β -order perceptually motivated MMSE estimation for speech enhancement

Abstract

This paper derives a β -order perceptually motivated minimum mean square error (MMSE) spectral estimator of the short-time spectral magnitude of speech, aimed to improve the performance in speech enhancement. The performance of the proposed estimators with different values of β has been evaluated by two kinds of objective measures. We have also compared the waveforms of perceptually motivated MMSE (PM) estimated speech and β -order PM estimated speech. The results showed the proposed estimator with different constant β value could focus on either noise reduction or preventing speech distortion. With an adaptive value of β, the proposed estimator have better performance in noise reduction than PM estimator.