Abstract
The basic motivation for employing trajectory models forspeech recognition is that sequences of speech features arestatistically dependent and that the e#ective and e#cientmodeling of the speech process will incorporate this dependency.In our previous work #1# we presented an approachto modeling the speech process with trajectories. In thispaper we continue our developmentofparametric trajectorymodels for speech recognition. We extend our modelsto include time-varying covariances and...

This publication has 4 references indexed in Scilit: