It's true that nostr as a protocol doesn't seem to have any real capacity for voice, but given a Discord-like community would probably "live" on a fixed relay, that server could also very easily provide something like a TURN server like Matrix clients use for voice and I think video support. The client could integrate support for it, and the typical clueless user wouldn't see the difference. For the more ephemeral nature of most voice communications, there's no real need to publish voice chat through Nostr events. It could be done, sort of, for any talks that need to be archived, but it's not a requirement for the vast majority of the voice chat happening on Discord anyway.