改进增益函数的MMSE语音增强算法

MMSE speech enhancement algorithm using modified gain function

讨论了基于语音短时对数谱最小均方误差(MMSE-STSA)的语音增强算法,将先验信噪比估计引入增益函数的计算中,有效消除噪声。在带噪信号模型中引入语音存在的不确定度,估计出每个频点的先验无声概率,对增益函数进行改进。通过客观与主观两种评价方法将改进算法与小波变换算法和MMSE估计算法进行比较,实验结果表明,改进算法能更好地抑制背景噪声并且使增强后的语音有较小的失真,增加语音清晰度和理解度。

The present study of speech enhancement is based on the MMSE short time spectral amplitude. In the study,subsequent to the prior SNR estimation＇s being introduced into gain function for effective elimination of noise,the uncertainty of speech-presence is introduced into the noise signal model so as to estimate the priori speech absence probability of frequency point and improve the soft-decision enhancement for gain function. According to the study,by comparing with the wavelet conversion algorithm and MMSE estimation algorithm through objective and subjective appraisal methods,the proposed algorithm proves to be superior in achieving equilibrium among noise reduction,small residual noise and enhancement of speech signals.