Something on the screen has gone wrong. He types harder, and harder, and then he stops and tells you about it. Drop in a voice clip and he says yours — the mouth follows the recording. Add a tune if you want one under it. Nothing leaves your device.
Every keystroke is synthesised on the spot — the click of the switch, the thock of the
cap bottoming out, the clack coming back up — and one seeded schedule fires the sound and dips the hand
at the same instant, so they cannot drift apart. The desk and the man are generated in Blender by
rage-src/build.py; there is no .blend anywhere.