| 摘要 |
PROBLEM TO BE SOLVED: To obtain high-quality synthesized speech, by adopting a low cost labeling method, such as automatic labeling. SOLUTION: The method comprises an elementary speech unit searching step for searching an elementary speech unit which is required for performing speech synthesis based on phonetic information, from a speech corpus; an evaluating step for evaluating goodness of fit of the elementary speech unit which is obtained in the elementary speech unit searching step; and a synthesis step for performing speech synthesis, by using an optimal elementary speech unit which is obtained in the evaluation step. The elementary speech unit searching step includes an elementary speech unit selecting step for selecting the elementary speech unit which is stored in the speech corpus; and a connection point search step for searching an optimal connection point, when the elementary speech units are connected together. In the evaluation step, the goodness of fit for the elementary speech unit which is selected in the elementary speech unit selecting step, and the goodness of fit for the connection point of the elementary speech units together are evaluated with a predetermined function having weight, and the elementary speech searching step and the evaluation step are executed repeatedly. COPYRIGHT: (C)2008,JPO&INPIT
|
