| 摘要 |
PROBLEM TO BE SOLVED: To correct a speaker's voice by an easy method if the voice is misrecognized or it is decided that the voice can not be recognized. SOLUTION: When a voice is inputted to a voice input part 2, a voice recognition part 1 takes a frequency analysis of the inputted voice to generate the feature pattern of words and match it against feature patterns of words registered in a recognition dictionary 6 and then outputs operation information corresponding to the feature pattern of matching or similar words to an operation part 4, thereby placing the operation part 4 in operation. When the operation of the operation part 4 is reflected on the intention of a speaker or when it is decided that the speech recognition part 1 can not recognize the voice, the same voice is inputted to the voice input part 2 and the operation part 4 is manually operated to select the operation contents that the speaker intends. Then the speech recognition part 1 reads in the operation information corresponding to the operation contents of the operation part 4 and the feature pattern of the words which are currently generated are additionally registered as the feature pattern of the words corresponding to the operation information read in an additional registration part 6b of the recognition dictionary 6.
|
