| 摘要 |
<P>PROBLEM TO BE SOLVED: To structure a more suitable speech synthesis dictionary in order to increase the sound quality of a synthesized voice since when a speech is synthesized by referring to a speech synthesis dictionary structured by learning based upon HMM from a speech database, the speech deteriorates in sound quality because of disorder of a spectrum caused before or during the learning. <P>SOLUTION: When the speech synthesis dictionary is structured, the learning based upon HMM is performed after spectrum analysis of speech data extracted from the speech database is taken and disorder of a generated spectrum is corrected. When the learning comprises multiple stages, learning of a next stage is performed after disorder of a spectrum generated in each stage is corrected. Thus, spectrum correcting operation is performed before or during the learning, so a more suitable speech synthesis dictionary is structured as a speech synthesis dictionary that a speech synthesizer refers to. <P>COPYRIGHT: (C)2008,JPO&INPIT |
