General Discussion

Nested
FreeCLAS - "Free Commons of Linguistically Annotated Speech".
User: kmaclean
Date: 9/4/2008 12:33 pm
Views: 7367
Rating: 9

From the a comp.speech.research post:

FreeCLAS (http://www.ihear.com/FreeCLAS) is a new project to build a a
data base of high-quality speech data. "High quality" means annotated
data that have been validated by humans. Building such a data base has
been expensive because it requires substantial investment of people's
attention. As a result, high-quality speech data is not generally
available.

FreeCLAS uses a wiki. This is a call for people to join the wiki to
build it. Embedded in the wiki is a tool, shva, which opens from your
browser to let you hear, view and annotate any utterance in FreeCLAS.
At this point, there is a seed data base of a small collection of
utterances annotated in en-US and IPA.

shva and other related software downloadable from FreeCLAS are all
Free Software, licensed under GPL or other compatible licenses. The
speech data is under the Creative Commons attribute-share-alike
license.

Their focus seems to be more collecting linguistic annotations of speech by getting users to provide/validate time stamps of utterances.  This is a little different what VoxForge is doing.  We are basically trying to collect speech prompts (15-20 words long), with little regard for accurate timings - since the HTK/SPhinx acoustic model training process can do this automatically (with short utterances)

What is really interesting (from VoxForge standpoint at least) is their ALingA (GPLv3) annotation Java applet.  I can't get the app the run on my PC (I have a 64-bit machine, which they don't provide support for...yet).  However, from the screen shots, it looks very impressive for a Java applet.  They use the JavaFX libraries, which is Sun's answer to creating rich Internet applications (RIAs)... i.e. Sun approach to creating a Flash-like environment.  It might be a useful starting point for a speech submission annotation validator for VoxForge (but just to allow other users to validate that an utterance matches the prompt line).

Ken

--- (Edited on 9/4/2008 1:33 pm [GMT-0400] by kmaclean) ---

Re: FreeCLAS - "Free Commons of Linguistically Annotated Speech".
User: Visitor
Date: 9/4/2008 4:21 pm
Views: 298
Rating: 3
Cool!

--- (Edited on 9/4/2008 4:21 pm [GMT-0500] by Visitor) ---

Re: FreeCLAS - "Free Commons of Linguistically Annotated Speech".
User: TEReisler
Date: 6/11/2009 8:24 pm
Views: 69
Rating: 3

Hi, greetings from a sister project Smile.

FreeCLAS has just been updated with new releases of all the software and the small database.

The web start Shva will now let you view 2D derived signals such as spectrograms. It now runs on 64bit architecutres, but you will need a late version of Sun's jre. If you have problems building the native libraries, drop a note on the wiki.

ALingA is the Ogg-contained multi-track format that interleaves the annotations with the original acoustic signal, as well as derived signals. The annotation format, http://www.ihear.com/FreeCLAS/wiki/LingA, is very general, and is capable of representing not only manual labelling, but also machine recognition results. Then you can use Shva to hear and view them.

I invite you all to visit and try out the tools, and see whether they can help you with your efforts.

 

--- (Edited on 6/11/2009 8:24 pm [GMT-0500] by TEReisler) ---

Re: FreeCLAS - "Free Commons of Linguistically Annotated Speech".
User: kmaclean
Date: 6/22/2009 12:33 pm
Views: 3494
Rating: 1

Hi TEReisler,

thanks for the update!

Ken

--- (Edited on 6/22/2009 1:33 pm [GMT-0400] by kmaclean) ---

PreviousNext