MVNHMM Toolbox

This page is no longer updated.

Newer releases are available for download from MVNHMM new home.


This toolbox contains algorithms for modeling multivariate time series with hidden Markov models.  The software has been developed on a Linux platform and has been used on Linux, UNIX (Solaris), and MacOS X.  A quick-start manual provides basic information required to install and run the software.

Among the capabilities of the toolbox:
  • Estimation of the model parameters (EM)
  • Computation of the most likely values of latent variables (Viterbi)
  • Computation of the log-likelihood
  • Simulation of the data given the parameters of the model
  • Model selection via cross-validation
Models implemented:
  • HMM
  • Non-homogeneous HMM
  • Mixture model
  • Non-homogeneous mixture model
Among the emission distribution types:
  • Conditionally independent variables or clusters of variables
  • Multinomial
  • Tree-dependent (Chow-Liu trees and conditional Chow-Liu forests)

Previous Release

mvnhmm-06242005.tar.gz (06/24/2005)

Current Untested Releases

mvnhmm-04232006.tar.gz (04/23/2006)

mvnhmm-06132006.tar.gz (06/13/2006)

mvnhmm-06232006.tar.gz (06/23/2006)
Please read and accept the licensing terms before downloading the software.

Changes are documented in the notes file in the mvnhmm directory.


A short manual contains a brief description of the software along with a few examples.  A full version of the manual will be available in the future.  README file contains the complete functionality of the toolbox.


Email Sergey Kirshner with questions, comments, and bug reports.