General spectral subtraction with postprocessing for speech enhancement

Abstract

Speech enhancement using spectrogram filtering techniques such as Generalized Spectral Subtraction (GSS) method, Blade method and Blade+Whipple method are described in this dissertation. All simulations are carried out on an IBM PC using Matlab, a technical computing software. This dissertation attempts to present an enhancement system that addresses the problems of residual noise and speech distortion in an effective manner. The higher speech quality is due to further reduction in musical noise. The proposed method (Blade+Whipple) is based on the timedependent frequency analysis. For speech enhancement, the proposed method practically removes all the musical noise while maintaining the speech intelligibility. The performance assessments such as time plots, spectrogram plots and objective tests confirm that the proposed method is effective for speech enhancement.