Voice activity detection using entropy-based method
Ning Xu, Chengcheng Wang, Jingyi Bao · 2015
Voice activity detection (VAD) is an imperative technique in many speech applications. An efficient and accurate VAD algorithm that is robust to background noise is proposed in this paper. By calculating permutation entropy (PE), the method can not only determine the presence or absence of speech, but also distinguish voiced and unvoiced parts of speech. Experiments under several noise cases have demonstrate that the proposed method can obtain conspicuous improvements on the aspect of false alarm rates, while maintaining comparable speech detection rates compared to the reference method.