The standard hidden Markov model (HMM) has often been pointed out for
its inappropriateness in capturing state duration behavior. Explicit s
tate duration modeling in the HMM has been developed but it is not suf
ficient for modeling the intrinsically dynamic, or nonstationary, tran
sition process. Nevertheless, most research efforts have been concerne
d with only within-state nonstationarity, e.g., variable state duratio
n and regional symbol correlation. In this paper we explore the nonsta
tionarity of Markov chains and propose a nonstationary HMM that is def
ined with a set of dynamic transition probability parameters A(tau) =
{a(ij)(tau)}, a function of time duration tau. The model, when compare
d to the traditional models, is defined as a generalization of the sta
ndard HMM and the state duration HMM, with the description being given
for discrete observation distributions. Through a set of experiments,
it has been shown that the proposed model is more capable of capturin
g the dynamic nature of signals with higher discrimination power in on
-line character recognition.