>Is possible to create speech recognizer  on phonetic-based by HTK if input

>is word or sentence?

See the VoxForge Tutorial - it uses uses HTK to train the acoustic model, and Julius for recognition, but HTK's HVite program can be used also (see the HTK book for more info)

--- (Edited on 3/6/2011 9:32 am [GMT-0500] by kmaclean) ---

I'm sorry if I explane not clear.

Is it possible to create phonetic recognizer. Is it right when I change only pronunciation dictionary from

(example)

normaly dictionary

I             iy
need       n iy d
some     s oh m
coffee    k ow f iy
.

I like to change to

iy     iy
n     n
d     d
s     d
oh    oh
m    m
k     k

But it has error when system will generate because HMM model error. I think it can't generate tied state (transision prob.) because I don't fix sequence in dictionary.

How to solve this problem.

Best

--- (Edited on 3/7/2011 7:44 am [GMT-0600] by Noah) ---

>Is it possible to create phonetic recognizer.

If you don't need real-time recognition of phones, check forced alignment on this page: Automated Audio Segmentation Using Forced Alignment (Draft);

--- (Edited on 4/15/2011 10:58 am [GMT-0400] by kmaclean) ---