Robust voice activity detection using feature combination

This paper presents a study of noise-robust voice activity detection (VAD) utilizing combination of feature vectors extracted from speech signals. Conventional VADs are sensitive to non-stationary noise especially in low SNRs. Also situations such as cutting off of unvoiced regions of speech and random oscillation of VAD decisions are unavoidable. To overcome these problems, the proposed algorithm utilizes measures such as energy differences, periodicity, zero crossing rate, and spectral differences between different sound frames. The performance of the proposed VAD algorithm is tested on real speech signals. Comparisons confirm that the proposed VAD algorithm outperforms the conventional VADs, especially in the presence of background noise.