Acephale Radio
4 channel AI powered radio-- talk show, conspiracy monologue, being able to call in and make requests, and generate music with lyria
Core trick
Persistent AI host agents script live segments that TTS voices, with Lyria generating ambient music between them on a continuous stream.
Description
4 channel AI powered radio-- talk show, conspiracy monologue, being able to call in and make requests, and generate music with lyria
README excerpt
# Acephale Radio Autonomous multi-channel AI radio station. Persistent AI agents generate live talk shows, DJ sets, conspiracy monologues, and generative ambient music around the clock. Listeners tune in with a rotary dial, hear static between stations, and can call in live. **[Listen live](https://erosika.github.io/acephale/)** ## Stations | Station | Mount | Host | Description | |---------|-------|------|-------------| | The Morning Zoo | `/zoo` | Buzz + Val | Two co-hosts who bicker about everything. Buzz is obnoxiously enthusiastic; Val is contemptuous and deadpan. | | Crate Digger | `/crate` | Crate | Obsessive music nerd pulling tracks from Radiooooo. Lectures lovingly about context, labels, and session musicians. | | Conspiracy Hour | `/conspiracy` | Nyx | Late-night paranoid monologues over dark Lyria ambient. Builds cross-session conspiracy threads, accuses other DJs. | | Request Line | `/request` | Echo | The people's DJ. Takes live voice calls, remembers every caller, plays requests via Radiooooo. | ## Stack - **Runtime**: Bun + TypeScript - **LLM**: Gemini 2.5 Pro (scripts, monologues) + Gemini 2.0 Flash (steering, transcription) - **TTS**: Google Cloud TTS (Chirp 3 HD voices: Aoede, Leda, Puck, Charon, Kore) - **Music Generation**: Lyria RealTime (48kHz stereo PCM via WebSocket) - **Music Curation**: Radiooooo API (decade/country/mood discovery) - **Memory**: Honcho SDK v2 (persistent per-station sessions, per-agent peers) - **Streaming**: Liquidsoap reques