Hearing Aid Speech Quality Evaluation Based on MARS

Abstract

Accurate and reasonable evaluation of hearing aid speech performance is especially important for hearing-impaired patients. In this paper, an objective evaluation algorithm for speech quality based on MARS (Multivariate adaptive regression spline) is proposed. In the algorithm, several basic features of speech signals are extracted, and MARS is used to select the key features that have great influence on speech quality. Then the optimal objective prediction model is built up to map the distortion measure of the characteristic parameter onto the subjective evaluation scores, which is substituted by PESQ. Finally, experimental verification shows that the correlation between the objective evaluation by the algorithm and the subjective evaluation scores is high, which is 0.978 and 0.9333 for training samples and test samples respectively.