Hi All,

How to UNDERSTAND and WRITE efficient JULIUS GRAMMAR and VOCA. I did referred the link - http://julius.sourceforge.jp/en_index.php?q=en_grammar.html but wish to have more examples or any document(papers) which suggest in writing INDEPENDENT acoustic grammar in ENGLISH.

I need some examples in ENGLISH to understand. Any pointer is well appreciated. To start with given below sentences as saved in audio .wav file -

[A] thank you so much for calling AOL this is tony speaking how can i help you
[B] tony my name is michael
[B] can you understand me because i have a very bad cold
[A] no don't worry i totally understand you perfectly
[B] you can understand me ok
[A] yes

Now I have written X.grammar file as -

S       :       NS_B    CONVERSATION    NS_E

CONVERSATION:   VERB
CONVERSATION:   PERSON  VERB ADVERB
CONVERSATION:   ADVERB  VERB
CONVERSATION:   ADVERB

whose X.voca file is -

% NS_B
<s>             sil

% NS_E
</s>            sil

% VERB
THANK           th ae ng k

% PERSON
YOU             y uw

% ADVERB
MUCH            m ah ch
SO              s ow

which only written to READ -> 'thank you so' the starting conversation by A but wish to write complete GRAMMAR and .voca file for above as having problems while writting with errors as I read word 'MUCH'. 

The o/p obtained is -

### read waveform input
Stat: adin_sndfile: input speechfile: ../train/wav/X.wav
Stat: adin_sndfile: input format = Microsoft WAV
Stat: adin_sndfile: input type = Signed 16 bit PCM
Stat: adin_sndfile: endian = file native endian
Stat: adin_sndfile: 48000 Hz, 1 channels
STAT: 2880000 samples (60.00 sec.)
STAT: ### speech analysis (waveform -> MFCC)
### Recognition: 1st pass (LR beam)
pass1_best: </s> THANK YOU SO <s>
pass1_best_wordseq: 1 2 3 4 0
pass1_best_phonemeseq: sil | th ae ng k | y uw | s ow | sil
pass1_best_score: -462639.750000
### Recognition: 2nd pass (RL heuristic best-first)
STAT: 00 _default: 9 generated, 9 pushed, 6 nodes popped in 17998
sentence1: </s> THANK YOU SO <s>
wseq1: 1 2 3 4 0
phseq1: sil | th ae ng k | y uw | s ow | sil
cmscore1: 1.000 1.000 1.000 1.000 1.000
score1: -462666.656250

The julian.jconf parameters are -

-dfa sample.dfa

-v sample.dict

-h hmm15/hmmdefs
-hlist tiedlist

-penalty1 5.0           # first pass

-notypecheck

-iwcd1 max     

-gprune safe     

-b2 200                 # beam width on 2nd pass (#words)
-sb 200.0      

-spmodel "sp"           # HMM model name

-nostrip 

-rejectshort 800

-lv 2000                # threshold of input level (0-32767)
-headmargin 300 # head margin of input segment (msec)
-tailmargin 400 # tail margin of input segment (msec)
-zc 60                  # threshold of number of zero-cross in a second

-smpFreq 48000

I am newbie so need some help. Any stardard way to WRITE JULIUS GRAMMAR in ENGLISH is well appreciated.

Also, completing above as an exercise would be good example to understand.

~ Thanks!!