> I've been tweaking visual speech recognition models (VSR) for the past few weeks so that I can "talk" to my agents at the office without pissing everyone off.

Wouldn't SilentWhisper do just as good a job?