采用小波变换的均方值滤波和门限值编码的语音端点检测

语音通信中语音噪声分离是一项艰巨而热门的研究课题.其中语音端点检测是最流行的方法之一.目前一种方法是检测短时平均幅度Mn和短时平均过门限率Zn.该方法的Mn和Zn参数检测不太准确.另一种是基于分形理论的检测方法.此方法要设置一个较佳的门限值通常比较困难.还有一种是基于DWT变换的方法.这种方法的互相关系数包络不能准确地表现原始语音信号的包络.为此，本文提出一种基于小波变换的均方值滤波和门限值编码的方法.本方法先对语音信号进行小尺度小波变换,然后进行均方值滤波,再进行门限值编码去确定语音端点.该方法的优点

Speech endpoint detection using mean square filtering and threshold encoding of wavelet transformation

In speech communication, speech-noise separation is a hard but hot research topic. In speech-noise separation, speech endpoint detection is one of the most popular methods. One of the methods for speech endpoint detection is to detect the short time average amplitudes Mn and the short time threshold exceeding rate Zn. In many cases, Mn and Zn can not be measured accurately by this method. Another method is based on fractals theory. In this method, it is difficult to determine an optimum threshold. Other one is DWT based, but the envelope of the correlation calculated by the method can not coincide with that of the speech signal. For this reason, a mean square filtering and threshold encoding this paper. In this method, the wavelet transform of a speech of small scale wavelet transform is proposed in signal is first carried out. Next, mean square filtering is applied to the transform. Then threshold encoding is performed to obtained the endpoints of speech. The advantage of this method is high accurate speech endpoint detection.