| 摘要 |
PROBLEM TO BE SOLVED: To provide a voice quality conversion apparatus for suppressing deterioration of similarity to a conversion destination speaker generated for learning voice quality conversion rule based on the restriction by enabling smooth voice quality conversion in a time direction considering the time direction of voice. SOLUTION: A voice elementary piece conversion part 1 includes a voice quality conversion rule storing part 11, a spectrum correction rule storing part 12, a voice quality conversion part 14, a spectrum correction part 15, and a voice waveform generating part 16. The voice quality conversion part 14 includes a voice parameter extracting part 21, a conversion rule selecting part 22, an interpolation coefficient determining part 23, a conversion rule generating part 24, and a voice parameter converting part 25. The voice quality conversion apparatus suppresses deterioration of similarity due to supposing an interpolation model by interpolating the voice quality rule selected in the start point of voice elementary pieces and the voice quality conversion rule selected in the end point to convert each voice parameter in the voice elementary piece and correcting spectra in the spectrum correction part 15. COPYRIGHT: (C)2008,JPO&INPIT
|
