VOICE ACTIVITY DETECTION BASED ON HIGHER ORDER CUMULANTS AND CONVOLUTION

Abstract

This paper refers to the application of higher-order statistical signal processing techniques (cumulant calculation) on noise reduction. The performed procedure, joined to a convolution process, results in the complete estimation (i.e., amplitude, frequency and phase recovery) of any corrupted periodic signal. The aim of this work lies in its application to the voice activity detection (VAD) for environments with high noise levels. The minimum signal to noise ratio for all experiments using the proposed algorithm was -5dB. Obtained results are highly satisfactory compared with existing models.