Dereverberation of Speech Signals Based on the Discrete Model of Speech Production

Authors: Marco Jeub, Peter Vary

Abstract:

Over the last years several algorithms for the dereverberation of speech signals based on the discrete model of speech production have been proposed. They have in common that they rely on a model consisting of an excitation source and a time-varying vocal tract filter. In this paper we investigate the application of the postfilter algorithm used in Code Excited Linear Prediction (CELP) speech coding for the purpose of speech dereverberation. We show that in case of a reverberant signal, the amplitudes of the unwanted peaks in the excitation signal are attenuated and that this approach is capable of reducing early reverberation. Furthermore, we compare the new approach with state-of-the-art speech source-model dereverberation algorithms and evaluate the proposed method in a single-channel approach.