A Silence/Voice Segment Detection Method of Speech Signal Using Wavelet Transform Parameters

As to the problem of low accuracy of the end detection in the commonly-used detection method of the speech end in recent years, this paper proposes a silence/voice detection method of speech signal using wavelet transform parameter. Using wavelet´s ability of frequency segmentation and energy focusing, the statistic parameters of the speech signals on different sub bands are extracted. Then importing parameters validity analysis based on fuzzy entropy, we get the most discriminable and stable parameters of different sub bands as the discrimination parameters. Simulation experiment results prove this method is stable and effective under different noise conditions, and get some improvements in precision and robustness compared with the method based on the traditional parameters.