If the quality is too low, though, the benefits to communication from seeing your partner's facial features synchronized with their voice disappear, and people won't bother using it.
If software identified the user's face in live video and transmitted just that at high resolution, discarding or compressing the rest, how much efficiency could be gained?
