| 摘要 |
<p>PURPOSE: To reduce the deterioration in speech quality and decrease the arithmetic quantity at the time of speech synthesis by cutting a pitch waveform with a window function as to all peaks in a specific section and making the length of the window function shorter than lengths to both adjacent peaks. CONSTITUTION: A management part 2 is connected to plural state holding parts 3, which are connected to sample read parts 5 one to one. Further, waveform holding parts 9 are connected to sample read parts 5 one to one, and the outputs of the sample read part 5 are put together in one, inputted to a weighted superposition part 6, and passed through an amplitude control part 4 and outputted from an output part 8. Further, a compressed speech element pieces DB10 is all connected to all the sample read part 5. Then speech element piece data are generated by cutting the pitch waveform with the window function which is shorter in length that the lengths to both adjacent peaks as to any of peaks present in a pitch cycle in a specific section of the speech waveform, and stored in the compressed speech element piece DB 10.</p> |
