Voice Activity Detection Algorithm with Low Signal-to-Noise Ratios Based on Spectrum Entropy

Abstract

This letter presents a robust voice activity detection (VAD) algorithm for detecting voice activity in noisy environments. The presented robust VAD utilizes the entropy measurement defined in band-splitting spectrum domain to exploit the formant frequency representation as a highly efficient, compact representation of the time-varying characteristics of speech. Additionally, Teager energy operator (TEO) can be employed to provide a better representation of formant information resulting in high performance of classification of speech/non-speech priori to entropy-based measurement. The results show that the proposed algorithm has an overall better performance than the standard ITU-T G.729B VAD and Shen's entropy-based VAD.