| 摘要 |
PURPOSE:To quickly generate a speech code book which is high in the classifying ability of categories at a high speed in consideration of the relation among different kind of categories. CONSTITUTION:In a phoneme code information amount calculation process 400, the separation extent of the categories and the information amount of phoneme codes are calculated for individual partial data divided in a dividing process 300 for division into partial sets and in a selecting process 500 for divided partial sets, partial data having the lowest separation extent is determined. In a dividing process 600 for the partial sets and an updating process 700 for code vectors, the partial data are divided repeatedly in order so that the category separation extent becomes large. Consequently, the code book can be generated at an extremely high speed in consideration of the classifying ability of the categories although the processes are simple. |
