This paper proposed a novel algorithm for text-independent voice conversion based on Chinese phoneme classification and kernel eigenvoices Gaussian mixture model. The phoneme classification can avoid the disturbance of linguistic information and spectral smoothing. A speaker adaptation technique of kernel eigenvoices was employed for performing spectral conversion between speakers for each category phoneme, adapting the conversion parameters derived for the pre-stored pairs of speakers to a desired pair, which can relax the parallel constraint effectively. Objective test on the spectral conversion accuracy demonstrated that the proposed kernel algorithm can effectively exploit the nonlinearity in supervector space. In subjective listening test, an ABX test was performed and the proposed algorithm was preferred to the existing eigenvoice algorithm by 4.75%, and improved quality by 10.91% in terms of mean opinion score (MOS). Both objective and subjective tests demonstrated that the proposed algorithm effectively enhanced speech quality and speaker individuality in a text-independent manner.
