Speech detection algorithm based on energy-entropy
XU Yong-hua · Jisuanji gongcheng yu sheji · 2005
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.