To provide the method and the device for voice synthesis in which a synthesized voice is generated that is close to a natural voice by selecting the connection units of voice pieces and feature parameters in accordance with the uttering speed and the uttering style and properly giving the continuation time of a sound syllable and a mora.
The voice synthesis method generates synthesized voices by connecting the feature parameters of voice pieces. Based on the uttering parameters, which include uttering style and/or uttering speed (S62), the connection method of the feature parameters is changed. The change in the connection method is conducted as follows: A silence interval length is decided based on uttering style and/or uttering speed (S64 and S65). Then, new feature parameters are generated (S66) by interpolating from different feature parameters against a same voice piece based on the silence interval length (S63 and S64) and the feature parameters are synthesized (S67). The feature parameters are made different in accordance with the different voice pieces of VCV, V, CV, VC or CV.
ASO TAKASHI
OTSUKA MITSURU
OKUYA YASUO