Question for the livekit team: right now there's no voice available from realtime model that could also be used in a regular agent audio pipeline, correct? As in, the same exact voice available via a realtime provider and via a tts plugin? I was thinking ultravox could be used, but then realized they're not on the TTS plugin list.

It also has to be a custom voice, not a predefined voice.

You can use a separate TTS with the realtime model as long as the model supports text modality, e.g. here's the docs for doing that with OpenAI realtime: https://docs.livekit.io/agents/models/realtime/plugins/openai/#separate-tts, I think that would be the way to go since you can't define a custom voice for realtime models.