A new method for voice activity detection based on sparse representation

This paper presents a novel approach for Voice Activity Detection (VAD), based on the sparse representation of an input noisy speech over a learned dictionary. For this purpose, we first generate sparse representations of the input noisy speech by Orthogonal Matching Pursuit (OMP) sparse decomposition method with an over-complete speech dictionary learned from clean speech using K-SVD. We then propose a criterion to recognize the speech frames from non-speech frames. Experimental results demonstrate that our VAD approach has a good performance in low SNR conditions and outperforms than current VAD methods.