Voice activity detection algorithm using radial basis function network

Abstract

A voice activity detection (VAD) algorithm using the radial basis function (RBF) network is proposed. The k-means clustering and least mean square algorithms are used to update the RBF network to the underlying speech condition. The inputs for RBF are code excited linear prediction coder parameters, which are stable under background noise. The RBF network output is compared to a threshold to determine the nature of the period (voice or silence). Experimental results show that the proposed VAD algorithm achieves better performance than G.729 Annex B at any noise level.