This is exactly right. Even though the quality is good enough to fool people in general, they've simply added a new set of measurements to the model, because th

This is exactly right. Even though the quality is good enough to fool people in general, they've simply added a new set of measurements to the model, because the mouth and throat do not conform to static dimensions when speaking. Fluid dynamics isn't a black box mystery, and already used in video game graphics, so this will get incorporated rather quickly. Probably in the very next iterations, since this article spells out pretty clearly what needs to be modelled next to make the voice sound / appear even more authentic under analysis.