Voice conversion using Viterbi algorithm based on Gaussian mixture model

Abstract

A novel method for voice conversion based on Viterbi algorithm is proposed in this paper, which uses the matrix of transition probabilities of the target speaker to represent the relationship between the subsequent frames, and then determines the most appropriate component of the GMM by utilizing the Viterbi algorithm for each frame of the source speaker. It avoids the spectral discontinuities between adjacent frames and the spectral averaging. Both objective and subjective evaluations have demonstrated that the proposed method improves the performance of the conventional voice conversion system based on GMM.