| 摘要 |
PROBLEM TO BE SOLVED: To provide a speech recognition device capable of reducing erroneous speech recognition, even when noise is superimposed on input speech. SOLUTION: A speech recognition device 10 performs speech recognition of inputted speech through steps in which: an acoustic likelihood calculating means 16 calculates the acoustic likelihood of an acoustic model 12a and the feature vector of input speech; a reliability calculating means 15 calculates the reliability of the acoustic likelihood, on the basis of a noise model 11a and the acoustic model 12a; an acoustic likelihood correcting means 17 corrects the acoustic likelihood on the basis of the reliability; and a search means 18 searches for the candidates of an output series of connected words or the like, on the basis of the corrected acoustic likelihood and a language model 13a. COPYRIGHT: (C)2005,JPO&NCIPI
|
