| 摘要 |
PURPOSE:To decrease possible character strings greatly, and to perform effective evaluation by inputting the reliability of each voice-recognized character string after it is normalized by the number of characters, and restricting the limit of evaluation to possible character strings which have higher reliability than specified. CONSTITUTION:A monosyllable recognition part 10 consists of a parameter extraction part for each KANA character voice input, a matching part for collation with standard features, and a discrimination part for collation results, and the discrimination part outputs possible strings for each character. They are inputted to a reliability processing part 20; a reliability calculation part 21 calculates the reliability of each character in the possible string firstly, and a normalized reliability calculation part 22 calculatea the normalized reliability of the mumber of characters in the possible string. This character string is inputted to a KANA and KANJI mixed sentence conversion part 30 together with the evaluated reliability to display a final possible character string, thus making a final decision. |
