基于短时平均幅度差函数的带噪语音端点检测算法

传统的基于自相关函数的端点检测算法有两个方面的问题,一是计算量大,二是要进行语音信号基音周期的提取.提出了一种改进的方法,用短时平均幅度差函数代替自相关函数,节约了计算量;利用浊音与噪声平均幅度差函数的区别省去了基音周期的计算,同时也避免了误差带来的问题.传统算法与改进算法的仿真比较表明,改进算法的检测曲线噪声容限大,所以在低信噪比下也表现出了较强的稳定性.

An algorithm of end-point detection of speech with noise based on short-time average magnitude difference function

Traditional end-point detection algorithms based on auto-correlation function have two major problems. One is large computational efforts, the other is the extraction of pitch period. Put forward an advanced method, which uses short-time average magnitude difference function to replace auto-correlation function and reduces computational amount. Meanwhile, by means of the difference of AMDF of voiced sound and noise, calculation of pitch period is avoided, so is the related problems provoked by inaccurate estimation of pitch period. By simulating and comparison of traditional method and advanced method, it is found that noise margin of detecting curve of the latter is bigger, so it shows high stability even in the low SNR（ signal to noise ratio） environment.