Speech enhancement using spectral weighting with EMD-based post filtering

This paper presents a novel speech enhancement technique using spectral weighting with data adaptive postfiltering method. One disadvantage of spectral weighting method is to introduce musical noise. The empirical mode decomposition (EMD) based post filtering approach is implemented to reduce the speech distortion and musical noise effects for a wide range of signal-to-noise ratios (SNR). The EMD is used to discriminate the speech signal into high and low energy segments and then the weighting gains are adaptively smoothed based on the results of such discrimination. The gain modification is made in all frequency components to obtain the enhance speech signal. The experimental results show that proposed method is significantly more effective in speech enhancement compared to recently reported techniques.