A probabilistic speech enhancement filter utilizing the constructive and destructive interference of noise

A new probabilistic speech enhancement filter is presented in this paper considering the three state possibilities of discrete cosine transform (DCT) coefficients of noisy speech: speech absence, speech and noise are constructive, and destructive. The conditional probabilities of the three events are calculated using Gaussian approximations. Unlike conventional fixed values, the speech presence or absence probability in different spectral coefficients is experimentally calculated. A novel set of gain functions is proposed for accommodation of the aforesaid three possibilities, and merged into one, called the expected gain. It is used on the noisy speech component for enhancement. Experimental results are presented to show the effectiveness of the proposed denoising filter.