Abstract

 This paper describes the theory and implementation of DYNAMIC Bayesian networks in the context of isolated digit recognition. The common statistical model used in isolated digit recognition is the hidden Markov model. Bayesian networks provide an expressive graphical language for factoring joint probability distributions. The principle of this approach is to build a speech model using the formalism of dynamic Bayesian networks. In this paper we will show that how triangulation methods affect inference algorithms. We present illustrative experiments and our experiments show that this new approach is very promising in the field of isolated digit recognition.