I'm actually building this exact thing right now. Using Gemini TTS with its multi-speaker capabilities. Soundtrack comes from a Suno Playlist which the tool will download the entire playlist at the start.

Plays 3-5 songs prior to a "break", where the logic will call out channel donators vs pick a random topic to discuss, then use an LLM to generate the script before sending to gemini.

In the end; its really just a wrapper around FFMPEG broadcasting to an SRT or RTMP stream. I can upload the latest branch when I get home to share