A robust low complexity voice activity detection algorithm for speech communication systems

We present a robust voice activity detection algorithm which enables further bit rate reduction for practical speech communication systems. The voice activity decision is based on a set of energy and spectral measures. The local frame measures are compared to a long term estimation of the noise. The voice activity decision is obtained using a pattern recognition approach with additional decision smoothing. This algorithm was adopted as part of Annex B for the ITU Recommendation G.729.