Coherence-based voice activity detector
Sofia Ben Jebara · Electronics Letters · 2002
Speech activity detection in noise is considered. It is based on the coherence measure between the noisy signal and its prediction residue. The decision rule is based on thresholding and is applied in different multi-frequency bands, according to speech characteristics. The proposed algorithm has the advantage of simplicity of implementation and a small number of parameters to adjust. The proposed algorithm is validated by simulations in a high noisy environment.