ICASSP
2009
IEEE
13 years 11 months ago
2009
IEEE
In the Weighted Finite State Transducer (WFST) framework for speech recognition, we can reduce memory usage and increase flexibility by using on-the-fly composition which genera...