Professor Hideyuki Sawada from Kagawa University in Japan was at Robotech 2011 showing off that incredibly bizarre robot mouth of his. It's based as closely as possible on a human mouth, complete with an air pump for lungs, eight fake vocal cords, a silicon tongue, and even a nasal resonance cavity that opens and closes. Like other robot mouths, it uses a microphone to listen to itself speak (or whatever you want to call it) and analyze what it hears to try to figure out how to be more understandable and less, you know, borderline nightmarish.
I know, there wasn't a demo in that vid. But I've got one right here for you, of this robot attempting to sing a Japanese children's song called "Kagome Kagome." You can hear what it's supposed to sound like over on Wikipedia before or after you listen to the robot have a go, but either way, you're not gonna recognize much. The action starts at about 30 seconds in:
Wonderful. Don't get me wrong, on principle this is some undeniably fascinating stuff. I have to wonder, though, whether the effort it would take to get this thing into a humanoid robot would really pay off relative to a voice synthesis system based on software and speakers. I guess there might be other advantages to a bionic mouth, but I'll leave the speculation up to you.