A Novel Prewhitening Subspace Method for Enhancing Speech Corrupted by Colored Noise

In this paper, we propose an improved subspace method for speech enhancement in the presence of colored noise, based on a novel prewhitening technique. The colored noise modeled as autoregressive (AR) process is first used for the AR parameter estimation. Then the speech model in colored noise is changed into the one in white noise, by multiplying the noisy speech by the whitening matrix constructed by the AR parameters. Because of the novel prewhitening technique, the proposed subspace method for speech enhancement can efficiently deal with colored noise. Compared with existing subspace method, the proposed subspace method overcomes difficulty in estimating covariance matrix of colored noise. Simulation shows that the proposed approach has better performance than three conventional algorithms.