Introduction

A hidden Markov model (HMM)[#!rabiner!#] is a statistical model commonly used to represent systems whose observable outcomes depend on the system's states, which are, themselves, not directly observable. HMMs have become very popular in several different fields, such as speech recognition, bioinformatics and computer networks.

Tangram-II's HMM Module allows users to create and work with two different classes of hidden Markov models: regular hidden Markov models[#!rabiner!#] and hierarquical hidden Markov models[#!fernando2006!#]. The regular HMM has, associated with each state, a probability distribution which determines the symbol's emission in that state. The hierarchical HMM, on the other hand, has associated with each state a Markov chain, which is responsible for the symbol emissions. TANGRAM-II supports four different types of HMMS (the regular HMM and three different hierarquical HMMs), and each is described, in details, in sections , , and of this manual.

In this chapter, we explain how an HMM model can be created with TANGRAM-II, and what interesting metrics can be generated with it.