Every time these come up, I just can't help but imagine the first time someone sneezes in one. It's not going to be pretty. I do agree though that they'd be useful, assuming that it is a model which form-fits and has two way filtration,
my ability to understand and parse speech efficiently has gone way down recently due to the exact problem these solve. However, head position will also matter for some of these models, it almost seems like you have to be looking straight at the person in order
to see through the window, and I expect they'll have a negative effect on acoustics beyond a regular mask as well.