[nitrogen]

If I remember right there was an app from the early 2000s that did something kind of like this. It would be cool to combine the 2D map with a speech-to-text word cloud over clusters of people so you could see if you wanted to join another conversation.

[hobs]

Mumble has positional audio for game stuff - https://wiki.mumble.info/wiki/Positional-Audio its also extremely low resource usage and you can install/manage it yourself.

I really prefer it to discord/whatever other voice client because you can also do stuff like hierarchical broadcasting, a parent channel can broadcast to children without children broadcasting back.