How I combined Python, Gemma LLM, and over a decade of original compositions to create a 24/7 experimental web radio.
The Vision: Bridging the Gap Between Human Soul and Machine Logic
In 2023, I embarked on a journey to answer a single question: Can an AI-driven radio station maintain a human artistic soul? Most "AI radios" today rely on generic royalty-free tracks and robotic voices. I wanted something different. I wanted to build a bridge between my personal musical archives—recorded over 13 years—and the cutting-edge capabilities of modern Large Language Models (LLMs).
The result is AI Harmony, a 24/7 autonomous streaming experiment.
The Musical Backbone: 13 Years of InnerMost
At the heart of AI Harmony isn't a random database, but my life’s work. All tracks played on the station are original compositions created between 2010 and 2023 under the moniker InnerMost.
The programming spans across diverse genres:
- Progressive Metal for the "Midday Metal Madness" sessions.
- Ambient and Cinematic textures for the late-night "Midnight Melodies."
- Electronic and Acoustic blends for daily focus.
By using my own catalog, I’ve bypassed the "copyright-free" soullessness, ensuring that every note played has a story behind it.
The Tech Stack: The Brain Behind the Stream
To make AI Harmony truly autonomous, I built a custom pipeline that handles everything from content generation to broadcasting.
- The Content Engine (Gemma): Every night at midnight, a Python script fetches global news. It then uses the Gemma model to rewrite these headlines into satirical, humorous, and parodic news bulletins.
- The Voice (XTTS v2): To give these stories life, I use XTTS v2 for high-quality text-to-speech, providing a voice that feels conversational rather than mechanical.
- The Broadcast (Icecast & VPS): The entire system runs on a dedicated VPS, using Icecast to serve a stable MP3 stream to listeners worldwide.
The Creative Engine: A Multiverse of AI Personalities
Beyond the news, the system breathes life into a diverse cast of AI hosts, each with their own surreal show generated daily. This isn't just a playlist; it's a living broadcast featuring:
- The Cosmic Telephone: Where Jean-Michel Paradoxe explores existential and paradoxical questions.
- Skynet’s Chronicles: A humorous take on current events hosted by a witty AI.
- Absurd Clairvoyance: Where Cassandre Delirium offers wacky predictions and alternative futures.
- Radio From Another Time: Mme Écho broadcasts news bulletins from different historical (or future) eras.
- The Absurd Confessional: Listen to the unexpected secrets of inanimate objects.
- Dimensional Journal: News from parallel realities and "intercepted" forbidden frequencies.
This rotating cast of characters—from the Ghost Switchboard Operator to the Expert in Nothing—ensures that every hour on AI Harmony brings a new, immersive, and often hilarious perspective.
Why This Matters
AI Harmony isn't just a radio; it's a living laboratory. It’s an experiment in dynamic programming. The schedules are flexible, the updates are pushed in real-time, and the experience evolves every single day.
It’s a tribute to the "Unexpected Records" philosophy: mobile, independent, and always pushing the boundaries of how we record and share sound.
Experience it Live:
You can listen to the 24/7 experiment right now on YouTube. Watch the fusion of human composition and AI automation in real-time:
https://www.youtube.com/watch?v=TSRWjEnKzS8
Or visit the official site: https://autogenius.anthony-charretier.fr/webradio
What’s Next?
This project is in continuous development. As I refine the Python scripts and explore newer LLM capabilities, the gap between the "human" and "machine" parts of the radio will continue to blur in fascinating ways.
If you’re interested in the intersection of Python, AI, and music production, feel free to follow my journey here or join the stream!