A recursive speech enhancement system employing Laplacian speech model

A new recursive approach for speech enhancement is proposed in This work. The speech is enhanced based on the Laplacian-Gaussian mixture model, where the Laplacian factor was estimated using the noisy speech signal. The proposed algorithm uses the enhanced signal to estimate the Laplacian factor resulting in a better estimate. Then the original speech is enhanced again with this estimate, recursively. The results show that the quality of the speech can be improved with this recursive approach, if the number of iterations is carefully selected.