Hi All,

I have pre-recorded audio file content whose HTK AM tiedlist and hmmdefs were perefectly generated.

The Julius (-v4.3.1) X.grammar file is as -

--

S : NS_B CONVERSATION NS_E

CONVERSATION: ACTION ADJECTIVE ADVERB AFFIRMATIVE COMMON CONJUNCTION DETERMINER EXCLAMATION INDEFINITE INFINITIVE INTERJECTION MODAL MONTH NEGATIVES NOUN NUMBER PARTICIPLE PERFECT PERSON PREPOSITION PRONOUN SUBJUNCTIVE VERB WEEK YEAR

--

and X.voca file is as -

---

% NS_B
<s>             sil

% NS_E
</s>            sil

% ACTION
HELP            hh eh l p
HOW             hh aw

% ADJECTIVE
BAD             b ae d
COLD            k ow l d
DELINQUENT      d ih l ih ng k w ih n t
FOR             f er
GOTTEN          g aa t en
NOTICE          n ow t ih s
OK              ow k
TALKED          t ao k t
TOTALLY         t ow t ax l iy
VERY            v eh r iy

% ADVERB
MUCH            m ah ch
PERFECTLY       p er f ih k t l iy
SO              s ow
THAT            dh ae t

% AFFIRMATIVE
WAS             w ah z
WE              w iy
WERE            w er
WHEN            w eh n

% COMMON
RECORD          r eh k er d
RECORDS         r eh k er d z

% CONJUNCTION
AND             ae n d
BECAUSE         b ax k ah z
BUT             b ah t
IN              ih n
IN              ix n
OR              ow r
SO              s ow

% DETERMINER
MUCH            m ah ch
THE             dh ah

% EXCLAMATION

% INDEFINITE
A               ax
A               ey

% INFINITIVE
I'D             ay d

% INTERJECTION
YES             y eh s

% MODAL
CAN             k ae n

% MONTH
DECEMBER        d ih s eh m b er

% NEGATIVES
NO              n ow
DON'T           d ow n t

% NOUN
CALLING         k ao l ih ng
CENT            s eh n t
CENTS           s eh n t s
COMCAST         k aa m k ae s t
DELINQUENT      d ih l ih ng k w ih n t
DEXTER          d eh k s t er
DOLLAR          d ao l er
DOLLARS         d ao l er z
MAIL            m ey l
NAME            n ey m
NATALIE         n ae t a l iy
NUMBER          n ah m b er
PAYMENT         p ey m ih n t
SPEAKING        s p iy k ih ng

% NUMBER
EIGHT           ey t
EIGHTY          ey t iy
FOUR            f ow r
SIX             s ih k s
SIXTY           s ih k s t iy
THIRD           th er d
THIRTEEN        th er t iy n
TWENTY          t w eh n t iy

% PARTICIPLE
CALLED          k ao l d

% PERFECT
HAD             hh ae d
HAVE            hh ae v
LAST            l ae s t
UNDERSTAND      ah n d er s t ae n d

% PERSON
HERE            hh iy r
I               ay
IS              ih z
MY              m ay
THIS            dh ih s
YOU             y uw

% PREPOSITION
TO              t ax
FOR             f er
ON              aa n

% PRONOUN
ME              m iy
YOU             y uw
IT              ih t
THAT            dh ae t
MUCH            m ah ch

% SUBJUNCTIVE
SITUATION       s ih ch uw ey sh ih n

% VERB
CHECK           ch eh k
WORRY           w er r iy
THINK           th ih ng k
LIKE            l ay k
HAD             hh ae d
MAILED          m ey l d
THANK           th ae ng k
CAN             k ae n
UNDERSTAND      ah n d er s t ae n d

% WEEK
FRIDAY          f r ay d ey

% YEAR

----

The X.dfa state as -

---

0 1 1 0 0
1 26 2 0 0
2 25 3 0 0
3 24 4 0 0
4 23 5 0 0
5 22 6 0 0
6 21 7 0 0
7 20 8 0 0
8 19 9 0 0
9 18 10 0 0
10 17 11 0 0
11 16 12 0 0
12 15 13 0 0
13 14 14 0 0
14 13 15 0 0
15 12 16 0 0
16 11 17 0 0
17 10 18 0 0
18 9 19 0 0
19 8 20 0 0
20 7 21 0 0
21 6 22 0 0
22 5 23 0 0
23 4 24 0 0
24 3 25 0 0
25 2 26 0 0
26 0 27 0 0
27 -1 -1 1 0

---

and X.term being generated as -

--

0       NS_B
1       NS_E
2       ACTION
3       ADJECTIVE
4       ADVERB
5       AFFIRMATIVE
6       COMMON
7       CONJUNCTION
8       DETERMINER
9       EXCLAMATION
10      INDEFINITE
11      INFINITIVE
12      INTERJECTION
13      MODAL
14      MONTH
15      NEGATIVES
16      NOUN
17      NUMBER
18      PARTICIPLE
19      PERFECT
20      PERSON
21      PREPOSITION
22      PRONOUN
23      SUBJUNCTIVE
24      VERB
25      WEEK
26      YEAR

--

but the Julius output gives as -

$ julius -input rawfile -filelist wavlist -C julian.jconf > JuliusOutput

------
### read waveform input
Stat: adin_sndfile: input speechfile: ../train/wav/X1.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> HOW GOTTEN PERFECTLY WHEN RECORDS BECAUSE MUCH
pass1_best_wordseq: 0 2 3 4 5 6 7 8
pass1_best_phonemeseq: sil | hh aw | g aa t en | p er f ih k t l iy | w eh n | r eh k er d z | b ax k ah z | m ah ch
pass1_best_score: -476364.281250
### Recognition: 2nd pass (RL heuristic best-first)
WARNING: 00 _default: hypothesis stack exhausted, terminate search now
STAT: 00 _default: 0 sentences have been found
WARNING: 00 _default: got no candidates, search failed
STAT: 00 _default: 0 generated, 0 pushed, 0 nodes popped in 17998
<search failed>


------

Queries:

1. Is my X.grammar and X.voca perfectly fine?

2. How to make sure that the audio format matches the acoustic model format?

3. I have 15 audio files (X1.wav - X15.wav), why in every julius output I have same pass1_best as "<s> HOW GOTTEN PERFECTLY WHEN RECORDS BECAUSE MUCH"?

I am willing to share the datas when asked.  Please help.

Thanks ..!!

---