MDCT domain postfiltering for speech quality enhancement

In this paper, we propose a MDCT domain postfilter to suppress the noise in speech spectral valleys and allow more noise in speech spectral peaks. Since peaks and valleys of MDCT coefficients don´t correspond to those of speech spectrum, we construct both the short-term and long-term postfilter using MDCT pseudo spectrum. The short-term postfilter is induced by smoothing the envelope that is obtained by a piece-wise linear interpolation between consecutive local minima of the MDCT pseudo spectrum. And the long-term postfilter is generated by smoothing the residual spectrum that is a difference between the MDCT pseudo spectrum and envelope. Experiments show that the proposed method outperforms the G.729.1 postfilter.