摘要 |
PURPOSE:To obtain high recognition accuracy by small registration labor by providing a means for discriminating whether a result of recognition is correct or not, registering automatically a voice pattern of an unknown input voice when its result is correct after a recognizing operation has been executed and shifting successively to the whole registration type. CONSTITUTION:Based on a syllable registration type as a reference, at first, a voice of about 100-150 syllables is brought to recognizing operation by only an utterance and a registration of a character-string. Subsequently, by showing a result of recognition, whether the result of recognition is correct or not is informed to a word pattern storage part 13 through a correct/incorrect input part 12. In case the result is correct, the word pattern storage part 13 registers additionally an unknown input voice pattern to a voice pattern part of a recognition object word dictionary 2 in accordance with the character-string of the result of recognition. In such a way, small labor for registering a voice will suffice, and high recognition accuracy can be secured by its use. |