摘要 |
PROBLEM TO BE SOLVED: To appropriately selects an example sentence for constituting a subset which is a part of a main set in order to register a phoneme and an accent pattern in a speech data base. SOLUTION: In a first step, the indicated number of example sentences is extracted from a main set file in which the example sentence of the main set is registered to make a temporary subset. In a second step, a first evaluation value for evaluating a degree of the dispersion of the appearance frequency of the phoneme, and a second evaluation value for evaluating a degree of the dispersion of the appearance frequency of the accent pattern in the example sentence of the subset. In a third step, when one example sentence of a residual set in which the subset is removed from the main set is exchanged with one example sentence of the subset, it is determined whether the degree of the dispersion calculated by the first and the second evaluation values is increased or decreased, and when the degree of the dispersion is increased, the example sentences are exchanged. The third step is repeated, until at least either of the first evaluation value and the second evaluation value reaches a predetermined value or larger, or an exchange frequency reaches a predetermined number. COPYRIGHT: (C)2010,JPO&INPIT
|