| 摘要 |
PROBLEM TO BE SOLVED: To provide a speech synthesizer and its program that can reproduce the voice quality of a target speaker even when a voice database of the target speaker is small in scale.SOLUTION: A speech synthesizer 1 includes target-speaker small-scale statistic model/clustering information generation means 10, different-speaker small-scale statistic model generation means 30, different-speaker clustering information generation means 60, different-speaker large-scale initial model generation means 70, mapping rule extraction means 80 for generating a large-scale statistic model from an initial model of a different speaker, and extracting a mapping rule between a large-scale initial model and the large-scale statistic model, target-speaker large-scale initial model generation means 90, statistic model expansion means 100 for expanding a small-scale initial model of the target speaker into a large-scale statistic model according to the mapping rule, input sentence analyzing means 110, statistic model enumerating means 120, acoustic feature quantity string generation means 130, and speech waveform conversion means 140. |
