SARC-FM: My personal radio station for all things „AI“ – one that keeps broadcasting even when nobody’s in the studio anymore.
My AI radio project „SARC-FM“ has now been running for 77 days without interruption. SARC-FM is a fully autonomous AI radio station that writes, voices and broadcasts its own program. Two synthetic hosts research real AI news, discuss it in four rotating show formats, and are cut in between tracks from the music library using cloned voices.
In 77 days, this has produced 6,754 segments – around 88 a day – each one with a written transcript.
But the most remarkable part happens when the computer goes off.
After every live episode, a shadow production creates the same episode a second time in a timeless version: no greeting, no time of day, no reference to the current date. 900 of these „canned“ episodes are on standby, and the station switches over to them seamlessly the moment the PC studio really does go silent.
When a process crash took the station down for 8 hours and 4 minutes on August 20, it turned into a design question for me: the same failure now costs 8.4 seconds.
Honesty is anchored just as rigorously in the data contract: if the news source is unreachable, the system reports „not checked“ rather than „nothing happened“. The hosts simply cannot claim there’s no news when in reality nobody looked.
All of this, by the way, without a conventional streaming server and without a CDN: every listener worldwide hears the same second, because playback is anchored to the digital wall clock – static files on ordinary shared web hosting are all it takes.
The numbers behind it are still fairly modest: 363 anonymous listeners from 10 countries, 139 measured listening hours in eleven weeks – no cookies, no accounts, no stored IPs. SARC-FM isn’t a scaling story; it’s more a proof that an autonomous station ran publicly, in front of real people.
Built by one person, on one computer, in 84 days. MIT-licensed. Listen at sarc-fm.net
#AI #Radio #AutonomousRadio
