Voice Conversion Using HMM combined with GMM

According to HMM´s strong representation capability of speech signal and GMM´s better transformation effect, a method for LSF conversion using HMM combined with GMM is proposed. The theoretical derivation and flow diagram of this algorithm are offered, and gauss model is introduced to achieve the prosodic features transformation. The experiment is applied on two segment speech, and the result reveals that the converted speech has good naturalness. The ABX test indicates that converted speech is 90.2% similar to the target speaker’s.