Speech Enhancement Using Adaptive MMSE Estimator Under Signal Presence Uncertainty and Laplacian Prior

In this paper we present an adaptive speech spectral amplitude estimator when Discrete Fourier Transform (DFT) coefficients of (clean) speech are modeled by Laplacian distributions. This estimator minimizes the Mean Square Error (MSE) under signal presence uncertainty. The performance of the proposed estimator is compared to the state-of-the-art spectral amplitude estimators that assume Gaussian and Laplacian priors for clean DFT coefficients. Comparative results demonstrate the superiority of the proposed estimator in terms of speech enhancement/noise reduction measures.