Perceptual Gain Function for Eigenspectral Domain Speech Enhancement

The goals of speech enhancement are to improve its perceptual aspects most commonly by applying a gain function to the noisy signal coefficients in a transform domain. This gain function is normally chosen to provide a good trade-off between suppressing the noise and avoiding speech distortion. In this paper, we identify some desired gain characteristics for better sounding enhancement and propose a method for choosing the gain transfer function based on perceptual criteria. We implement our approach in the eigenspectral domain and compare our results with those from selected eigenspectral-based transfer functions.