Catalogue of Artificial Intelligence Techniques
Hidden Markov Models
Author(s): Steve Renals
Hidden Markov modelling is a powerful stochastic approach to speech recognition. It is based on the assumption that the speech signal may be approximated by a first-order Markov process. Although this assumption is erroneous, their tractability, firm grounding in statistics and the existence of a powerful and provably convergent training algorithm have made hidden Markov models the dominant technique in speech recognition. Hidden Markov models are `hidden' since each state in the model does not necessarily correspond to a particular portion of the speech signal, but contains an output probability distribution over the input space.
- Rabiner, L.R. and Juang, B.H., An Introduction to Hidden Markov Models IEEE ASSP Magazine 3 (1986) no.1, 4--16.