Mmse Speech Spectral Amplitude Estimators With Chi and Gamma Speech Priors

Abstract

We present two novel algorithms for optimal MMSE estimation of the speech spectral amplitude, assuming it has been corrupted by additive and uncorrelated noise. The noise DFT coefficients are modelled as Gaussian random variables, while the speech spectral amplitude is modelled using either a Chi or a Gamma distribution. The influence of the priors' shape parameter is investigated. Results from simulations that demonstrate the performance of the proposed algorithms for different noise types and SNRs, as well as a comparison with previously developed MAP estimators are presented.