A New Intelligent Technique for Sound Quality Evaluation of Nonstationary Vehicle Noises

Abstract

A new intelligent technique for sound quality evaluation, the so-called wavelet pre-processing neural network (WT-NN) model, is investigated in this paper. Based on pass-by vehicle noises, the WT-NN sound quality evaluation model was developed by combining the techniques of wavelet analysis and neural network (NN) classification. A wavelet-based, 21-point model for vehicle noise feature extraction was established. Verification shows that the trained WT-NN models are accurate and effective for sound quality evaluation of nonstationary vehicle noises. Due to its outstanding time-frequency characteristics and intelligent functions, the WT-NN model is proved more advanced than the in-situ sound quality evaluation models in common use. The proposed WT-NN model can be applied to both stationary and nonstationary signals and even to transient ones. The WT-NN technique is suggested not only for the prediction, classification, and comparison of the sound quality of pass-by vehicle noises, but also for applications in other sound-related engineering fields, in place of conventional psychoacoustical models.