Abstract: Spectral subtraction for speech signal processing is a classic method for decreasing noise which easily leads to error by regard fixed silence fragment as noise sample. Spectral entropy is an effective voice activity detection method. But detect effect will be much lower with these methods in low signal noise ratio environment, and threshold estimation make use of fixed silence fragment. Therefor，this paper presents a synchronous method of decreasing noise and voice activity detection. The experiment result proves that the method has better detection result.

Key words: spectral subtraction, spectral entropy, decreasing noise, voice activity detection

摘要： 语音信号处理中减谱法是一种传统的降噪方法，但减谱法利用固定的无音片段作为噪声样本容易产生误差。谱熵法是一种有效的端点检测方法，但在低信噪比环境下，检测效果将大大降低，并且门限估计也采用初始的固定无音片段。为此，本文提出了一种降噪和端点检测同步的方法。实验结果表明，本文的方法可以得到较高正确率的端点检测结果。

关键词: 减谱法, 谱熵, 降噪, 语音端点检测