A dual-microphone subband-based Voice Activity Detector using higher-order cumulants

The paper proposes a robust dual-microphone algorithm for Voice Activity Detection (VAD) suitable for detecting speech arriving from random directions. The algorithm is based on the use of higher order statistics (HOS) in the complex subband domain in order to effectively detect voicing segments and distinguish them from nonharmonic noise. Metrics based on new established properties of the 2^nd and 4^th-order cumulants of complex exponentials are derived. The pros and cons of each of these are analyzed and validated through simulation in various SNR conditions. The results show the proposed scheme is effective in discriminating voiced speech segments, and is robust to Gaussianlike and real-life recorded noises, even in low SNR.