Read Aloud: Text to Speech

Text to speech (TTS) for the Omarchy bar. Read aloud any text you copy with a local AI voice (Kokoro). 54 voices, 9 languages, works offline after setup. Helps with dyslexia and long reads.

Powered by Kokoro: GitHub · 🤗 Hugging Face

Controls

• Left-click the speech bubble: read the clipboard. Click again to stop.
• Right-click: pick a voice, switch male or female, and open Settings for speed, language and keys.
• While a voice is getting ready, the icon turns into a spinning ring. While it reads, it shows a stop square.

Picking a voice plays a short sample. Choices are saved and apply to every read.

Keybind

Add a toggle to ~/.config/hypr/bindings.lua. Copy the shortcut setup in Settings (right-click the bubble) copies these lines for you. They fire on a tap of AltGr + Right Ctrl:

o.bind("CTRL + ALT + Control_R", "Read clipboard aloud / stop", '$HOME/.local/share/omarchy-say/say --clip', { release = true })
o.bind("CTRL + ALT + Alt_R", "Read clipboard aloud / stop", '$HOME/.local/share/omarchy-say/say --clip', { release = true })

Read a screenshot aloud

Drag a box over any text on screen and hear it. Omarchy's built-in OCR copies the text, then Read Aloud reads it. Add this line too:

o.bind("SUPER + ALT + PRINT", "Read screenshot aloud", 'omarchy-capture-text && $HOME/.local/share/omarchy-say/say --clip')

First run

The first click sets up the voice by itself: Python 3.12 and the Kokoro engine through uv, then the 350 MB voice model, checked against fixed SHA-256 hashes. A progress bar shows in the panel, and it says "Voice ready" when done. Everything lands in ~/.local/share/omarchy-say/.

Needs PipeWire (pw-play) and wl-clipboard (wl-paste), which Omarchy ships, and uv, which some installs lack. Check with command -v uv, add it with omarchy pkg add uv.

Install

omarchy plugin add https://github.com/remigius-labs/omarchy-read-aloud --enable

The icon shows in the bar after the shell reloads: omarchy restart shell, or log out and back in.

Install with an AI assistant

Paste this page into your assistant (Claude Code, Codex, …) and ask it to install Read Aloud. Each step has a check, so it knows when to move on.

1. Check the tools. Run command -v omarchy uv pw-play wl-paste. All four should print a path.
   • If uv is missing, run omarchy pkg add uv. If that fails, follow uv's official install guide (https://docs.astral.sh/uv/getting-started/installation/), then open a new shell.
2. Add the plugin. Run omarchy plugin add https://github.com/remigius-labs/omarchy-read-aloud --enable --yes.
   • Check: omarchy plugin list | grep remi.say shows enabled.
3. Set up the voice. Run ~/.config/omarchy/plugins/remi.say/engine/setup. It installs Python 3.12 and the engine through uv, then downloads the 350 MB model. This takes a few minutes.
   • Check: ls ~/.local/share/omarchy-say/.ready exists. If setup stops, it prints the reason; fix that and run it again.
4. Reload the bar. Run omarchy restart shell. The speech bubble shows in the bar.
5. Test the voice. Run ~/.local/share/omarchy-say/say "Read Aloud is ready". You should hear it.
6. Optional keybinds. Add the lines from Keybind to ~/.config/hypr/bindings.lua, including the screenshot one.

Remove

omarchy plugin remove remi.say
rm -rf ~/.local/share/omarchy-say

The second line deletes the voice engine and model (about 350 MB).

Limits

• Kokoro has English (US, UK), Portuguese, Spanish, French, Japanese, Italian, Mandarin and Hindi.
• First sound takes about a second on a laptop CPU.

How it works

Copy text, tap the key, and it reads it out.

The voice engine runs with no network access at all (unshare -n), so nothing you read can leave the machine. This also blocks the usage telemetry that onnxruntime, the library Kokoro runs on, tries to send to Microsoft. On a system with unprivileged user namespaces switched off, it runs normally instead.

License

MIT. Kokoro model weights are Apache 2.0. Packages the setup installs keep their own licenses: kokoro-onnx and onnxruntime (MIT), phonemizer and espeak-ng (GPL-3).