耦合MMSE和WEDM幅度谱估计的语音增强方法

Coupled MMSE and WEDM spectral amplitude estimations for speech enhancement

为同时在高SNR和低SNR条件下获得较好的语音质量,提出了一种耦合MMSE和WEDM幅度谱估计的语音增强方法。此方法利用Sigmoid映射函数将后验SNR映射到[0,1]之间。然后,根据此映射函数,提出一种自适应的谱恢复增益函数计算方法。在高SNR条件下,为避免语音,所提方法采用较大的映射函数值,从而使MMSE谱估计增益函数为谱恢复增益函数。反之,在低SNR条件下,为有效抑制含噪语音中的噪声,所提方法使用较小的映射函数值,从而选择WEDM谱估计增益函数为谱恢复增益函数。实验结果表明,所提算法性能在客观性能测试方面要优于参考算法。

In order to achieve better speech quality both in high SNR（signal-to-noise ratio）and low SNR conditions,this paper proposes a coupled MMSE（minimum mean square error）and WEDM（weighted euclidean distortion measure）spectral amplitude estimations method for speech enhancement.This method employs sigmoid mapping function to map a posteriori SNR into[0,1]range.Thus,according to this mapping function,we develop an adaptive gain function calculation method for spectral restoration.In higher SNR conditions,to prevent speech distortions,the proposed method adopts a larger value of mapping function and makes MMSE spectrum estimation gain function be the gain function for spectral restoration.On the other hand,in lower SNR conditions,to more effectively remove noise from noisy speech,the proposed method uses a smaller value of mapping function and selects the gain function of WEDM spectrum estimation to be the gain function for spectral restoration.Our experiments demonstrate that the proposed method is superior to the reference methods.