I keep a fairly large MP3 collection. Music was an important part of how I grew up; my earliest memories involve a radio playing in the background.
I have been doing some archivist work recently, generating playlists that are deeply personal to me. Steven Wright had an old joke:
My socks don't match, but to me they're the same, because I go by thickness.
Music is kind of like that for me. I build playlists around the feels. So on my way home last night, I thought about the parts of the radio experience that I missed, and thus was Radio AI born. It plays shuffled playlists with a few.. additions.
Mimika Studio gives me voice cloning, so I pulled four voices and I feed them generated text passed through qwen3:8b like this:
| Voice | Purpose |
|---|---|
| Al Roker | Weather updates fetched from https://api.open-meteo.com |
| Tom Brokaw | Tom reads me the news from two sources: an MCP server that gives me server statistics, who's watching what, download stats, from my Media Server, as well as actual news from https://newsapi.org |
| Car Salesman | Extracted from Samson Auto ads that I grew up with, Car Salesman reads items from my ToDo list in Obsidan via that MCP server, and pitches them at me like it's trying to sell me a car. |
| Wolfman Jack | The DJ, telling me what songs are about to play, or played. I can't get his iconic howl out of Mimika, but the kinds of things he said and general voice works extremely well. I might trade him out for Kasey Kasim, though. |
.. and that's app. I wish I could host this somewhere and stream it, but alas, my laptop is the only machine I have with the horsepower I need to get all this running (which is less work on the machine pound for pound than I anticipated), so it goes where I go.