A study of voice conversion based on sophisticated control of speaker identity founded on tensor analysis.

点击翻译按钮获取中文摘要

In this study、 we have developed voice conversion methods which realize sophisticated and flexible control of speaker identities. These techniques can be applied to welfare services and entertainment software. In this study, we have proposed a method to construct a speaker space using tensor analysis. In this method, various information included in speech utterances are properly decomposed, and these decomposed factors can be utilized for various applications in speech processing. As one of the applications of this method, a style conversion system from speaking style to singing style has been developed.