How Believable Are Real Faces? Towards a Perceptual Basis for Conversational Animation