中文摘要：

端点检测是语音识别系统中很重要的一部分，端点检测的准确性直接影响语音识别系统的性能。提出了一种新颖的两阶段端点检测方法：第一阶段采用基于短时能量的边缘检测方法得到第一阶段端点；第二阶段采用统计直方图聚类削波求取阈值，并得到最终准确的端点。实验结果表明，与传统的端点检测方法相比，该算法在噪声环境中有更好的效果。

英文摘要：

Endpoint detect is the essential component of the speech recognition system. Accurate location of the endpoints is important for the recognition performance. In this paper＇ a novel two-phase algorithm of endpoint detect is proposed. In the first phase, coarse endpoints are acquired according to the edge detect method. In the second phase, the threshold and the exact result is got by the statistics histogram method. From the experimental result, the algorithm has a better performance than the traditional method.