Speech enhancement in joint time-frequency domain based on real-valued discrete gabor transform

In this paper, we propose a new speech enhancement method in joint time-frequency domain. Noisy speech is first transformed into the joint time-frequency domain by fast Real-Valued Discrete Gabor Transform (RDGT) where the Gaussian window is used as the transform kernel due to its superior local energy assembling ability. The MMSE based log-amplitude estimator is derived under speech presence uncertainty hypothesis and also with the assumption that speech and noise are statistically independent Gaussian random variables. Clean speech estimate is then got by inverse transform of RDGT. Experimental results show that the proposed method is very effective in avoiding the musical residual noise and retaining weak speech components.