I've done a fair amount of audio engineering, and my perception today of Meet was that it wasn't the simple threshold / noise gate that I would expect. For instance, I could see her moving her lips and putting down dishes, which I'm sure were noisy, but until she spoke loudly enough for a recognizable voice to overcome the dish sounds, she was essentially muted by the app. I think it was being gated based on vocal range or actual identifiable speech, not just volume. That was what seemed new to me.