HMMConverter - a toolbox for hidden Markov models

by Tim Yin Lam and Irmtraud M. Meyer

HMMConverter is a software package for setting up probabilistic hidden Markov models and pair hidden Markov models as well as their generalized versions.

The user defines the model itself and the algorithms to be used via an XML-file which is translated directly into C++ code. The software package provides several sophisticated features, many not available elsewhere:

  • the Viterbi algorithm and the Hirschberg algorithm for generating predictions
  • linear-memory algorithms for Viterbi training, Baum-Welch training and posterior sampling training for automatic parameter training
  • banding to reduce the search-space along the sequence dimensions which can be combined with any of the prediction and training algorithms
  • posterior probababilities to incorporate prior information on the input and training sequences into the prediction and training algorithms

HMMConverter paper and software:

Updated: May 20, 2009, Irmtraud Meyer