Silicon Waves

An AI radio host for a cyberpunk-style internet radio station. It watches your AzuraCast stream, generates short lead-ins with OpenAI, turns them into speech with ElevenLabs, then uploads the audio and pushes it to the front of the queue so it plays right after the current track.

How it works

1. Polls AzuraCast every 15 seconds for now-playing and time remaining.
2. When 40–70 seconds are left on the current track, it:
   • Asks GPT (gpt-4o-mini) for a short, cynical, atmospheric host line between the last track and the next one.
   • Synthesizes that text with ElevenLabs (multilingual v2).
   • Uploads the MP3 to AzuraCast into the ai_voiceovers folder.
   • Requests that file to be played next (front of queue), then deletes the local file.
3. Waits ~70 seconds after a successful run to avoid duplicate segments.

Requirements

• Python 3.10+
• AzuraCast instance with API access
• OpenAI API key
• ElevenLabs API key and a voice ID

Setup

1. Clone the repo and create a virtual environment:

   python -m venv .venv
   source .venv/bin/activate   # Windows: .venv\Scripts\activate
   pip install -r requirements.txt

2. Create a .env in the project root with your keys (example below). Do not commit .env — it’s in .gitignore.

Environment variables

Variable	Description
AZURA_API_KEY	AzuraCast API key (from Profile → API Keys)
STATION_ID	AzuraCast station ID (default: 1)
BASE_URL	AzuraCast API base URL, e.g.
OPENAI_API_KEY	OpenAI API key
ELEVENLABS_API_KEY	ElevenLabs API key
VOICE_ID	ElevenLabs voice ID (e.g. from Voice Library)

Example .env:

AZURA_API_KEY=your_azuracast_api_key
STATION_ID=1
BASE_URL=https://your-azuracast.example/api

OPENAI_API_KEY=sk-...
ELEVENLABS_API_KEY=...
VOICE_ID=pNInz6obpgDQGcFmaJgB

Run

python main.py

Leave it running; it will keep polling and generating host segments when the time window (40–70 seconds left) is hit.

Tuning

• Segment window: In main.py, the condition 40 < rem < 70 controls when a segment is generated. Adjust to match your track lengths and encoding delay.
• Poll interval: time.sleep(15) is how often it checks now-playing; time.sleep(70) is the cooldown after a successful segment.
• GPT style: Edit the prompt in generate_script() to change the host personality and tone.
• Voice: Change VOICE_ID in .env to use another ElevenLabs voice.

License

MIT