| 摘要 |
PROBLEM TO BE SOLVED: To effectively remove noise other than speech, such as environmental noise without learning of an acoustic model. SOLUTION: Speech recognition is performed on noise by a single phoneme except a silence phoneme, and the single phoneme whose likelihood is high is extracted. Speech recognition is performed to noise by a phoneme string to which the same or other single phoneme is added to the extracted single phoneme whose likelihood is high, irrespective of whether or not, it is meaningful as a language. The phoneme string whose likelihood is high is extracted again, and it is stored in a dictionary of unnecessary words. COPYRIGHT: (C)2011,JPO&INPIT
|
