A voice activity detection algorithm for multichannel acoustic echo cancellation exploiting fundamental frequency estimation

Voice activity detection algorithms are intended to discriminate vocal segments from silence or other background noise periods. This is an important aspect in many speech communication systems and applications, e.g. teleconferencing systems. Therefore, a new method for detecting voiced segments in a multichannel scenario is proposed in this work by exploiting the fundamental frequency tracking through adaptive notch filters previously introduced for interchannel coherence reduction in multichannel acoustic echo cancellation systems. Novel expedients are presented in this paper for obtaining a more accurate tracking and thus for guaranteeing a correct voiced/unvoiced decision. The great advantage of the presented solution is related to the possibility of deriving a self-contained solution for multichannel acoustic echo cancellation where all the involved algorithms are based on the same fundamental frequency estimation. Experimental results prove the effectiveness of the proposed approach in correctly identifying vocal segments and silence and background noise periods.