I repurposed an old Google Voice Kit to use @openai Realtime API with . Vibecoding failed, so I did some old school research+trial&error.

My main learnings:
· Pass all the same WSS messages via Data Channel
· Ephemeral tokens not always needed
· API is a lot like WHIP (no SDK needed)
· Best practices for managing gUM and RTCPeerConnection still apply (not in docs)
· Use speech events for voice functions

Full code walkthrough on webrtcHacks here: webrtchacks.com/the-unofficial

0

If you have a fediverse account, you can quote this note from your own instance. Search https://fosstodon.org/users/chadwallacehart/statuses/114186025413236512 on your instance and quote it. (Note that quoting is not supported in Mastodon.)