中文摘要：

首先分析讨论了小波变换的原理，在此基础上提出了一种利用小波系数方差识别含噪语音信号中静音与语音的新算法。算法首先对含噪语音进行小波分解，观察各层小波系数的统计特性，提取它们的方差作为检测特征，从而进行语音端点检测。对该算法进行了仿真实验，并与传统的基于能量与过零率的端点检测算法进行了比较。实验结果表明：该算法在低信噪比条件下也能够有效分割语音。

英文摘要：

Speech endpoint detection is a key technology for speech recognition. It is difficult to exactly detect endpoint under low SNR, especially in silence segment or before pronouncing or after pronouncing. This paper first discussed the principle of wavelet transform, based on which, a new speech segmentation algorithm using the variance of the wavelet coefficients was proposed. Speech signal with noise was decomposed by wavelet to investigate the statistic characteristics of wavelet coefficient and different characters were obtained to detect speech signal. Simulations were made under different signal-to-noise ratios and were compared with traditional speech endpoint detection algorithm based on energy and zero-crossing rates. The results show that this method is efficient to segment noisy speech even at a low signal-to-noise ratio.