Hello,

I have found a script which can be used for training/testing phone recognition in HTK. Link - http://www.humanities.mcmaster.ca/~tjyoon/resource/htk.m

Currently, it's doing CI phone training. If I modify it so that CD models are created instead, will those models be properly used during decoding? The reason I'm asking this is the gram file for decoding contains all phones in parallel. This probably means that each phone is considered as a word. So situation is different from phones *inside* a word. Does this mean the triphone models are somehow not used effectively during decoding? Can somebody please tell me if this is a proper way of doing phone recognition?

Thank you.

--- (Edited on 11/12/2012 4:30 am [GMT-0600] by voxUser123) ---