一类新的实时语音端点检测方法

A Robust Algorithm for Real-Time Detection of Speech Endpoints in Noisy Environment

该文对语音端点检测的能量状态变迁算法进行了改进,同时把能量差、Lpc系数和倒谱系数应用在改进后的状态机中,得到了新的语音端点检测算法.实验表明,这些改进算法的检测性能很好,端点检测结果准确度高.模拟实验表明,这些算法具有较强的鲁棒性,其中基于倒谱系数的能量状态方法性能最好.

The classical energy state method for detection of speech endpoints is modified, and a new algorithm for speech endpoint detection is presented. Three methods are proposed by combining energy difference, Lpc coefficients, and cepstral coefficients into the new algorithm, respectively. Experimental results show that these methods are effective with both high detection precision and robustness against noise. Experiments also show that the method based on cepstral coefficients is the most effective among the three methods.