Two-speaker Voiced/Unvoiced Decision for Monaural Speech

Abstract

This paper presents a method for multi-speaker voiced/unvoiced decision in the case of monaural speech. This approach is based on the multi-scale product (MP) analysis of a composite signal. It consists in calculating the distances between the maxima and the minima of our proposed MP analysis technique. Then, we analyze these distances in order to make the voicing decision for both speech signals forming the mixture. Experiments are performed using Cooke and Keele databases and some mixtures from the GRID database. The results show the robustness and effectiveness of our proposed approach.