A Voice Conversion System can change the speech of source speaker into the target speaker's without changing the meanings. Recent years, the research of Voice Conversion System becomes a key area of the speech signal process, because it will improve and push the development of other area of SSP, like speech recognition, speech synthesis, etc. Based on the special feathers of Mandarin, This paper will give a new combined enframing algorithm during pre-processing period, a new GMM training strategy based on Phoneme during training period, and in transform period, a combined method of GMM and codebook mapping algorithm will be used, which will highly improve the quality of the transformed voice. |