An incremental dialogue system concurrently listens and speaks. That's how it's supposed to be, but there is one problem: on the most basic level, the ASR does not know whether the TTS is speaking and hence will "tune in" to its own words and listen to itself. What's worse, a polite incremental dialogue system will stop speaking when interrupted (or speak up to keep the turn). This may either lead to the system feeling interrupted all the time (by itself), or speaking louder and louder in a positive feedback loop.... read more
After a very successful tutorial last week in Bielefeld, I've finally decided to push InproTK to it's 1.0 release!
Also, I have moved our sourcecode management to Git. I'm thankful to Sourceforge for hosting us for so long, but our sourcecode is moving on to Bitbucket, thanks to their free academic license. Git will hopefully enable us to publish our intermediate, closed versions of InproTK more frequently and to encourage all the young folks to submit patches as well.
There will be an Interspeech Tutorial on incremental spoken dialogue processing, held by Timo Baumann and David Schlangen, that will be largely based on InproTK. Come visit us to learn more and to start using InproTK!
Slides are available at the tutorial website: http://www.dsg-bielefeld.de/is13-tutorial
A student project on spoken dialogue systems at the University of Hamburg has used InproTK as part of a hybrid dialogue system.
The system combines a standard, non-incremental system (based on DialogOS) with an incremental mode (built with InproTK, of course) for positioning puzzle pieces. it turns out that positioning is much easier and more flexible in the incremental mode. Also, it's quite amazing how they integrated the two systems (different ASRs, different DM, different everything) into one application. Well done!... read more
I've finally been able to import the InproTK code from our restricted SVN to Sourceforge. This was difficult because the code was living along with documents, code, and other stuff not meant for public circulation (especially not including all the revision information).
It turns out that selectively exporting from a subversion repository is difficult as soon as you have started to rename things. To be precise, you'll have to implement it yourself, or use my implementation svncleaner.pl (which I'll upload shortly). Because I wanted to keep revision numbers identical across the Sourceforge and the proprietary repositories, the import contains many empty revisions. Sorry for that, but it will help us to more easily track bugs in the long run.... read more
Tomorrow, David will be presenting our Poster which describes the upcoming release (also due tomorrow).
Right now I'm still wrestling with GIT to get our code in shape. I don't want to upload it without the full history, but I'm forced to make sure that there aren't even traces of some files which we can't/don't want to release. Anyone who's very intimate with GIT who's interested in helping?
Our (David Schlangen's and Timo Baumann's) system demonstration of the incremental speech synthesis component that is part of InproTK was accepted for demonstration at ACL. We'll really have to get the code in shape before that can happen. The code for all demonstrations is in the inprotk code repositories, in src/demo.