一种复杂噪声环境下的语音端点检测方法

噪声环境下的语音端点检测在语音识别系统中占有十分重要的位置。为了提高端点检测的鲁棒性和实时性，本文提出了一种延迟分割策略：以能频比为特征参数确定粗端点，并在此基础上使用排列熵算法确定精确端点，以精确端点为起始点分割语音信号，对所得到的语音片段信号按照分类标准消除噪声信号带来的错误分割。在TIMIT连续语音库与NOISEX-92标准噪声库上的实验表明，文中提出的方法比基于常规的基于零能与谱熵的方法有更好的检测效果，特别是在低信噪比的情况下，效果尤为突出。同时由于排列熵算法的简单易实现，算法的实时性表现非常好，能够为嵌入式移动通信设备提供精确快速的语音端点检测技术。

A Robust Algorithm for Detection of SpeechEndpoint in Noisy Environment

Accurate Speech endpoint detection in adverse environments is very important for robust speech recognition. To improve the noise robustness and real-time performance of endpoint detection, this paper proposes a delay strategy using energy frequency ratio to find a coarse endpoint and using permutation entropy to locate the precise endpoint. And then use classification criteria to eliminate the incorrect results. The results of experiment based on the TIMIT and NOISEX-92 database show that the proposed algorithm can locate the endpoint more accurate and quicker than the traditional method.