Visualization of continuous density hidden Markov models

Continuous density hidden Markov models (CD-HMMs) are doubly stochastic processes which are extensively used in speech and image signal processing. Especially in case of isolated spoken word recognition systems, the spoken words are usually modeled using HMMs. While CD-HMMs are in extensive use, to most of the speech community the HMMs remain abstract in the sense there has been no nice way of visualizing them. In this paper, we give a visual representation for an HMM. These visuals serve two purposes, (a) they give the beginner in the area of speech technology a feel for the HMMs and (b) the HMMs of words can be visually compared quickly to check if the HMM models of any two words are similar, which could in turn cause confusion during recognition.