The performance analysis of Chinese speech endpoint detection based on continuous multi sub-band spectral features

Exact speech endpoint detection is very important to the integral creation of a speech recognition model. After reviewing some shortcomings of the temporal endpoint detection method, we propose two Chinese speech endpoint detection algorithms based on continuous multi sub-band spectral features, and test their performance by experiment. The result shows the detection rate of the algorithms is better than that of the temporal algorithm, which explains that they can get more effective information from speech. It also shows that the detection performance based on the mutual correlative coefficients of the continuous multi sub-band spectrum is much better, with the average detection rate more than 97%. Besides, the algorithm is low in complexity and even less in the constrained condition. Such algorithms can be used to detect the endpoint of the speech with higher SNR.