To provide a voice synthesizer capable of obtaining synthesized voice having a natural intonation.
A voice synthesizer comprises: target F0 deformation amount calculating means for calculating at least one deformation amount deforming a target F0 pattern; target F0 deforming means for generating at least one target deformation F0 pattern by shifting the target F0 pattern as much as the frequency of the deformation amount; F0 deformation sub-cost calculating means for calculating an F0 deformation sub-cost corresponding to the deformation amount; sub-cost calculating means for calculating a plurality of sub-costs expressing distortion between the target deformation F0 pattern and a candidate fragment, when the target deformation F0 pattern and the candidate fragment stored in a voice database are an input; and search hypothesis expansion means that calculates per search hypothesis a total cost for a whole text to be synthesized by inputting the plurality of sub-costs and the F0 deformation sub-cost, and that outputs a candidate fragment number column which minimizes the total cost.
MIZUNO HIDEYUKI
JP2004138728A | 2004-05-13 | |||
JP2000194390A | 2000-07-14 | |||
JP2001092482A | 2001-04-06 | |||
JP2005091747A | 2005-04-07 |
Yukio Nakamura
Yoshimura Munehiro