| 摘要 |
PURPOSE:To improve the recognition rate within a limited processing time by providing a means which changes the number of recognition candidates, which are selected by a candidate extracting means, in accordance with the time length of the pattern of an input voice. CONSTITUTION:The input voice pattern is temporarily collated with register patterns in a register pattern dictionary 6 in a linear collating part 4 by a simple collation method, and candidates are ranked by a sorting part 5. A number (n) of candidates determined by a candidate number determining part 10 is used to extract n-number of high ranked candidates by a candidate extracting part 7. A DP collating part actually collates the input voice pattern with these n-number of extracted candidates. Thus, the recognition rate is improved within a limited processing time. |
