Damn it - Fujitsu Laboratories have implemented this tool before me... :-/
I was planning to do a tool like this for a MS thesis back in late 2007... :D
The first application would have been for marking material from books and papers without tedious retyping or physically marking the paper, and then quick citation.
However I was to work without the projector - I couldn't have one to play.
It was for another bigger project I've mentioned here since long years - the Research Accelerator, and it is possible that I make it to its first users release this, or at worst - next year, after completing a few other milestone projects. (Google's and Kurzweil's project are related competitive projects, some other NLP-ML projects which are already commercial - too...)
Obviously Solvable Engineering Problem
I was suggested by my supervisor, that it's too "engineering" a problem, it should have been more "scientific".
I agreed - I'd call it an "obviously solvable" problem, it just has to be implemented.
At the time I've been quite busy with the ASIC engineering work, and for the MS thesis I designed an improved version of my freshman Bulgarian TTS "Glas 2004" and proposed a number of improvements for consecutive generations of synthesis. I thought that it would be easier to go on that path.
It wasn't implemented at the time though, only very few new experiments, I didn't have time. And actually now I don't know whether the gesture selection wouldn't have been easier. :)
Anyway, as you know, if you read my blog, that extended text-to-speech synthesizer is almost released now - in fact yet it's a subset of the proposed functions.
I have some work going on related to this tool, too, as I'm building a general infrastructure for both the Research Accelerator and for the Ultimate Goal - the thinking machine.
Stay tuned, I will unveil probably when I have a complete application that can be offered to end users. :)