Joint Soft Threshold and Statistical Estimation for Speech Enhancement

This paper presents a novel method for speech enhancement based on the combination of sigmoid shrinkage and bayesian estimator. The main idea is to apply a joint detection and estimation to noisy speech before using a standard minimum-mean-squared-error (MMSE) estimator. Hence, the proposed method can take advantage of two basic approaches for improving the quality of noisy speech. Experiments performed on stationary and non-stationary noisy speech signals show that the proposed approach is promising when compared to classical methods, in terms of objective and pseudo-subjective measurements.