Offline and online identification of hidden semi-Markov models

M. Azimi, P. Nasiopoulos, R.K. Ward
2005 IEEE Transactions on Signal Processing  
We present a new signal model for hidden semi-Markov models (HSMMs). Instead of constant transition probabilities used in existing models, we use state-duration-dependant transition probabilities. We show that our modeling approach leads to easy and efficient implementation of parameter identification algorithms. Then, we present a variant of the EM algorithm and an adaptive algorithm for parameter identification of HSMMs in the offline and online cases, respectively. Index Terms-Expectation
more » ... imization (EM) algorithm, recursive maximum likelihood (RML), recursive prediction error (RPE), semi-Markov models. Rabab Kreidieh Ward (S'71-M'72-SM'85-F'99)
doi:10.1109/tsp.2005.850344 fatcat:jpgd7ja6kjf3bdabtq2i6mqhbe