摘要 |
PROBLEM TO BE SOLVED: To provide a speech recognition device having high performance of speech recognition without increasing the data amount of feature values with respect to unnecessary words. SOLUTION: This speech recognizing device 100 is provided with a speech analysis part 103 for extracting the feature value of an uttered speech divided into a prescribed frame, a first similarity calculation part 106 for calculating the similarity of feature value on the basis of feature value patterns of a plurality of keywords to be recognized, stored in a keyword model database 104 and the feature value of each frame, and a second similarity calculation part 107 for calculating the similarity of feature value on the basis of feature value patterns of unnecessary word elements to be recognized, stored in a garbage model database 105 and the feature value of each frame. The second similarity calculation part 107 adds a correction value calculated on the basis of ambient sounds of the uttered speech by a correction value processing part 108 to calculate a similarity to recognize the keyword included in the uttered speech. COPYRIGHT: (C)2004,JPO
|