基于DWT变换的语音端点检测

Speech Endpoint Identification Based on DWT

根据语音特性,提出一种基于离散子波变换(DWT)的语音端点检测方法.该方法利用(数字)语音信号的多尺度能量分布特性和不同分辨率下重构语音信号的相关特性来刻画(数字)语音信号,从而在噪声中检测出语音信号的端点.仿真试验结果表明:基于DWT变换的端点检测法与常用的端点检测方法相比,具有更好的抗噪性和识别稳定性.

A detection method of speech start-end points is presented. In terms of speech properties, the method is based on multi-scales energy distribution and the cross-correlation function of the reconstruction digital speech signal in the different scales. Therefore, authors can keep track with the energy change along speech duration to check start-end points. Comparison with classical endpoint detection are given to show how the way performs under different signal-noise ratios. The experimental results show the method is more stability and immunity to noise.