Homogeneity Test Based Voice Activity Detection

In this paper a new approach for voice activity detection (VAD) is proposed. This technique is based on homogeneity test of two autoregressive (AR) processes; each one models a speech window and involves the measure of a defined distance. The homogeneity test is formulated as a hypothesis test with a threshold derived analytically according to a userdefined false-alarm probability. Results using Aurora database shows the effectiveness of the proposed technique compared to other methods and standards