基于多窗谱估计谱减法和能熵比法的语音端点检测算法

A STUDY OF ESTIMATING THE SPEECH ENDPOINT DETECTION Al GORITHM OF SPECTRAL SUBTRACTION AND ENERGY-ENTROPY RATIO BASED ON MULTI-TAPER SPECTRAL METHOD

语音端点检测是语音处理中非常关键的一个环节,目前主要的语音端点检测算法都侧重于语音特征参数的提取而忽略了之前的语音增强。论文提出一种基于多窗谱估计谱减法和能熵比的语音端点检测复合算法,该算法利用多窗谱估计谱减法将有噪声环境下的语音信号减噪,提高性噪比,达到语音增强的效果,再结合能熵比法进行端点检测。仿真结果表明,算法在低信噪比情况下,可以提高语音端点检测的正确率。

Speech endpoint detection is a very important part of speech processing. At present, the main endpoint detection algorithms mainly focus on the extraction of phonetic characteristic parameters but ignore the previous speech enhancement. In this paper, a speech endpoint detection composite algorithm based on multi-taper spectral estimation of spectral subtraction and energy entropy ratio is proposed. The algorithm can denoise the speech signal in noisy environment and improve the SNR（ signal-noise ratio） to achieve the effect of speech enhancement by using multi-taper spectral estimation of spectral subtraction.Then the energy entropy ratio method is used for endpoint detection. Simulation results show that under the condition of low SNR, the algorithm can improve the accuracy of speech endpoint detection.