continuous hidden Markov model

    Jun 22 2009 | 7:34 pm
    I am considering writing a HMM toolbox for Max that would support continuous observations and include a modified Viterbi algorithm that can work in real time.
    I've looked thoroughly, but can't find any continuous observation HMMs for Max. Paul Kolesnik and Marcelo Wanderley put out a discrete HMM several years ago (although never in universal binary, I believe); and I know there's a discrete HMM for FTM -- but has anyone seen anything else?
    Thanks for your help!

    • Sep 20 2009 | 1:03 am
      Hi! I'm trying to understand that in order to do gestures recognition in Max. There is a few things in about hmm. I'm working with a wii, and I'm treating right now the raw data.
      best wishes, dc
    • Nov 02 2009 | 8:48 am
      I've developed a set of Max patchers interfacing with a Java library to train and use a semi-continuous HMM that I'm about to finish this week.
    • Nov 03 2009 | 5:30 am
      Well, that's just teasing, isn't it?
      I would be most interested in seeing your results - please share.
      - Arne
    • Nov 05 2009 | 7:51 pm
      discrete time models will be just fine with me, and in best case as abstraction.
      is there a deadline already or should we set you one?
    • Nov 06 2009 | 7:45 am
      I think I set the deadline for myself (this week)
      Anybody interested can send me an e-mail and I can return the patchers, Java libraries and usage instructions zipped. It would be way to soon to put it somewhere for anyone to download. That way I can keep track of who uses it and we can discuss improvements (like beta testing).
    • Dec 07 2009 | 8:26 pm
      Hello, I am interested in beta testing and checking out your HMM package. I am working on a project using HMM and I have been using MatLab, but I would be very interested in trying it out in Max/MSP. Marc, I could not find your email to contact you. You can email me at If you have time to send a zip I would be most greatful!