Hey,

I'm wondering if anyone has any experience using Julius + HTK with Gaussian Markov Models (GMM). I'm looking for an example/tutorial anything really that explains how to approach this. 

Many Thanks,

A

--- (Edited on 5/5/2010 10:59 am [GMT-0500] by abnev) ---

You mean Gaussian Mixture Model (GMM)!

Why do you want to use GMMs? So you don't want to do ASR with that?! Do u want to do event detection or so? Otherwise using just GMMs makes no sense.

You could try to use a HMM prototype which looks like this:
~o <VecSize> 39 <MFCC_0_D_A>
~h "proto"
<BeginHMM>
      <NumStates> 3
      <State> 2
         <Mean> 39
          0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
         <Variance> 39
          1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0
      <State> 3
      <TransP> 3
          0.0 1.0 0.0
          0.0 0.0 1.0
          0.0 0.0 0.0
<EndHMM>

This should be similar to a GMM!

Greez

--- (Edited on 5/25/2010 4:52 am [GMT-0500] by Visitor) ---