摘要 |
PROBLEM TO BE SOLVED: To provide a new speech recognition method based upon prosody. SOLUTION: A car navigation device executes a speech recognition program 30 to calculate an acoustic feature quantity from an inputted speech signal and also calculate an instantaneous frequency feature quantity (an acoustic analysis part 31 and an instantaneous frequency analysis part 32), calculate phonetic likelihoods of a plurality of recognition candidate words on the basis of the calculated acoustic feature quantity (a phonetic likelihood calculation part 33 and a phoneme HMM part 34), calculate prosodic likelihoods of the plurality of recognition candidate words on the basis of the calculated instantaneous frequency feature quantity (a prosodic likelihood calculation part 35 and a prosodic HMM part 36), calculate a combination likelihood of a word as a recognition candidate on the basis of the calculated metric likelihoods and phonetic likelihoods, and select one of the plurality of recognition candidate words which has the highest calculated combination likelihood. COPYRIGHT: (C)2006,JPO&NCIPI
|