VoxForge
Hi!
I gernated an acoustic model (HTK compatible) with the current German voxforge speech corpus and ralfherzogs GPL dictionary.
You can download it here (and use it with simon as static base model if you want to):
http://hotfile.com/dl/38763548/707f8a0/voxforge_de_simon-2010-04-20.tar.bz2.html
The model was generated completely automatic with simon and altough I didn't do any manual optimization, I noticed that the model optimizer of simon took quite long, removed a couple of recordings and a few words from the dictionary.
I tried it with a small vocabulary and altough the recognition rate is obviously not perfect it seems to work alright.
Maybe you could make it available as a download?
Greetings,
Peter
>Maybe you could make it available as a download?
will do!
thanks,
Ken
Hi!
I improved the phoneme set a little bit.
Most obvious change: Introduced the "gls" phoneme for a glotal stop ("?" in SAMPA).
Download: http://hotfile.com/dl/39038161/e23b3f0/voxforge_de_simon-2010-04-21.tar.bz2.html
I also designed the first scenario using the model:
http://kde-files.org/content/show.php/%5BDE%2BVF%5D+Firefox?content=123527
Saldy I needed to work around some words whose phonemes were not covered ("Lesezeichen" and "Ok" most notably) so the vocabulary is not perfect. More scenarios are planned :)
Greetings,
Peter
Hi Peter,
Thanks again,
Here is the link:
voxforge_de_simon-2010-04-21.tar.bz2 21-Apr-2010 10:11 10.1M
Ken
Thanks but I'm afraid that this'll be outdated (again) in a couple of hours.
Found another source of improvement and am compiling a new model for the past couple of hours.
If you try it: Which version are you going to use? There is the first alpha (Sourceforge) which should be quite stable despite a couple of issues. If you want to live on the edge you could also check out the current git version which is a bit more experimental but does include cool new features like a first run wizard to guide you through the initial setup. Your choice.
Sadly I have to wait for Qt 4.6.3 for a new release because QtMultimedia in 4.6.2 still has some blocker bugs on windows...
Greetings,
Peter
____________
New version:
http://hotfile.com/dl/39109642/c3046f9/voxforge_de_simon-2010-04-21_2.tar.bz2.html
I also added another scenario (window management):
http://kde-files.org/index.php?xcontentmode=692
>Thanks but I'm afraid that this'll be outdated (again) in a couple of hours.
no worries. I've updated the links to:
voxforge_de_simon-2010-04-21_2.tar.bz2 21-Apr-2010 13:39 10.1M
Thanks!
>Which version are you going to use? There is the first alpha (Sourceforge)
>which should be quite stable despite a couple of issues.
I will take a look at this one for now... after I get through the upgade to FC12, which may or may not be clean, given all the junk on my dev box.
Ken
Ok, I am at five scenarios and done for now.
They are all created to work with the german voxforge model I built and will work together so you can load all at once without clashing commandonames / grammars.
They all work but the recognition is obviously not perfect. All scenarios also include trainingstexts to train the model but using the static voxforge model is a cool place to start off and see the system in action before diving into the training.
The scenarios cover:
Controlling mouse
Controlling keyboard (pressing individual buttons)
Controlling Firefox
Controlling Amarok
Controlling Windows (specific functions for KWin but the basics should work everywhere including Windows)
Controlling XBMC (XBOX Mediacenter)
To set up a simon system with those scenarios and the voxforge model follow those steps:
1. Settings > Configure simon > Model settings:
Select "Static model" and set the files (HMM definitions, tiedlist, etc. to the respective files in the downloaded voxforge model tarball)
2. Get the scenarios you want:
Manage scenarios > Import > Download
3. Enjoy
Greetings,
Peter
I trained the model a bit (1170 samples targeted toward the scenario vocabulary) and re-built it.
Download: http://hotfile.com/dl/39406523/f8ba664/voxforge_de_simon-2010-04-23.tar.bz2.html
Btw, you need the current simon version (at least 0.2.93) for the current scenarios. The last released alpha version is only 0.2.91.
@ken: If you don't want to compile your own simon version to try it, I'll release a new version once Qt 4.6.3 is released (we switched to QtMultimedia and I need for these bugfixes: http://bugreports.qt.nokia.com/browse/QTBUG-9100 and http://bugreports.qt.nokia.com/browse/QTBUG-9766).
Greetings,
Peter
@Ken: Would it be possible for you to update the links to the latest version:
I uploaded it again:
http://hotfile.com/dl/41925215/32be652/voxforge_de_simon-2010-04-23.tar.bz2.html
Greetings,
Peter
>Would it be possible for you to update the links to the latest version:
Done... see:
http://www.repository.voxforge1.org/downloads/de/Tags/AcousticModels/