Speech enhancement using a minimum mean-square error log-spectral amplitude estimator

In this correspondence we derive a short-time spectral amplitude (STSA) estimator for speech signals which minimizes the mean-square error of the log-spectra (i.e., the original STSA and its estimator) and examine it in enhancing noisy speech. This estimator is also compared with the corresponding minimum mean-square error STSA estimator derived previously. It was found that the new estimator is very effective in enhancing the noisy speech, and it significantly improves its quality.