I would imagine it does the analysis locally and sends some sort of "room signature" to the server, rather than a complete audio stream, which would be an awful

I would imagine it does the analysis locally and sends some sort of "room signature" to the server, rather than a complete audio stream, which would be an awful waste of bandwidth.