Combining Sub-bands SNR on Cochlear Model for Voice Activity Detection

In this paper, we proposed a novel approach to combine sub-bands SNR for Voice Activity Detection. In the proposed algorithm, a nonlinear method based on cochlear model is used to divide sub-bands. In each sub-band, two Order Statistic Filters are used to estimate the signal SNR. Through the use of the above methods, the sub-bands SNR is combined by a linear dicriminant function calculated under the MSE criterion. The effectiveness of proposed method has been evaluated on RASC863 corpus. It is shown that the proposed algorithm has more robust ability against the common VAD methods, and the non-speech hit rate is significant improved under the proposed algorithm.ε