摘要 |
PROBLEM TO BE SOLVED: To provide a speech recognition and method, capable of creating utterance register vocabulary information to facilitate coping with change of an acoustic model, without requiring re-utterance of a user. SOLUTION: In the utterance registration mode, after input voice by utterance of the user is acoustically analyzed in an acoustic analysis section 3, it is input to a recognition vocabulary creation section 6, and phoneme string information is created depending upon the used acoustic model, and feature amounts 15 corresponding to a plurality of the acoustic models are stored. Then, when the acoustic model is changed, a feature amount/vocabulary conversion processing section 16 creates a user register recognition vocabulary corresponding to the changed acoustic model by using the feature amounts 15. COPYRIGHT: (C)2008,JPO&INPIT
|