I've played around with microphone arrays with a Kinect and
http://hark.jp and it's amazing what you can do even with 4 microphones.
-----------------
I can figure out where people are talking spatially.
I can figure out how people are moving around when speaking.
I can voiceprint people.
I can cancel voices individually, or mute all noise past a certain distance from the center of the mic array.
I can 3d map rooms by recording over time how people talk and how they move.
I can connect multiple nodes together to create wider areas of coverage of above.
I can convert all sounds to text if appropriate. I can send tiny messages that include voiceprint, position, and content.
I can tell how many people are in your house at any one time.
I can tell if you have friends over, and when. If they have similar system, can collaborate backend data across all of above
------------
So yeah, there's plenty of things people aren't thinking of. I doubt they even have the experience to know how possible it is with open source tooling and cheap hardware. But all of these things are possible, and my guess is being done.
The distinction with phones is hard. Apple has stayed within the "we support privacy" aspect. And with things like limeSDR being cheap and common, wouldn't be as hard to attack and prove. Android is tricky, but with some of the anti-crapware systems and basebuild efforts going on, can be considered possibly trustworthy (if edited).
And in general, phones move a lot. Amazon tunacan/tomatocan doesn't. Just add a $1 accelerometer, and detection of movement is trivial. Phones can have this, but doing the above would be a magnitude harder.