High Quality Voice Conversion through Combining Modified GMM and Formant Mapping for Mandarin

A novel voice conversion system using formant mapping based on modified GMM technique is proposed in this paper. Compared with the traditional GMM technique, our modified GMM technique selects the stable frames automatically in each vowel phoneme for parameter extraction to avoid using the parameters in the transition part. With the spectral parameters extracted from the stable frames, phoneme-based GMM model is built. In the transformation procedure, the transformed formant frequencies at the main vowel phonemes are obtained using the corresponding GMM model. After this, the spectrum of the test speech is warped in frequency axis using warping functions determined with the transformed formant frequencies and original ones. Besides, fundamental frequencies, average energy and speaking rate are also transformed. Finally the converted speech is re-synthesized with the transformed parameters by high quality speech manipulation framework STRAIGHT. Perceptual results for F-M and M-F conversion show that our MOS score of the converted voice is improved from 3.8 to 4.1 and ABX score from 3.3 to 3.8 compared with IBM´s system. Comparisons with other systems are also given in this paper.