Hello,

I have completed step 10 and I have followed the instructions on the "Run Julian" just after step 10.

I can run Julian using the old sample .dfa/.dict file with the hmms15/hmmdefs and tiedlist that was created.

What files should I use to replace the .grammar and .voca to build a new .dfa/.dict file?  I can see that a dict file was created in steps 1-10.  Should I use this file, but if do, what .dfa file should I use in the example.

Should I just create these files from scratch?

Thanks,
Mike

--- (Edited on 4/1/2010 3:53 pm [GMT-0500] by gothrog) ---

>What files should I use to replace the .grammar and .voca to build a new

>.dfa/.dict file?

See Step 1 - Task Grammar and replace the voca file with words and pronunciations from the pronunciation dictionary you used to create your acoustic models.  Step 1 also explains how to create a new grammar file.

Ken
 

--- (Edited on 4/7/2010 9:53 am [GMT-0400] by kmaclean) ---

I have 2 problems.

1.  I have just realized that it wasn't using the julian_LINUX.conf that I created even though I thought I told it to use it.  Unfortunately it used the default julian.conf.  I'm not sure why that happened.
I ran this command in dos:  run_julian.bat -input mic -C julian_LINUX.conf

2.  Now that it is using the correct .conf file. I'm getting a TON of errors during the execution of:  run_julian.bat -input mic -C julian.conf

Here is a small portion of the error message:

uw-sp+* or biphone uw-sp
v-sp+* or biphone v-sp
w-ey+sp
y-ax+sp
z-iy+r
z-sp+* or biphone z-sp
zh-ax+sp
zh-er+sp
zh-ih+n
zh-sp+* or biphone zh-sp
//////////////////////
error in reading MIKE/test.dict: 112 words failed out of 114 words
ERROR: failed to read dictionary, terminated
Terminated

I'm not sure what I'm doing wrong.  I created a .grammar and .voca on my linux virtual machine.  Then ran the mkdfa.pl script to create the .dfa and .dict. 

I thought the hmmdefs defines the acoustic model so I don't understand why it is getting those errors.

My julian.config has this set:

-dfa MIKE/test.dfa
-v MIKE/test.dict
-h MIKE/TEST/hmm15/hmmdefs
-hlist MIKE/TEST/tiedlist
-smpFreq 48000        # sampling rate (Hz)

Any idea's what I'm doing wrong?

--- (Edited on 4/7/2010 1:42 pm [GMT-0500] by Visitor) ---

>I'm getting a TON of errors during the execution of:  run_julian.bat -input

>mic -C julian.conf

Where are you getting the phonemes for you grammar/voca files?  You need to use the phonemes from same pronunciation dictionary you used to create your acoustic models.

The phone list for the pronunciation dictionary in the Tutorial and the phone list in the for the pronunciation dictionary  in the VoxForge nightly builds/QuickStart are not the same...

Did you change the tree.hed script in any way?

--- (Edited on 4/8/2010 3:13 pm [GMT-0400] by kmaclean) ---

>Where are you getting the phonemes for you grammar/voca files? 

I get it from the Voxforge/Manual/dict file.  However I massage it to look like the .voca file.

[mike@wamf Manual]$ ls
aligned.mlf    global.ded  hmm2       mkphones0.led  prompts    triphones1
codetrain.scp  hmm0        hmm3       mkphones1.led  proto      wav_config
config         hmm1        hmm4       mktri.hed      sil.hed    wintri.mlf
dict           hmm10       hmm5       mktri.led      stats      wlist
dict-tri       hmm11       hmm6       monophones0    store      words.mlf
dlog           hmm12       hmm7       monophones1    tiedlist
flog           hmm13       hmm8       MyLog          train.scp
fulllist       hmm14       hmm9       phones0.mlf    tree.hed
fulllist1      hmm15       HVite_log  phones1.mlf    trees

The .grammar looks like this:

[mike@wamf Manual]$ cat ../VRTest/test.grammar
S : NS_B WORD_LOOP NS_E
WORD_LOOP: WORD_LOOP WORD
WORD_LOOP:  WORD

>Did you change the tree.hed script in any way?

No, I didn't change the tree.hed at all.  I have an automated script that builds steps1-10.  I'll double check that it follows what the instructions say.  Everything builds without error, but I will double check to see if I messed up any files.

--- (Edited on 4/8/2010 3:47 pm [GMT-0500] by gothrog) ---

--- (Edited on 4/8/2010 3:47 pm [GMT-0500] by gothrog) ---

>dict           hmm10       hmm5       mktri.led      stats      wlist

The dict file is created in Step 2 from all the words in your prompts file.  The phonemes for each word are taken from the voxforge_lexicon  file (in the VoxForge Tutorial). 

Are you using words taken from the voxforge_lexicon in your '.voca' file?

--- (Edited on 4/8/2010 5:20 pm [GMT-0400] by kmaclean) ---

>Are you using words taken from the voxforge_lexicon in your '.voca' file?

I didn't think that I can use the voxforge_lexicon file because I didn't create a .wav file for all of those words.

I didn't copy the words from there; only out of the dict file since it was alot smaller.  It looked like they had the same values for the pronouncation.

OHHHhhhhh......  I just took a closer look.  The line in the Voxforge/Manual/dict file has the 'sp' in there.  I guess that is what is messing me up.  I have to go right now, but I will fix that using the voxforge_lexicon.  And let you know what happens.  Thanks.

--- (Edited on 4/8/2010 4:57 pm [GMT-0500] by gothrog) ---