Quality improvement of enhanced speech in DCT domain using modified a priori SNR

This paper illustrates the superiority of using time-frequency varying averaging factor to estimate the a priori SNR for speech enhancement using spectral subtraction rules. An expression for the smoothing parameter is derived in the DCT domain. Improvements over conventional spectral subtraction rules in terms of objective quality measures e.g., log-area-ratio (LAR), output SNR and average segmental SNR are shown.