Two-Channel Speech Enhancement System Based on Auto Gain Control

We propose a new method of speech enhancement based on auto gain control using two channel inputs. Auto gain control is considered to be less effective for reducing noise superimposed on speech, however, it has advantages on the problems of musical noise and spectral distortion of the speech. In this method, two operations are combined for obtaining accurate gain, of which, one is spectral subtraction (SS) and the other is self offset of the noise with pre-whitening. The proposed method is evaluated in the experiments across three noise conditions where (i)impulsive noises, (ii)stationary car noise and (iii)speech noise present respectively. Objective measures, spectrograms and informal listening tests demonstrate significant improvements over other twomicrophone based methods.