A New Formulation of Coupled Hidden Markov Models

摘要: Among many variations of more complex hidden Markov models, coupled models (CHMM) have recently attracted increased interests in practical applications. This paper describes a new CHMM formulation which the joint transition probability is modeled as linear combination marginal probabilities and weights used to capture interactions among multiple HMMs. The greatly reduces parameter space for CHMM. New approximated forward procedure training algorithm are proposed reduce computational complexity level. Experimental results show that our perform better recognition task on both artifical real data compared non-coupled And still converges local maxima even though there no theorectical proof thus provides an efficient formulation.