Thanks guys!
This runs fine on my rather modest laptop (1.4 GHZ, 480mb RAM, no 3D card), although the demo is using only 500 polygons (more polygons = more detail, but also requires more processing power) and no textures.
In theory you can do graphics that are near Pixar-quality this way, but that requires a workstation that costs between $10,000-20,000 and Blender's real-time graphics have some serious limitations right now, but its developers are in process of addressing that.
This
Bad Language Mario video has a good example of a digital puppet "talking head". The Disney theme parks have an attraction called
Turtle Talk With Crush where Crush the Sea Turtle from Finding Nemo appears in an "aquarium" - actually a big video screen - and interacts with the audience. Both of these use the same basic principles I am working with.
What's really cool (I think anyway) about the approach I am trying is that because it's tied in to Blender you can use all of Blender's animation tools too, which are really powerful. You can perform in real-time, then edit your performance and correct any mistakes with Blender before rendering the final work out like conventional 3D animation that's very high quality.
I'm about halfway through setting up a bunch more features on the head, I'll make up another video and post it as soon as I am done.