基于模糊域奇异值分解的语音增强方法

提出了一种遵循语音非平稳特性的增强方案。它不同于以往语音增强惯用的前提。首先，使用时频表征中的模糊函数描述被高斯白噪声恶化的语音信号；其次，对其进行奇异值分解后。采用尺度因子矩阵调整模糊函数矩阵的奇异值达到抑制与噪声相关成份的目的；最后，将模糊域下经过处理的余留成份变换为时域信号。仿真实验表明，这种消噪方法具有良好的性能。

Speech Enhancement Approach Based on Singular Value Decomposition in Ambiguity Domain

The paper produces a nonstationarity-keeping solution to the subject of speech enhancement,which has been implemented from the different point of view from the supposition of the conventional noise reduction approaches.First,we employ the ambiguity function of time-frequency representation(TFD)to describe the speech degraded by Gaussian white noise;Second,apply singular value decomposition(SVD)to it so as to suppress the components related to noise by adjusting the singular values depending on scaling factor matrix;Finally,the left components in ambiguity domain are translated into the time domain.The simulation experiments demonstrate that the noise reduction method embodies the effective performance.