Analysis updated 2026-05-18
Run a fully automated, always-on AI radio station with rotating AI hosted shows.
Split show writing from performance across two AI tiers to keep a 24/7 content stream cheap.
Broadcast a simulated sports league with live play by play commentary that stays consistent with the game state.
| 42kyynfqjv-dot/deepseekradio | 13127905/deep-learning-based-air-gesture-text-recognition- | 6xvl/paralives-plugins-index | |
|---|---|---|---|
| Stars | 15 | 15 | 15 |
| Language | Python | Python | Python |
| Setup difficulty | hard | moderate | easy |
| Complexity | 4/5 | 3/5 | 2/5 |
| Audience | developer | developer | general |
Figures from each repo's GitHub metadata at analysis time.
Requires an OpenRouter API key, a self-hosted Kokoro TTS setup, and Icecast/Caddy for streaming.
The Frequency is a fully automated AI radio station that runs around the clock without any human at the controls. A rotating cast of AI hosted characters run different shows throughout the day, take fake caller phone calls, bring on invented guests, and riff on pre written comedy bits, and the whole thing is generated continuously and streamed live as audio. The setting is a fictional 51st US state called Wending, invented by the project so the comedy stays purely absurd rather than touching real people or events, with its own town, local government, and sports league all made up. The show schedule mirrors how real radio stations work: the same time slot has the same tone and format every day, morning drive, an afternoon culture hour, a late night show, and an overnight call in program, but the specific cast and jokes rotate so it never repeats itself exactly. Under the hood the system splits writing from performing to keep costs low while staying entertaining. A head writer AI runs occasionally to sketch out each show's outline, its jokes, its guest, and callbacks to earlier storylines, while a separate performer AI runs constantly, turning that outline into spoken in character dialogue and inventing callers and guests as needed. Both roles currently run on the same cheap language model, but the code is built so either one can be swapped independently. Text is turned into speech using a self hosted, free voice synthesis tool called Kokoro running on ordinary CPU, with custom male voices created separately since Kokoro did not ship any strong male voices by default. A persistent lore file keeps running jokes, feuds, and multi day storylines consistent across shows and days. The station also broadcasts a twice weekly hockey show for a fully simulated sports league, with a season engine tracking standings and scores, and a safeguard system that keeps the commentary from contradicting what actually happened in the simulated game. The whole setup costs roughly twenty to twenty five dollars a month in AI usage plus around twenty euros a month for the server it runs on, and has been streaming live in production since July 2026.
A fully automated, 24/7 AI radio station with AI hosts, fake callers, invented guests, and a simulated sports league, streaming live.
Mainly Python. The stack also includes Python, OpenRouter, Kokoro TTS.
Setup difficulty is rated hard, with roughly 1h+ to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.