Enhancement of noisy speech signal based on variance and modified gain function

Single-channel Speech enhancement algorithms are widely used to overcome the degradation of noisy speech signals. Speech enhancement gain functions are typically computed from two quantities, namely, an estimate of the noise power spectrum and of the noisy speech power spectrum. The variance of these power spectral estimates degrades the quality of the enhanced signal and smoothing techniques are, therefore, often used to decrease the variance. In the proposed method Adaptive threshold is estimated using the variance in the time index. Using this threshold the gain and the speech spectrum are updated. Further the gain is modified based on the adaptive threshold estimated in the frequency bins and Enhanced signal is obtained from the product of modified gain function and the updated speech spectrum. By this method definite improvement in SNR can be obtained. Compare to the conventional method Mean square error (MSE) is much reduced in the proposed method.