基于自相关最大值和过门限率的语音端点检测

语音处理中，在噪声环境尤其是在非平稳噪音环境下进行端点检测是很困难的。在低信噪比的情况下，传统用于端点检测的特征参数不能充分描述语音信号的特征，导致端点检测的效果严重退化。为此，笔者从语音信号的时域或频域出发，提出了一种把短时自相关函数最大值和短时过门限率相结合的方法。实验表明，该方法弥补了自相关函数最大值方法和过零率的不足，提高了语音端点检测的性能。

A Speech Endpoint Detection Algorithm Based on Maximum of Auto-correlation Function and Amended Threshold-crossing Rate

In speech processing, endpoint detection is difiqcult in noisy environments, especially in the presence of non--stationary noise. The traditional characteristic parameters for the endpoint detection can not be adequately described the characteristics of speech signals, resulting in severe degradation of the effect of endpoint detection in low SNR. So a novel method that maximum of autocorrelation function is combined with the amended zero-crossing rate is presented. Experimental results show that the method can compensating the drawback of the maximum of auto-correlation method and the zero-crossing method so that the performance of the detection is improved.