ReadAloud

Select text anywhere, press Ctrl+Super+S — or just wiggle the mouse — and hear it.

Three platforms, each the native way:

• Linux (this folder) — the original, fully tested: Google voice + offline fallback, top-bar tray, wiggle-to-read.
• windows/ — built-in Windows voice, system-tray icon, Ctrl+Win+S, wiggle via mouse hook. One-liner: irm https://raw.githubusercontent.com/IAFahim/ReadAloud/master/windows/setup.ps1 | iex
• macos/ — built-in say voices, clipboard-safe Cmd+C capture. One-liner: bash <(curl -fsSL https://raw.githubusercontent.com/IAFahim/ReadAloud/master/macos/setup-mac.sh)

The ports compile clean and pass their logic self-checks, but were written on Linux — not yet run on real Windows/macOS. First thing to run there: ReadAloud --self-check.

• Press again with a new selection: interrupts and reads the new one.
• Press with nothing selected: stops talking.
• Voice engines (tray → Engine):
  • Google — free Translate voice (needs internet); falls back if it fails.
  • Inflect — local neural voice (Inflect-Micro-v2), ~40 MB ONNX under ~/.local/share/readaloud/. Warm worker keeps the model loaded. English, one fixed voice. Install once: publish/ReadAloud --install-inflect (or tray → Inflect → Install / update). Needs uv + hf once.
  • Offline — spd-say robot, always available, never pretty.
• Check what is wired: publish/ReadAloud --engine-check
• Want Claude Code replies auto-spoken? Deliberately NOT installed by default — with several agent sessions open the voices trample each other. Select + wiggle instead. To opt in anyway, add a Stop hook running publish/ReadAloud --claude-hook (timeout 600, async).

Top-bar controls

A speaker icon sits in the GNOME top bar (ReadAloud --tray, started at login by ~/.config/autostart/readaloud-tray.desktop). Click it for:

• Stop speaking — shuts the voice up right now.
• Speed — 1.0x to 3.0x, the current one is ticked.
• Pitch — Lower / Natural / Higher.
• Engine — Google / Inflect (local neural) / Offline.
• Inflect — Install / update, Warm up, Stop worker, Steady / Natural / Expressive delivery.
• Wiggle to read — select text, shake the mouse left-right, it speaks (with a little pop). Needs one-time access: sudo usermod -aG input $USER, then log out and back in. Sensitivity is on the tray (Wiggle → Sensitive / Normal / Firm / Stubborn). Too twitchy? pick Firm or Stubborn. Fine-tune the four knobs in settings.json if you want.
• Open settings file — opens ~/.config/readaloud/settings.json in your editor.
• Quit — removes the icon (and stops the Inflect worker) until next login.

Every click writes that file, and every read picks it up, so nothing needs a restart. Edit the file by hand if you prefer; the menu re-reads it each time you open it.

Hack it

Speaker logic in Program.cs + Engines.cs, top-bar icon in Tray.cs, shared knobs in Settings.cs → ~/.config/readaloud/settings.json (speed, pitch, engine, language, shortcut). Rebuild after edits:

dotnet publish -c Release -o publish

The GNOME shortcut runs publish/ReadAloud; change the key in Settings → Keyboard → Custom Shortcuts.

New PC setup — one line

bash <(curl -fsSL https://raw.githubusercontent.com/IAFahim/ReadAloud/master/setup.sh)

Installs everything: packages, .NET 10, the app, the shortcut, the top-bar icon, autostart, and mouse permission for wiggle. Or manually:

git clone https://github.com/IAFahim/ReadAloud
cd ReadAloud && ./install.sh

install.sh builds, registers the shortcut (the Shortcut field in ~/.config/readaloud/settings.json, default Ctrl+Super+S), installs the tray autostart entry, downloads the Inflect voice (skip with READALOUD_SKIP_INFLECT=1), and starts the tray. No Claude Code hook is installed.

Where next

See IDEAS.md — 15 ranked upgrades (speech daemon, per-agent Piper voices, gist-first summaries, OCR, dictation, karaoke overlay…).