continuous recognition next up previous
Next: Contributions from this work Up: Suggestions for further work Previous: A non-restricted Recurrent Net

continuous recognition

The system can easily be modified to handle the continuous speech. It is the HMM which should be modified mainly. It is possible to modify the HMM stage using the techniques described in section 1.6. Using the eqn.1.48 error derivative for the observation probabilities can be obtained. These can easily be propagated to the other stages in a similar way as in the isolated mode.

Narada Warakagoda
Fri May 10 20:35:10 MET DST 1996

Home Page