Speech perception involves two senses, hearing and seeing: we hear and lip-read speech. The synchrony between acoustic and visual speech is a driving force for the development of facial animators, which predict facial motion given an acoustic speech input. A computationally and conceptually simple model of audio-visual speech production is examined, one that relates acoustic speech and visual speech through a linear transformation. The model is tested with a large group of generic sentences, as well as sentences composed of highly controlled contexts to elicit specific types of articulator motion. On average, the transformation can account for 70% of facial motion, but this result is influenced by the context of the sentence. Results are interpreted in terms of the model assumptions in approximating the speech production system. This research is relevant to animating faces for multimedia or rehabilitation purposes, and falls in the framework of current speech perception theories.