:-) A (very) old project I've done in the past was actually "augmenting" a DIY theremin using it as a controller: http://www.youtube.com/watch?v=c5nOK44Z02c

In this system the user can define a gesture by performing it just once. So users can easily define their gestures for every surface in few seconds. The next step would be to automatically recognise that the surface has changed so to avoid to re-train the system for every surface. Thanks for the interest!

Yes you can train the system with your own gestures :-)

The code is actually standard c++, hope to be able to share it soooon....

In the video we see that different gestures are mapped to different sounds. For example the coin enables a certain abstract sound whist the bare fingers enables the bass sound and the nails enables the pad... this is done thanks to gesture recognition techniques (a modified version of Hidden Markov Models to work in realtime). Apologies whether it is not clear enough in the video though...