An Introduction to Bayesian Networks for Automatic Speech Recog

Bayesian Networks are a particular type of Graphical Models, providing a general and flexible framework to model, factor, and compute joint probability distributions among random variables in a compact and efficient way. For speech recognition, a BN permits each speech frame to be associated with an arbitrary set of random variables. They can be used to augment well-known statistical paradigms such as Hidden Markov Models by decomposing each state into several variables, outside acoustics representing for instance articulators or speech rate. Factoring joint probability distributions may potentially lead to more meaningful state representations as well as more efficient processing. Bayesian networks have also been applied for language modeling. Bayesian networks are rather new in the field of automatic speech recognition. Within the scope of a term paper, we provide an introduction to their main properties and give some examples of their current use.