一种基于自适应谱熵的端点检测改进方法

An Improved Speech Endpoint Detection Method Based on Adaptive Band-partition Spectral Entropy

在低信噪比的环境下,为增强与噪声的区分度,提出了一种适应于低信噪比环境的语音端点检测方法。通过改进语音端点检测的特征参数,更好地区分语音信号与噪声信号,提高在低信噪比环境下的端点检测正确率。基于子带谱熵,引入正值常量对基本谱熵参数进行算法改进,得到改良的负谱熵特征,并结合自适应子带选择方法,得到一种新颖的特征参数——自适应子带常量负谱熵。特征在低信噪比的情况下有较强的抗噪能力,并能够准确地检测出语音端点。实验结果表明,不仅快速有效,具有较强的鲁棒性,而且适合低信噪比的语音端点检测。

This paper proposes a novel speech detection method used with lower signal-to-noise ratio（SNR）.The method improves the discriminability between speech and noise,and the speech detection ratio is increased with lower SNR.To enhance the efficiency of speech endpoint detection,a positive constant is introduced to basic Band-partition Spectral Entropy（BSE） to get an improved negative spectral entropy.Combined with Adaptive Band Selection（ABS） method,a novel feature parameter called Adaptive Band-partition Constant Negative Spectral Entropy（ABCNSE） is achieved.Speech endpoint can be accurately detected through this feature.Experiment results reveal that this method is robust and valid under lower SNR.