In this paper we discuss the extension of an isolated word recognition system based on tied density hidden Markov models towards a digit string recognition system. Starting from the models for isolated digits we compare different schemes developed in order to adapt the models. In particular, the modular concept of the recognition system based on tied density hidden Markov models allows to modify either the parameters of the underlying densities or only the weights of the given densities in the model states.
Bibliographic reference. Euler, S. (1991): "Adaptation techniques in tied density hidden Markov models", In EUROSPEECH-1991, 919-922.