Continuously variable duration hidden Markov models for automatic speech recognition

The solution proposed here is to replace the probability distributions of duration with continuous probability density functions to form a continuously variable duration hidden Markov model (CVDHMM) which is ideally suited to specification of the durational density.