A hidden Markov model (HMM) is a statistical model in which the system being modeled is assumed to be a Markov process with hidden states. This model has been widely used in speech recognition and biological sequence analysis. Viterbi algorithm has been proposed to compute the most probable value of these hidden states in regards to an observed data sequence. Constrained HMM extends this framework by adding some constraints on a HMM process run. In this paper, we propose to introduce constrained HMMs into Constraint Programming. We propose new version of the Viterbi algorithm for this new framework. Several constraint techniques are used to reduce the search of the most probable value of hidden states of a constrained HMM. An implementation based on PRISM, a logic programming language for statistical modeling, is presented.
Proceedings Des 5ème Journée Francophone De Programmation Par Contraintes, 2009