Radioo

A desktop audio broadcaster for Windows. Capture a microphone and system audio, run them through a live effects chain, and stream the mix to an Icecast server (and optionally record to disk or write HLS segments at the same time).

Built with Electron + Svelte 5. All DSP runs in the browser via the Web Audio API; ffmpeg handles encoding.

What it does

• Two sources: pick any microphone and capture system audio.
• Per-process system audio: instead of grabbing everything that plays on the default output, target specific apps. Three modes — All (everything except Radioo itself), Include (only these processes), Exclude (everything except these processes). A native WASAPI sidecar (proc-audio-capture.exe) does the capture; Radioo auto-excludes its own renderer so monitor beeps don't leak into the stream, and hides itself from the process list.
• Live effects chain on the mic: input boost, noise gate, compressor, presence EQ, megaphone/telephone filters, convolver reverb (load your own impulse response), and a brickwall limiter.
• Auto-ducking: music and system audio drop automatically while you talk, or hold Ctrl+Shift+D to duck manually.
• Music player: load a local audio file and play it into the mix. A short beep warns you ~15s before the track ends.
• Disconnect alert: a continuous 1s beep plays on your speakers whenever the stream disconnects, for any reason (manual stop, server drop, error).
• Streaming: Icecast SOURCE protocol over raw TCP — mp3, ogg/vorbis, aac, opus, or flac. Listener count polled every 10s.
• Recording: save the encoded stream to a file while broadcasting.
• HLS output: optionally write segmented HLS to a local directory.
• Multilingual UI: English, Spanish, Portuguese, French, German, Italian, Japanese.
• Accessible process picker: the "Add process" dialog is a proper listbox — ↑/↓ to navigate, Space to toggle, Enter to apply, Escape to cancel.

Requirements

• Windows 10/11 (the provided build targets win-x64).
• ffmpeg on PATH at runtime. A bundled copy is included in the unpacked build under resources/ffmpeg/.

Running

Grab the latest win-unpacked build and run Radioo.exe. No installer.

Keyboard shortcuts

Global shortcuts (marked ★) fire even when Radioo isn't focused.

Shortcut	Action
Ctrl+M	Toggle mic mute
Ctrl+Shift+M ★	Toggle system-audio capture on/off
Ctrl+D	Toggle auto-ducking
Ctrl+Shift+D ★	Momentary duck (hold)
Ctrl+S	Start/stop streaming
Ctrl+R	Start/stop recording
Ctrl+O	Open a music file
Ctrl+P	Play / pause music
Ctrl+I ★	Music volume up (+5%)
Ctrl+K ★	Music volume down (−5%)
Ctrl+Shift+R	Announce music progress (time remaining)
Ctrl+Shift+C	Announce current mic status (no device / muted / silent / live)
Ctrl+L	Announce current listener count
Ctrl+Shift+S ★	Toggle screen-reader process in the capture list (adds/removes NVDA/JAWS/Narrator)

In the Add process dialog: ↑/↓ to move, Home/End for ends, Space to toggle, Enter to apply, Esc to cancel.

Developing

pnpm install
pnpm build:sidecar   # builds native/proc-audio-capture (Rust) — needed once before pnpm dev
pnpm dev             # Vite + Electron with hot reload
pnpm build           # production renderer + main bundles
pnpm dist:win        # zipped Windows build via electron-builder
pnpm test            # vitest suite (audio worklet, IPC, store, i18n, sidecar)

Building the sidecar requires a Rust toolchain (rustup default stable). The binary lands in native/proc-audio-capture/target/release/proc-audio-capture.exe and is copied into resources/proc-audio-capture/ for packaging.

Unpacked build (no zip):

pnpm build
pnpm exec electron-builder --win --dir
# output: release/win-unpacked/Radioo.exe

Architecture

Three-process split:

• Renderer (src/) owns the Web Audio graph and UI. Mic plus one worklet per captured PID feed into per-source gains, then the ducking stage, effects chain, master bus, and a PCM capture worklet that emits raw f32 frames.
• Main (electron/) owns ffmpeg, the Icecast client, file I/O, settings, dialogs, listener polling, and the sidecar lifecycle. PCM frames from the renderer arrive over IPC and are piped into ffmpeg stdin; encoded output is fanned out to the Icecast socket and (optionally) a recording file. In 'all' mode it passes the renderer's OS PID to the sidecar as --exclude-pid so the app's own audio never reaches the capture.
• Sidecar (native/proc-audio-capture/, Rust) runs AudioClient::new_application_loopback_client and writes raw f32le stereo 48 kHz PCM to stdout with JSON status on stderr. One child per captured PID in include mode; a single child in all/exclude mode.

Settings persist to %APPDATA%/Radioo/webice-settings.json. A debug log (both main and renderer console output) is written to %APPDATA%/Radioo/radioo-debug.log.