Bayesian spectral amplitude estimation for speech enhancement with correlated spectral components

In Bayesian short-time spectral amplitude (STSA) estimation for single channel speech enhancement, the spectral components are traditionally assumed to be uncorrelated. However, this assumption is not exact since some correlation is present in practice. In this paper, we propose a STSA estimator with correlated frequency components. Since its closed-form solution is not readily available, we alternatively derive closed-form expressions for corresponding upper and lower bounds. Three new speech enhancement estimators are proposed based on those bounds: one for each bound and one that is a combination of both. Results of PESQ and informal listening experiments indicate that the proposed estimators give better performances than earlier estimators.