Dusk Download

Guide · Linux

Voice dictation on Linux, including Wayland

Dusk is a dictation app for Linux that runs entirely on your computer. Hold Ctrl + Alt, speak, and let go: clean, punctuated text is typed at your cursor in any application. It works on GNOME's Wayland session, Ubuntu's default, and on X11, and installs from a .deb or .rpm package.

Last updated October 27, 2026

What works on each desktop

Wayland limits what one application can do to another, so what Dusk can do depends on the session you log in to.

GNOME on Wayland X11, any desktop Other Wayland desktops (KDE Plasma and others)
Dictation typed at your cursor Yes Yes Yes
Shortcuts, including Ctrl + Alt alone, and mouse buttons Yes Yes Yes
Commands that press keys, open apps and websites, and edit text Yes Yes Yes
Switching between windows by voice Yes Yes Not yet
Clicking links and buttons by voice Most Yes Most

How Dusk works on Wayland

On Wayland, an application can't watch other applications' keys, type into them, or list their windows. That protects you from keyloggers, and it's why most Linux dictation tools only work on X11. Dusk uses what GNOME offers for each of these:

On Wayland, Dusk can't click at a position on screen, so it clicks only links and buttons that report an action of their own through accessibility. Most do.

Install

Ubuntu, Debian, Linux Mint, Pop!_OS

Download the .deb and open it with App Center, or install it from a terminal:

sudo apt install ./Dusk_amd64.deb

Fedora, RHEL, Rocky Linux

Download the .rpm and install it from a terminal:

sudo dnf install ./Dusk_x86_64.rpm

What you need

Speech recognition that runs on your computer

Dusk uses open speech and language models. Setup checks your memory and graphics card, recommends models, and shows each model's license before you download it. The models available in October 2026:

Model Does Languages License
NVIDIA Nemotron 3.5 ASR Streaming 0.6B Speech, transcribed while you speak English, Arabic, Chinese, French, German, Japanese, Korean, Spanish, Ukrainian, Vietnamese OpenMDW 1.1
NVIDIA Nemotron Speech Streaming English 0.6B Speech, transcribed while you speak English NVIDIA Open Model License
NVIDIA Parakeet TDT 0.6B v2 Speech, transcribed when you finish English CC BY 4.0
Moonshine Base Speech, transcribed when you finish English MIT
Qwen3.5 0.8B, 2B and 4B Clean-up and commands English clean-up Apache 2.0

Questions

Does Dusk need Xorg?

No. It works on GNOME's Wayland session, which is the only session on Ubuntu 26.04, and on X11 sessions.

Does it work on KDE Plasma and other Wayland desktops?

Dictation, shortcuts and most commands work there. Commands that switch between windows need GNOME on Wayland or an X11 session, for now.

Is Dusk open source?

No. Dusk is proprietary, with a free plan that has every feature and a weekly word limit. The speech and language models it downloads are open models, each under its own license, shown in the app.

Does it work on ARM computers?

Not yet. The packages are for 64-bit x86 computers.

Does anything I say go online?

No. Speech recognition and clean-up run on your computer. See what stays on your computer for the few connections Dusk makes.