摘要 |
PURPOSE:To generate a large amount of data in a short time and to improve the reliability of the data. CONSTITUTION:Source speech pitch pattern data 12 are separated into a phrase component and an accent component by a method of the least squares. Then a component extraction part 13 extracts a phrase component parameter by a extracting process 14. A phrase component pattern is obtained from the parameter, extracted by the extracting process 14, is obtained by a process 15. The phrase component parameter obtained by this process 15 is subtracted from the source speech pitch pattern to obtain an accent pattern by a process 16. An accent component parameter is extracted from the pattern obtained by the process 16 by an extracting process 17. Then the respective parameters are stored in phrase and accent component parameter data bases 18 and 19. |