Perceptual Subspace Speech Enhancement with Variance Normalization

In this paper a perceptual subspace speech enhancement method using masking property of human auditory system with variance normalization is presented. The masking property of the human auditory system is used while deciding the gain parameters for the algorithm. Spectral Domain Constrained estimator was employed in determining the filter coefficients and colored noise was handled by replacing the noise variance by Rayleigh quotient. Variance normalization is further done to remove the spikes in the values so as to avoid abrupt increase or decrease in power of the output samples making the output more intelligible. The objective measures SNRLoss and SNRLESC were chosen for performance evaluation based on their efficiency in determining the intelligibility of the output. The results show an improved performance of the proposed method over some of the existing speech enhancement methods in terms of intelligibility.