In these days, there have been many research works on voice conversion system. The goal of voice conversion system is to transform or modify individualities in human voices.
A voice conversion system converts the speech of speaker into a new one which is alike to that of the target speaker. Therefore, the smoothness of the converted voice, and the approximity between the converted voice and the target's are the requirements of the system.
In this thesis, we propose a new voice conversion method using codeword bigram of the target speaker to improve the quality of the converted voice. We use the Viterbi algorithm to find the optimal codeword of the target speaker by using codeword bigram.
We compared the proposed method with the previous one which does not use codeword bigram of the target speaker. The performances of the systems are evaluated through the subjective evaluation method and the objective one. As the result, the quality of the voice converted by the new procedure is proved to be better, compared with previous one.