I guess the solution will be similar to what we have in protocols like IRC: you can create a query with someone, you can always open a room with two people, but if you want to talk with N people at the same time, then you have two options in decentralized networks: you can connect everyone with everyone by making N*(N-1) connections or you can create some "server" node connected with everyone and change that "server" at any time if needed. In IRC the latter is used and I expect it will be similar here, just voice will be transmitted instead of text. Also, as far as I know IRC can transfer voice, but last time I tried it was hard to set up everything.