MMSE Log-Spectral Amplitude Estimation for Single Channel Speech Enhancement Under Speech Presence Uncertainty by Weibull Speech Priors

Minimum Mean-Square Error short-time Log-Spectral Amplitude (MMSE-LSA) estimator methods extend widely in speech enhancement because of their satisfying performance. It is shown that speech enhancement methods performance can be improved by speech presence uncertainty (SPU). In this paper, we present an MMSE-LSA estimator in both the SPU mode and without SPU. Here we consider the Weibull probability density function (PDF) for the discrete Fourier transform (DFT) amplitudes of clean speech signals under an additive Gaussian noise assumption. In order to demonstrate merit of the performance of proposed approach we degraded clean speech signals by various additive non-stationary noise sources then, enhanced signals are evaluated. The evaluation results in terms of segmental signal-to-noise ratio (SEG-SNR) and perceptual evaluation of speech quality (PESQ) indicate the outperformance of the proposed method comparing MMSE estimator by Weibull PDF and MMSE-LSA estimator by Rayleigh PDF.