| 摘要 |
<P>PROBLEM TO BE SOLVED: To easily create synthesis voice of high sound quality by using a limited number of phonemes. <P>SOLUTION: A phoneme sequence composed of phonemes, one phoneme for each segment, is selected from a phoneme group stored beforehand so that a weight sum of a target cost for indicating difference of phoneme environment of target voice and rhythm information, and a connection cost for indicating a shift amount of connection between phonemes, may become the minimum. A consecutive two phonemes in which the connection shift amount is relatively large is selected as a correction object phoneme pair, and the phoneme pair in which the connection shift amount is smaller than that of the correction object phoneme pair in the phoneme group is selected as a phoneme pair for correction. The correction object phoneme pair is corrected by adding the phoneme pair for correction, and the correction object phoneme pair by making a weight for the phoneme pair for correction larger, as they get closer to a connection border. The synthesis voice is created by connecting the phonemes in the phoneme sequence including the correction object phoneme pair after correction. <P>COPYRIGHT: (C)2010,JPO&INPIT |
