The main idea behind HMMs is to integrate a simple and efficient temporal model and the available statistical modeling tools for stationary signals into a sound mathematically tractable framework. The majority of applications have been in speech recognition [Rabiner, 1989,Juang and Rabiner, 1991], but successive results are also reported in other fields, for example, in the handwritten script recognition [Cho and Kim, 1995] and in the modeling of the protein chains [Baldi and Chauvin, 1996].