A new voice activity detection method using maximized Sub-band SNR

Abstract

This paper presents a novel voice activity detection (VAD) method using Maximum Values of Sub-band SNR (MVSS) as the detection feature. The proposed new feature MVSS has different distributions between speech and non-speech signal, which is helpful for separating the speech signal from heavy noise. An adaptive threshold is applied to improve VAD accuracies and track the noisy signal rapidly without complex computation. Experimental results show that the proposed method achieves better performance than the conventional ETSI AMR VADs under the NOISEX-92 database.