| 摘要 |
PROBLEM TO BE SOLVED: To provide a speaker selecting device for voice quality converting voice synthesis that can select a registered speaker very close to a conversion target speaker by preparing a mapping code book, taking account of the whole acoustic space of a plurality of registered speakers, and provide a voice synthesizing device using the speaker selecting device. SOLUTION: A memory 13 is prestored with a voice data base including the acoustic feature parameters of a plurality of registered speakers, and its code book. A mapping code book generating part 6 computes a plurality of moving vectors between each acoustic space of a plurality of registered speakers and the acoustic space of a target speaker on every registered speaker using a moving vector field smoothing method on the basis of a voice signal of at least one word of the inputted target speaker, the voice data base and its code book, computes an average moving vector which is the average vector of the moving vector of the respective registered speakers, and selects the registered speaker, shortest in the distance between the average moving vector and a plurality of moving vectors, as a selected speaker closest to the target speaker whose voice quality is to be converted. |
