A novel Chinese continuous speech endpoint detection method based on time domain features of the word structure

Abstract

Based on the time domain features of Chinese words, which are short time peak-valley energy and zero crossing, we propose a novel method for continuous speech endpoint detection. It is simple and easy to use, with high detection rate and low computational complexity. The effectiveness of the method has been verified by experiments performed on some continuous Chinese speech. Experimental results show that 96% successful endpoint detection rate can be reached for 863 speech. We have also found that the endpoint detection rate could be enhanced further if we take measures to overcome the effect of some inherent speech factors such as the speaker's speaking style, speed, coarticulation, stressing or muting.