2000 Volume 21 Issue 1 Pages 17-24
When a conventional word database is used to synthesize words, it was thought that a lack of pitch variety in synthesis segments was the major reason why the sound quality degradation occurred in the synthesized words.In order to reduce the sound quality degradation caused by pitch modification, we developed a word synthesis method taking into consideration the pitch variety in the database.The synthesis method uses a VCV-balanced word database consisting of words uttered forcibly in type-0 and type-1 accent patterns.This database contains the VCV-segments that are the same with regard to phoneme sequence but have different pitch patterns.We also improved the pitch modification process by using the techniques that the spectral features of each pitch waveform are adjusted according to the pitch modification, and the micro prosody inherent in the VCV-segment is retained.A pitch pattern matching measurement and a subjective evaluation confirmed the effectiveness of the new synthesis method as follows.(1)The total average of the pitch pattern matching distance obtained when the proposed database is used is approximately two thirds of that obtained when the conventional database is used.(2)The sound quality of synthesized words using the proposed database has marks more than 75% of preference score against that using the conventional database.
