Spectral Subtraction Speech Enhancement with RASTA Filtering

Spectral subtraction based speech enhancement methods are known to be effective for the suppression of additive stationary, broadband noise. Tonal noises such as car horn sounds are found to cause serious degradation of the output speech quality. A method is proposed in this work that incorporates RASTA processing within the framework of spectral subtraction in order to achieve better suppression of tonal noises. It is shown that the proposed method significantly outperforms both, spectral subtraction and RASTA speech enhancement methods, in the presence of simultaneous broadband and tonal noises.