基于熵和能量的语音端点检测算法

Speech detection algorithm based on energy-entropy

短时能量与熵是语音端点检测的常用方法,但在低信噪比环境中都不能有效定位端点。因此给出结合这两种方法检测语音段的位置,同时采用自适应于不同的噪音背景下的判决准则,经实验证明该算法行之有效,对于连续数字音,准确率较原有算法平均提高16%,单个数字音提高26%。

Short-time energy and entropy features are two approaches, which have been used for locating the endpoints of utterance. However, each of them cannot locate efficiently endpoint for speech in the lower SNR. Therefore, a novel method was proposed to combine these two approaches to locate endpoint intervals, and a newly adaptive decision rule was made to various noise environment. Experiments show that the algorithm is effective and detection accuracy at 16% improvement for connected speech and at 26% improvem ent for signal number speech than previously method.