I am using julius to decode location names. it decode correctly but only detect singleword leaving other untouched

e.g wav file contains name "LIBERTY GULBERG"

and output is

### read waveform input
Stat: adin_file: input speechfile: test_wav/cle_10_50.wav
STAT: 21551 samples (1.35 sec.)
STAT: ### speech analysis (waveform -> MFCC)
### Recognition: 1st pass (LR beam)
.....................................................................................................................................pass1_best: <s> LIBERTY
pass1_best_wordseq: 0 2
pass1_best_phonemeseq: sil | L I B A R TT II
pass1_best_score: -3888.517822
### Recognition: 2nd pass (RL heuristic best-first)
STAT: 00 _default: 16 generated, 16 pushed, 4 nodes popped in 133
sentence1: <s> LIBERTY </s>
wseq1: 0 2 1
phseq1: sil | L I B A R TT II | sil
cmscore1: 1.000 0.999 1.000
score1: -3854.045166

--- (Edited on 10/9/2013 8:30 am [GMT-0500] by ) ---

I guess that it behaves like this because you selected the "Isolated word" modality:

### Isolated Word

-w dictfile

Indeed with that option Julius will output only one word (from "dictfile") found in the input audio. This means that if input file says "liberty gulberg" and you have "LIBERTY" and "GULBERG" words in your dictionary, the ASR system will output only one of those two.

One way to solve this is to create a dictionary with words like this:

LIBERTY L I B A R TT II

GULBERG G  U L B E R G

LIBERTY_GULBERG L I B A R TT II G  U L B E R G

... ...

Thus, Julius will have a chance of outputting "LIBERTY_GULBERG" with isolated word modality.

Another (more complex) solution is to design a specific grammar to be recognized, or train a LM model out of some sample text.

--- (Edited on 10/9/2013 10:57 am [GMT-0500] by azeem) ---

well thanks for reply.

but problem is that i have used grammar model (dfa,dict). so there is nothing to do with language model.

--- (Edited on 10/10/2013 12:51 am [GMT-0500] by ) ---

Oh, OK. Then you might have either

1) just got a bad recognition from Julius

or

2) your grammar does not describe what you need;

To make sure you're not in the latter case, if you want, you could share your model (dfa, dict) in this thread so we could take a look and review it.

--- (Edited on 10/10/2013 3:01 am [GMT-0500] by azeem) ---