VOICE ACTIVITY DETECTION IN NOISE SPECTRUM DOMAIN UNDER RAYLEIGH DISTRIBUTION
Ming Chen · Signal Processing · 2009
In this paper we propose an improved voice activity detection(VAD) algorithm by modeling the noise power spectrum density(PSD) distribution with Rayleigh model.It is well known that the probability density function(PDF) of noise PSD obtained from Welch approach has the ' tail' characteristic.It can be modeled more accurately by using Rayleigh distribution.Under the Rayleigh distribution a new expression for the decision threshold update is derived.Due to the analytical expression of the false alarm probability, the computation of the inverse complementary error function is avoided and the computational complexity of the proposed VAD is reduced. Experimental results show that the proposed VAD outperforms the VAD scheme mentioned in[5]under non-stationary noise environments.