Hi all!

I have a doubt and I need your help.

I am trying to recognize different HMMs by permitting change the order of the basic units within it. So, once I have trained the system, I need to change, manually, the transition matrix (i.e. I can return a before state and after I can jump to final state).

To change manually these probabilities I need to know what corresponds to what state. So, I need to know the segmentation (or alignment), I think made by HVite (if I am not wrong), to know what features frames correspond to what state for every utterance.

How can I do it?

Thanks a lot.

--- (Edited on 3/2/2011 5:41 am [GMT-0600] by iloes) ---

> I need to know the segmentation (or alignment)

this might help: Automated Audio Segmentation Using Forced Alignment (Draft)

--- (Edited on 3/2/2011 8:35 am [GMT-0500] by kmaclean) ---