语音端点检测中能零比方法的改进

Speech Endpoint Detection Method Based on Improved Energy-Zero Ratio

传统的基于语音信号短时能量与短时过零率之比的单参数双门限端点检测方法对高信噪比的语音信号能实现较好的检测,而在低信噪比的情况下检测正确率却很低。本文在研究了语音信号的非线性分析方法后,提出了一种改进的端点检测方法。首先,对分帧加窗后的每一帧带噪语音信号进行经验模态分解求其短时Teager能量;然后,求每一帧的短时过零率,平滑处理之后进行归一化;最后,求出短时Teager能量与归一化短时过零率之比用于端点检测。经过仿真实验证明,本文提出的改进方法能够在低信噪比的带噪环境下实现比传统能零比方法更好的端点检测效果。

The traditional single-parameter double-threshold endpoint detection method based on the ratio of short-time energy and short-time zero-crossing rate can achieve better performance of high signal-to-noise ratio speech signal,but the correct rate is very low under low SNR circumstance.In this paper,after studying the nonlinear analysis method of speech signal,an improved endpoint detection method is proposed.Firstly,the framed and windowed noisy speech signal is subjected to empirical mode decomposition to obtain the short-time Teager energy;then,the short-time zero-crossing rate of each frame is obtained,and normalized after smoothing;finally,calculate the ratio of the short-time Teager energy to the normalized short-time zero-crossing rate for endpoint detection.After a lot of simulation experiments,the improved method proposed in this paper can achieve better performance than the traditional energy-zero ratio method in the noisy environment with low SNR.