Improved speech enhancement method based on auditory filterbank and fast noise estimation

This paper addresses speech enhancement based on the auditory filterbank and fast noise estimation. A major problem in speech enhancement based on auditory filterbanks is the incorrect estimation of the noise components. To solve this problem, a fast noise estimation technique is proposed, which works based on speech presence probability. The enhancement algorithm is realized simply by multiplication the output of the synthesis filters by a factor calculated from noise statistics in each channel. The experimental results show that the proposed method is very effective in nonstationary environments and colored noise interference with improved speech quality and a very low level musical noise.