Speech endpoint detection in noisy environment using Bark wavelet transform

This paper proposes an effective speech endpoint detection method on the basis of the modified Bark wavelet transform. A new detection parameter is constructed by introducing a scale parameter into the nonlinear mapping, and combining the singularity information of the Bark wavelet transform coefficients in different sub-bands. Besides, appropriate threshold selecting rules are tested through simulation. According to the simulation results, this method outperforms the traditional methods based on short-time energy and cepstrum distance in noisy environment.