Acoustic Model Discussions

Flat
German AM where is the dict (Julius)?
User: psc
Date: 12/15/2011 7:47 pm
Views: 7790
Rating: 9

Hi,

I think I found the AM:
http://www.repository.voxforge1.org/downloads/de/Trunk/AcousticModels/

BUT where is the dict in Julius format, something like:
ZURICH          [ZURICH]        z uh r ix k

Thanks!

 

--- (Edited on 12/15/2011 7:47 pm [GMT-0600] by psc) ---

--- (Edited on 12/15/2011 7:58 pm [GMT-0600] by psc) ---

--- (Edited on 12/15/2011 8:16 pm [GMT-0600] by psc) ---

--- (Edited on 12/15/2011 9:09 pm [GMT-0600] by psc) ---

Re: German AM where is the dict (Julius)?
User: kmaclean
Date: 12/16/2011 9:18 am
Views: 344
Rating: 9

>where is the dict in Julius format

It is here: [   ] Lexicon.tgz 25-Sep-2010 05:52 5.1M (Ralf's German dictionary 0.2.1)

You can also take a look at the sphinx acoustic model, in the etc folder for a german pronunciation dictionary in Sphinx format

--- (Edited on 12/16/2011 10:18 am [GMT-0500] by kmaclean) ---

Re: German AM where is the dict (Julius)?
User: Pepe
Date: 4/7/2012 5:41 pm
Views: 71
Rating: 8

Hm...
The archive does not contain any *.grammar or *.voca files.

How do you tell Julius to use these files?

regards,
Pepe

--- (Edited on 4/7/2012 5:41 pm [GMT-0500] by Visitor) ---

Re: German AM where is the dict (Julius)?
User: kmaclean
Date: 4/7/2012 6:51 pm
Views: 81
Rating: 10

>How do you tell Julius to use these files?

see the VoxForge Tutorial: Running Julian Live

--- (Edited on 4/7/2012 7:51 pm [GMT-0400] by kmaclean) ---

Re: German AM where is the dict (Julius)?
User: Pepe
Date: 4/8/2012 3:10 am
Views: 91
Rating: 8

Hi Ken,

> see the VoxForge Tutorial: Running Julian Live

ok. I did that. However, the configuration file wants

sample.dfa
sample.dict

hmm15/hmmdefs
tiedlist

files. The tarball (Lexicon.tgz) consists of the following files:

Lexicon
Lexicon/dewik.rawpronSortedForms.current
Lexicon/voxDE20090209.xml
Lexicon/dewik.rawpron
Lexicon/voxDE20071211.pr0n
Lexicon/voxDE20080615.xml
Lexicon/voxDE20090209.pr0n
Lexicon/voxDE-lexicon.xml
Lexicon/dewik.rawpronSortedForms
Lexicon/dewik.output
Lexicon/german-dictionary.xml

So, either I'm too stupid (which might be the case - than I'm sorry)
to relate the files of the tarball to the appropriate configuration parameter,
or there must be a step in-between to create the hhmmdefs - or whatever
is required.


Any help is apprciated. Thanks in advance,
Pepe

--- (Edited on 4/8/2012 3:10 am [GMT-0500] by Visitor) ---

Re: German AM where is the dict (Julius)?
User: kmaclean
Date: 4/8/2012 9:51 am
Views: 69
Rating: 7

> However, the configuration file wants...

The first post in this thread has a link to:

 [   ] voxforge_de_simon-2010-04-21.tgz 12-May-2010 15:37 12M  

where you will find: 

 

  • hmm15/hmmdefs
  • tiedlist

You will have to make your own grammars(sample.dfa and sample.dict).

The VoxForge tutorial will give you an idea (see step1).  

The link to the lexicon (which is a pronunciation dictionary) gives you the pronunciations you need to create your own grammars in german.

 

 

 

--- (Edited on 4/8/2012 10:51 am [GMT-0400] by kmaclean) ---

Re: German AM where is the dict (Julius)?
User: Visitor
Date: 4/8/2012 6:41 pm
Views: 144
Rating: 9

ah... ok. I though it replaces it.

Thanks. I think I know how to go on from here. :-)

 

--- (Edited on 4/8/2012 6:41 pm [GMT-0500] by Visitor) ---

Re: German AM where is the dict (Julius)?
User: Pepe
Date: 4/9/2012 11:41 am
Views: 112
Rating: 9

Ok, most of the words I define in the voca file are reognized. :-)

However, there are some - very common - words which do not work, e.g.:

/Lexicon # egrep -e '^Wer\>' *pr0n
voxDE20071211.pr0n:Wer    v e: r

julian complains:

Error: voca_load_htkdict: line 13: triphone "v-e:+r" not found
Error: voca_load_htkdict: line 13: triphone "e:-r+*" or biphone "e:-r" not found
Error: voca_load_htkdict: the line content was: 9       [Wer]   v e: r

Any hints?

--- (Edited on 4/9/2012 11:41 am [GMT-0500] by Visitor) ---

Re: German AM where is the dict (Julius)?
User: kmaclean
Date: 4/9/2012 7:44 pm
Views: 287
Rating: 10

>Error: voca_load_htkdict: the line content was: 9       [Wer]   v e: r

include at least 3 to 5 recordings of words with the word Wer, retrain your acoustic model to include these additional recordings.

--- (Edited on 4/9/2012 8:44 pm [GMT-0400] by kmaclean) ---

Re: German AM where is the dict (Julius)?
User: Pepe
Date: 4/10/2012 1:39 am
Views: 2222
Rating: 8

Ken,

> include at least 3 to 5 recordings of words with the word Wer,

>retrain your acoustic model to include these additional

> recordings.

ok. This means these combinations are not found in the hmmdefs file and need to be recorded, right?

Ok... this time not my fault :-)

I think I will built my complete voca set and collect all missing words (there are more) and record them then.

Pepe

--- (Edited on 4/10/2012 1:39 am [GMT-0500] by Visitor) ---

PreviousNext