| 摘要 |
PROBLEM TO BE SOLVED: To provide a speech recognition model parameter creation device capable of improving a speech recognition rate. SOLUTION: When receiving an input of non-noise speech data (S11), the speech recognition model parameter creation device creates a plurality of noise superimposing data (S12), and calculates a feature amount of the created noise superimposing data (S13). When calculation of the feature amount is finished, a normalized feature amount is obtained by normalizing the calculated feature amount. Concretely speaking, the feature amount is filtered by using a band pass filter (S14) at first, and when filtering of the feature amount is finished, the filtered feature amount is divided by its maximum amplitude value (S15). Then, a speech recognition model parameter is created by a learning section by using each normalized feature amount (S16), and the speech recognition model parameter is stored in a storage section of the speech recognition device (S17). COPYRIGHT: (C)2011,JPO&INPIT
|
