explicit duration HMM

Started by April 4, 2014
hi,
   I'm looking for a methodology to find the optimal state sequence given an
observation sequence (using viterbi or otherwise) in an explicit duration HMM.

   All the papers on explicit duration HMM give an optimal state at time step t,
gvien an observation sequence. They give is as q(t) = argmax(gamma(i)), where gamma
is combination of the forward-backward variables alpha and beta.

The problem with the above approach is that the final resulting state sequence can
contain state transitions which have transition probability = 0.

Any help or pointers to literature is appreciated.

thanks
Ganesh
On Friday, April 4, 2014 11:28:37 PM UTC+13, ganes...@gmail.com wrote:
> hi, > > I'm looking for a methodology to find the optimal state sequence given an
observation sequence (using viterbi or otherwise) in an explicit duration HMM.
> > > > All the papers on explicit duration HMM give an optimal state at time step t,
gvien an observation sequence. They give is as q(t) = argmax(gamma(i)), where gamma is combination of the forward-backward variables alpha and beta.
> > > > The problem with the above approach is that the final resulting state sequence can
contain state transitions which have transition probability = 0.
> > > > Any help or pointers to literature is appreciated. > > > > thanks > > Ganesh
Is this for a Ph.D?