一种基于改进能零法的连续语音端点检测方法

The Continuous Speech Endpoint Detection Algorithm Based on Improved Energy-Zero Method

端点检测是语音识别和语音情感识别系统中极其关键的一步，其检测的效果直接关系到后续的参数计算和识别的结果．在分析了经典的基于短时能量和短时过零率的端点检测算法（能零法）的基础上，总结了其不足之处，并提出了改进的方法．改进后的算法通过对相邻两帧信号的短时能量正向做差来确定语音信号的起始点，反向做差来确定语音信号的终点；并且利用信号与背景噪声的短时过零率之比来修正语音信号的终点．MATLAB仿真结果表明，改进后的算法具有很好的端点检测效果．

Endpoint detection is a crucial step in the speech and emotion recognition system, its test results are directly related to the calculation of the follow-up and identify the results. Based on the analysis of the classic short- time energy and short-time zero-crossing rate of endpoint detection algorithm （ energy-zero method）, we summed up its deficiencies, and proposed improvement measures. Improved algorithm ascertains the starting point of speech signal through making positive difference of short time energy between two neighboring frame, on the contrary, making negative difference ascertain the ending point, and amend location of the ending point using the rate of the short time zero-crossing between signal and background noise. The simulation results of using MATLAB showed that the improved algorithm has a good endpoint detection effect.