| 摘要 |
PROBLEM TO BE SOLVED: To obtain high recognition performance in various noisy environments in an automobile etc. SOLUTION: A speech recognition device has a noise data decision part 14 which inputs data representing the travel state of the automobile, an ambient environment, and the operation state of mounted equipment and decides which of (n) previously classified kinds of noise data current noise belongs to, (n) kinds of sound models M1 to Mn generated by generating (n) kinds of noise superposed speech data by superposing standard speech data on (n) kinds of noise data respectively and then performing noise removal processing, a noise removal processing part 16 which inputs a noise superposed speech with the superposed noise from a microphone 11 and also inputs the noise kind decision result, and removes noise from the noise-superposed speech, and a speech recognition processing part 18 which performs speech recognition for the noise-removed speech by using the sound model corresponding to the noise kind decided by the noise data decision part 14 among the (n) kinds of sound models. COPYRIGHT: (C)2004,JPO&NCIPI
|
