| 摘要 |
PURPOSE:To select the standard pattern information according to the overlapping state of pattern distributions among categories to each other, by using the input pattern information having the nearest distance as the limit standard of malformation among those input pattern information belonging to different categories. CONSTITUTION:An input part 1 extracts the feature quantity of an input voice and produces a pattern data to store it to a pattern data storage part 6. When the production is over for various types of pattern data of plural speakers, an interpattern distance calculating part 2 performs the calculation for the distance between the pattern data. Then the result of calculation is stored in an inter-pattern distance storage part 7. The radius is calculated for each pattern data and then stored in a radius storage part. Then a group calculating part 4 calculates the group to store the result in a group storage part 9. Finally a standard pattern selection part selects the gathering of standard patterns, and not only the selected standard pattern information but the malformation limit distance information are stored in a standard pattern storage part 10 to be used as a standard pattern dictionary. |
