A CODEBOOK-DRIVFN SPEECH ENHANCEMENT METHOD BY EXPLOITING SPEECH HARMONICITY

In this paper, we present a novel codebook-driven speech enhancement method, which applies speech harmonic structure. Our algorithm is able to remove more residual noise in harmonic bands because we combine speech harmonic structure and codebook. Firstly, we utilize the speech harmonic structure to estimate prior speech presence probability. Then we use it to appraise the noise autoregressive (AR) spectral shapes for speech enhancement application. In addition, the prior speech presence probability is also used to modify the Wiener filter. Finally, by combining the clean speech AR spectral shape codebook, we can build the modified Wiener filter to acquire enhanced speech signal. Comparing with conventional codebook-driven method, the experiments derived from four typical noises show that the proposed algorithm is able to achieve a better performance.