Improved prior SNR estimation for sound enhancement based on DCT

A improved algorithm named MTSNR (Modified TSNR) in DCT domain was proposed in this paper. At the second step in original TSNR with assumption of Gaussian model in the signal, this method used MMSE (Minimum Mean Square Error) technique to estimate the prior SNR at current speech frame. The proposed approach is simple and fast and requires no any prior condition on the gain function of the speech enhancement system. And it can eliminate the drawback of the TSNR method while deep its advantages. Experimental results show that the improved approach has excellent performance under different noise background.