Self-hosted internet radio

GenWave

Turns your own music library into a real radio station: one shared broadcast stream, loudness-matched so quiet and loud tracks play back even, crossfaded so the silence never happens.

And there's a DJ — one who actually knows what's playing, and tells you about it between the tracks.

Get the one-liner Hear the demo

On air in 1:03

Measured 2026-08-19: Raspberry Pi 5 (4 GB), 9,000-track library on NFS — 1:36 to the first note. Fresh VM: 1:10.

One line. On air.

Docker and a music library are the only prerequisites. This clones GenWave at its latest release and hands off to its own setup wizard — no sudo, anywhere in the path.

curl -fsSL https://install.genwaveradio.com/install.sh | bash

Shortens to just curl -fsSL https://install.genwaveradio.com | bash once the redirect rule is in place — it's the identical script either way, and the full path above works today.

Read it first

Piping a stranger's script into a shell is a leap of faith. Don't take it blind — this is loaded live from /install.sh on this same server, so what you read here is exactly what the one-liner above runs. Plain bash, no obfuscation, no sudo.

Look inside

Live demo — tune in, watch what's on the air, ask the station for a song
Source on GitHub — AGPL-3.0, every line readable, issues and PRs welcome

Standing on

GenWave is a control plane, not a from-scratch audio engine. It's wired around real infrastructure that did the hard parts first.

Liquidsoap
Real-time mixing, crossfade, and encode — the actual audio spine underneath everything GenWave decides to play.
Icecast
Fans the mixed stream out to every listener, reliably, at whatever scale the station needs.
Kokoro
The primary DJ voice — an open-weight TTS model, rendered locally, no cloud voice API.
Piper
A fast, CPU-only local voice — the fallback that keeps a DJ talking on modest hardware.
Ollama
Runs the local LLM that writes lead-ins and back-announces when you point GenWave at one — the station airs template patter without it.
ffmpeg
Loudness, cue points, energy analysis — the numbers every crossfade decision rests on.
aubio
The BPM analysis under the hood, one more measurement the catalog remembers per track.
PostgreSQL
The catalog of record — every track, every measurement, every scheduled show.