| 摘要 |
PROBLEM TO BE SOLVED: To reduce the shift of fundamental frequency distribution, caused between learned voices, when the accuracy in calculating a fundamental frequency parameter is insufficient.SOLUTION: A voice synthesizer 100 generates synthesized voice waveforms from voice-synthesizing information for describing the kind of unit voice contained in a series of unit voice sequence. The voice synthesizer includes: a generator 134 for generating a first fundamental frequency time series data by predicting the first time series data, using distribution information of a first feature vector, based on a given voice synthesizing information; a generator 135 for generating a second fundamental frequency time series data by predicting the second time series data, using criterial distribution information of a second feature vector, based on a given voice synthesizing information; and a data corrector 136 for correcting the first fundamental frequency time series data using the second fundamental frequency time series data. The voice synthesizer generates a voice waveform synthesized based on the corrected first fundamental frequency time series data. |
