ROBUST PARAMETRIC MODELING OF DURATIONS IN HIDDEN MARKOV-MODELS

A major weakness of conventional hidden Markov models is that they imp licitly model state durations by a geometric distribution, which is us ually inappropriate. This paper presents a modified Viterbi algorithm that, by incorporating proper state and word duration modeling, signif icantly reduces the string error rate of the conventional Viterbi algo rithm for a speaker-independent, connected-digit string task. The algo rithm has essentially the same computational requirements of the conve ntional Viterbi algorithm.