Comparitive Analysis of Speech Enhancement Techniques: A Review

Speech signal enhancement in a non-stationary noisy environment depends on the speaker characteristics and temporal dynamics. Various techniques have been used which help in maximally extracting these features for speech enhancement. Corpus based speech enhancement methods have shown promising results. Newer methods predict cleaner speech estimates by recognizing long segments of speech as complete units. Clean speech sentences have been taken from a speech corpus for the speech recognition. Matching segments are identified between the noisy sentence and the corpus sentences. A wide range of speech enhancement methods such as minimum mean square error (MMSE), the super Gaussian joint maximum a posteriori (JMAP) estimator and the log spectral amplitude estimators have also shown good speech approximations.