Time Series Classiication Using Hidden Markov Models and Neural Networks

| This paper gives a brief overview on statistical classiication of time series using Hidden Markov Models. To overcome some limitations and assumptions made in the Hidden Markov Model framework two successful hybrid extensions using Neural Networks are presented. At rst, a short description of traditional stochastic-based Hidden Markov Model classiiers with diierent kinds of emission probabilities are given. The rst proposed hybrid approach uses a neural network that approximates arbitrary emission densities in a model-free way. The second hybrid system uses discrete models and a neural network that is trained to work as optimal vector quant-izer. The paper compares both systems and integrates them in the traditional stochastic model framework. In some experiments dealing with time series derived from speech signals and pen traject-ory based handwriting signals these methods are compared.