AI that listens.
AI that understands.
Voice dictation for Linux that runs entirely on your computer. Press a key, speak, and your words appear wherever you're typing.
Not released yet. We’ll email you once, when it is.
How it works
As easy as one, two, three.
Press
Tap your hotkey, or hold it down.
Speak
Say what you mean, in your language.
Done
It is typed exactly where your cursor is focused.
Why Quoth
Dictation that behaves like a local tool, because it is one.
Everything happens on your computer. Your voice and words never leave the device, there is no account to make, and crash reports stay off unless you turn them on.
Quoth uses your NPU or GPU when you have one and keeps the model warm, so the text lands the moment you stop talking.
Editors, browsers, chat, terminals. Whatever has focus gets the text. No plugins, no per-app setup.
Deutsch
English
Speak in any of 90+ languages and get fluent English text, straight into the field you are typing in.
Choose a pill or an orb. Use one of the four included themes, or make your own.
Fits right in
Made for the Linux desktop you already have.
Get Quoth
Your distro, your way.
Every distro
SoonOne click from Flathub.
Arch
SoonFrom the AUR, with the NPU path set up out of the box.
Debian, Ubuntu, Pop!_OS
SoonA .deb from the GitHub releases page.
Can I install Quoth yet?
Not yet. The first release is being prepared for Flathub, the AUR and as a .deb. Leave your email above and we’ll write once, when it’s out.
Does anything I say leave my computer?
No. Your speech is transcribed on your own computer with whisper.cpp, and the audio is never uploaded. Quoth only goes online to download a speech model and, if you turn them on, to check for updates or send a crash report. It never tracks how you use it, and there is no account.
Which desktops does it work on?
Linux desktops running Wayland: COSMIC, KDE Plasma, Sway and other wlroots compositors, and GNOME. On GNOME and KDE, typing needs a one-time setup of ydotool, and GNOME doesn’t show the floating pill. X11 support is planned.
What do I need to run it?
A 64-bit x86 processor with AVX2, which most PCs from 2015 on have, and a Wayland desktop. The default speech model needs about 300 MB of memory and 150 MB of disk space; the largest needs about 2 GB of memory and 1.6 GB of disk. ARM support is planned.
Do I need a GPU?
No. Every model runs on the processor. A graphics card from any vendor, used through Vulkan, or an Intel NPU makes the larger, more accurate models fast enough for everyday dictation. NPU support comes with the AUR and .deb packages, not the Flatpak.
Which languages does it understand?
Quoth detects the spoken language on its own and lets you pick from 19 languages, including English, German, Spanish, French, Chinese, Japanese, Arabic and Hindi. It can also translate what you say into English as you dictate.
How do I dictate?
Press the hotkey, Pause by default, speak, and press it again. Or hold it while you talk. The text appears in whichever window has focus, pasted through the clipboard, which Quoth restores afterwards. You can also bind any key to quoth --toggle in your desktop’s shortcut settings.
What does it cost?
Nothing. Quoth is free software under the GPL-3.0-or-later. There is no paid tier and no subscription. If it earns a place on your desktop, you can support its development.