Matt Gilbert developed the Gestural Speech Synth, a speech synth controller which puts expression before language.

Matt writes:

Speech synthesis is usually about converting text to speech, but what if you approached it differently? With this experimental speech synth controller, I started by providing control over inflection instead of starting with discrete language. I was curious about the different expressive possibilities that would come out of this, and I also wanted to make an electronic instrument that would make for an interesting live performance.

The synth can only control vowel sounds, but it moves fluidly between vowels through different gestures, allowing for more variation in inflection and emphasis.

Simply create the shape of your throat with hand and forearm.

The Gestural Speech Synth runs on a Max patch (based on research by Brad Story dealing with vocal tract shapes and voice quality) which acts as a filter. The sound input is passed through a simulated vocal tract which is picked up from the hardware input (Max/MSP -> PIC).

You can see a video of Matt playing around with it on his website where you will also find the Max patches and PicBasic program available for download.

Link via Make:Blog