摘要 |
PURPOSE: To generate a precise pitch pattern and to synthesize voice with a natural rhythm even in the case of a sentence with many accented phrases in pitch pattern generation in voice rule synthesis. CONSTITUTION: This device is provided with a feature amount extraction part 24 extracting a bias and a dynamic level as a feature amount by using label information from a voice file 21, an accented phrase list forming part 23 forming an accented phrase list corresponding to the accented phrase among the voice file 21, a statistically processing part 28 statistically processing the extracted feature amount and the formed accented phrase list and a pitch pattern generation part generating a pitch pattern model from the statistically processed result. |