Poor recognition after training a word model

[Anonymous]

Hi,
I trained a word model for 26 words. Basically I ran all the perl scripts mentioned in the tinydoc.txt
I converted them to S2 formats using make_s2_models.pl.  I need some clarifications.

1. After conversion I got .ccode, .xcode, .d2code, & .p3code files apart from .chmm, phone, map & sendump files. In the Acoustic model(model/hmm/6k directory) downloaded from the sphinx these .ccode, .xcode, .d2code & .p3code files are NOT present. Are these files needed & if yes how are they useful?

2. I used the S2 model obtained for speech recognition but the result was disastrous because any word I spoke I got the same word as the result. Please tell me whether all the perl scipts mentioned in tinydoc.txt ( make_feat, scripts 00.verify, 01.slave.VQ.pl ... make_s2_models.pl) have to be run for a WORD model?

Thanks
Edison

[Anonymous]

On running the sphinx2-batch ( force alignment ) I got the following output  -

   Use CMU Sphinx2 to perform phonetic alignment of an audio file
   to a known text transcription.  Shows an example.
D:\sphinx\sphinx2-0.4\src\libsphinx2\time_align.c(478): rcsid == $Id: time_align
.c,v 1.13 2001/12/11 00:24:48 lenzo Exp $
D:\sphinx\sphinx2-0.4\src\libsphinx2\time_align.c(495): state_bp_table size 937K

D:\sphinx\sphinx2-0.4\src\libsphinx2\time_align.c(499): phone_bp_table size 156K

D:\sphinx\sphinx2-0.4\src\libsphinx2\time_align.c(503): word_bp_table size 15K
0 compound words found
0.248 = AGC MAX
0.275 = AGC MAX
0.206 = AGC MAX
0.278 = AGC MAX
0.214 = AGC MAX
0.259 = AGC MAX
0.302 = AGC MAX
0.294 = AGC MAX
0.258 = AGC MAX
0.241 = AGC MAX
0.253 = AGC MAX
0.186 = AGC MAX
0.252 = AGC MAX
0.251 = AGC MAX
0.235 = AGC MAX
0.193 = AGC MAX
0.300 = AGC MAX
0.276 = AGC MAX
0.284 = AGC MAX
0.293 = AGC MAX
0.253 = AGC MAX
0.230 = AGC MAX
0.337 = AGC MAX
0.197 = AGC MAX
0.259 = AGC MAX
...

Can someone tell if this output is normal(healthy)?

Thanks in advance
Edison