Speech enhancement using modified minimun mean square error short-time spectral amplitude estimator [minimun read minimum]

To improve the effect of speech enhancement in low signal-to-noise ratio (SNR), we propose a new speech enhancement scheme using modified minimum mean-square error short-time spectral amplitude (M/sup 3/SE-STSA) estimator. We propose two methods for estimating the probability of speech absence that is used to control the soft-decision gain in each bin and in each frame. Evaluation shows that the proposed estimators can suppress the residual noise effectively and avoid" musical noise" phenomena in low SNR.