality metric is suggested which employs the four parameters used by Zwicker and is based on the Mahalanobis distance in order to predict subjective responses in sound quality evaluation. In addition, in order to calculate the Mahalanobis distance more accurately, after using data from a number of vehicles, sound quality metrics were reselected to remove those that do not require correlation analyses between each metric. Finally, we verified that the logarithmic Mahalanobis distance can be used not only as a new sound quality metric through correlation analysis with a jury test but also as a criterion to determine the vehicle quality. In order to verify the reliability of the regression equation, arbitrary vehicle data are applied to the regression equation. The regression equation using the logarithmic Mahalanobis distance is validated by the listening results, and the regression results after applying arbitrary data are similar.