An efficient solution to improve the spectral noise suppression rules

Typical spectral noise suppression rules employ a multiplicative gain function in the frequency domain. In this paper, we propose a modified gain function to improve the speech enhancement. The general principle is based on a theoretical development of the classical gain function in order to derive an efficient estimator for the noise reduction system. The idea combines the performances of spectral subtraction method and statistical model estimator. The obtained combination involves two parameters. These parameters control the rate of noise reduction and speech distortion. Performances of the idea were evaluated using a number of tests in various environments through speech distortion measurement. Experimental evaluations show that the modified gain function yields better results as compared to the conventional gain functions.