摘要 |
A speaker adaptation method and an apparatus thereof are provided to sufficiently accumulate observation information of a user to improve speech recognition performance. An acoustic model for each user is generated through a user confirmation process and registered with a user acoustic model database(S100). When the user speaks, the user acoustic model is loaded from the user acoustic model database to perform speech recognition(S210), and information required for speaker adaptation is output(S250). The information is received and observation data required to speaker adaptation is accumulated in a user accumulator database(S300). When the user finishes speech recognition service(S410), speaker adaptation is performed using information accumulated up to the present.
|