VoxForge
VoxForge
Hi, am a total noob with acoustic models and cmusphinx. i downloaded sphinxbase and pocketsphinx and immediately saw it work in action.
I then (as i am dutch) found this great website and wanted to recognise dutch words but failed.
however, using the acoustic model in
/usr/local/share/pocketsphinx/model/hmm/en_US/hub4wsj_sc_8k/
and
some data generated using the lmtool online webservice i got at least something back allhtough as expected not a lot of accuracy considering i was feeding in dutch words.
when i compare the files in /usr/local/share/pocketsphinx/model/hmm/en_US/hub4wsj_sc_8k/ and the files in the dutch acoustic model i can see all is there. i just can't get it to work.
can anyone send me some pointers on how to use the dutch acoustic model with pocketsphinx.
i.e. command to run which file to point the -hmm switch to.
For the moment just a single recognised word would already make me a happy guy and pull through as i am pulling my hairs out for the last 4 nights with no effects, well... less hair 
.
hopefully someone can me enlighten me as i'd love to see it in action in my native language.
keep up the good work ,and thanks in advance.
--- (Edited on 4/26/2012 6:07 pm [GMT-0500] by pivposie) ---
>how to use the dutch acoustic model with pocketsphinx.
best place to look is on the CMU Sphinx site... this page you give you a pretty good idea of what you need to so: Training Acoustic Model For CMUSphinx
--- (Edited on 4/27/2012 10:54 am [GMT-0400] by kmaclean) ---
Thanks for your response. i followed your advice and followed several tutorials including some other explorations.
My gut feeling is that i am missing some simple instructions (maybe overreading them) when i apply any step in the tutorial to the dutch accoustic model and test it.
I am trying to use pocketsphinx with the dutch accoustic model.
Is there anything missing from the dutch acoustic model for usage with pocket sphinx?
My main objective or goal would be to simply record some audio at the correct 16hz, 16bit and get at least some words recognised.
Something that works well with the default model installed but fails when using the dutch model.
i am pretty sure i am doing something wrong and would love some pointers from someone who actually used the model with pocket sphinx. i.e. a simple command and test wav file will give me the confidence to proceed.
More than happy to submit to tutorials , record audio or do whatever but if i can't even get the simplest thing to work i am lost.
A bit desperate currently after the effort i put in and as mentioned before this is likely my bad but i really want this to work.
--- (Edited on 5/6/2012 4:45 pm [GMT-0500] by pivposie) ---
> who actually used the model with pocket sphinx. i.e. a simple command and test wav file will give me the confidence to proceed.
Inside the voxforge-nl-0.1.tar.gz there is a script test/test.sh which shows how to run pocketsphinx. What problem do you have using it?
To recognize big file you can also run pocketsphinx_continuous -infile file.wav with the same arguments as described in documentation. That should also work.
--- (Edited on 5/7/2012 15:00 [GMT+0400] by nsh) ---
Thanks nsh, there seems to be a mismatch in the files i downloaded earlier there was no test.sh file. The tests run fine.
I will now start building an LM file.
Is it correct that i cannot use the webservice as the webservice produces english based output only?
thanks again.
--- (Edited on 5/8/2012 2:52 am [GMT-0500] by Visitor) ---
Excellent, i am up and running now.
It took some serious time and there is loads of things to read up on and explore. i will share my queste as soon as I get some tests sorted but at least i can now run them.
thank you for your help.
Ps. my first exporations where using the archive as can be found here.
www.repository.voxforge1.org/downloads/Dutch/Trunk/AcousticModels/AcousticModels.tgz
which lacked the tests.
better get the model (for dutch) from here.
http://heanet.dl.sourceforge.net/project/cmusphinx/Acoustic%20and%20Language%20Models/Dutch%20Voxforge/voxforge-nl-0.1.tar.gz
not sure what went wrong there but the tutorials + tests inside the acoustic model will get you there.
--- (Edited on 5/8/2012 8:17 am [GMT-0500] by pivposie) ---
Hello, did you get any further with your Dutch language explorations ? Do you know of a language model (lm , DMP) ?
I'd like to have a dutch language model. Any help would be appreciated.
Regards,
Harm
--- (Edited on 11/14/2014 11:06 am [GMT-0600] by ) ---
reply from pivposie sent to voxforge email address:
Hi , i am sorry but i did not.
--- (Edited on 11/16/2014 2:59 pm [GMT-0500] by kmaclean) ---