| 摘要 |
PROBLEM TO BE SOLVED: To reduce creating time of a weighted finite state transducer (WFST) for recognizing that words are added. SOLUTION: When the weighted finite state transducer (WFST for recognition) which is used for speech recognition is created, the WFST is divided into a WFST in which state transition is to be added; a WFST, in which state transition is not to be added, and the WFST in which state transition is not to be added is combined and/or optimized to set as a basic WFST, beforehand; and the WFST, in which state transition is to be added is set as the basic WFST, as is, and they are respectively stored. When state transition is added, state transition is added to an indicated basic WFST in each of the basic WFST, and then, combination/optimization calculation is performed on both the WFST and a final WFST for speech recognition is obtained. COPYRIGHT: (C)2011,JPO&INPIT
|
