Speech to text for Linux that types into any app

Good speech to text on Linux usually means wiring up a model, a hotkey daemon and a typing tool yourself. Vurb packages that into one app: a dictation key, a small on-screen pill while you talk, local transcription, and text typed into the focused window — terminal, editor, browser or chat client.

The default key is Right Control. Change it to Left Control, Alt, the right Super key or Right Shift in settings. Double-tap for hands-free dictation, tap once more to stop, and press Esc to discard a recording.

Vurb currently needs an X11 session to listen for the dictation key and type into other windows. Wayland isn't supported yet. If your distribution defaults to Wayland, choose the X11 (or “Xorg”) session at the login screen.

We'd rather say this up front than have you find out after installing.

Vurb runs OpenAI's open-source Whisper model with whisper.cpp on your own machine. Pick from Fastest (English only, 148 MB), Fast (148 MB), Balanced (190 MB, the default on Linux) or Best (574 MB). The multilingual models handle 90+ languages and detect them automatically.

Once a model has downloaded, speech to text works completely offline. Your audio is transcribed and discarded, and your history stays in Vurb's local data folder. Optional AI cleanup sends text only, to Anthropic, with your own API key.

Because Vurb types into whatever window has focus, it fits the way Linux users already work. A few common uses:

• Commit messages and pull request descriptions, typed straight into the terminal or your Git client.
• Long prompts for coding assistants, which are tedious to type and better when they're specific.
• Replies in chat and email clients, documentation, and notes in your editor of choice.

Add project names, libraries and acronyms to custom vocabulary so they come out spelled correctly. Optional AI cleanup, using your own Claude API key, removes filler words and fixes punctuation if you want tidier text. More examples are on the developers page.