Step 1: run Whisper (pip or whisper.cpp)
- Official package: install FFmpeg (sudo apt update && sudo apt install ffmpeg on Ubuntu/Debian, or sudo pacman -S ffmpeg on Arch), then pip install -U openai-whisper. On x86_64 Linux, the package's requirements also pull in triton.
- whisper.cpp: clone, download a ggml model (sh ./models/download-ggml-model.sh base.en), and build with cmake -B build. Add -DGGML_CUDA=1 for NVIDIA or -DGGML_VULKAN=1 for Vulkan. The README also covers AMD ROCm.
- Both tools are MIT-licensed. Model size is your main resource choice; see the model table in our Windows guide, which applies to every OS.
Step 2: find your microphone (ALSA, PulseAudio, PipeWire)
Most current desktops route audio through PipeWire, which also serves PulseAudio and ALSA applications. Older setups may use PulseAudio directly. Check that your microphone is visible before you blame the model.
- arecord -l lists ALSA capture devices.
- wpctl status shows PipeWire devices and the current default source; wpctl set-default ID changes it.
- Record a test clip whisper.cpp can read, which means 16-bit mono at 16 kHz: arecord -f S16_LE -r 16000 -c 1 -d 10 note.wav.
- whisper.cpp's live whisper-stream example captures the microphone through SDL2 (sudo apt-get install libsdl2-dev, then build with -DWHISPER_SDL2=ON).
Step 3: get the text into the app (X11 vs Wayland)
Whisper gives you text in a terminal or a file. Typing it into another window is a separate step, and your display server decides which tools work.
| Session | Tool | What its project says |
|---|---|---|
| X11 | xdotool type "…" | Simulates keyboard input through X11's XTEST extension. Its README warns that typing does not work correctly on Wayland |
| Wayland | wtype "…" | Described as "xdotool type for wayland" |
| Any (X11, Wayland, console) | ydotool + ydotoold | Emulates an input device through the kernel's uinput. The daemon needs access to /dev/uinput, which usually requires root permissions |
| Wayland clipboard | wl-copy / wl-paste | Copies text to the clipboard so you paste it yourself |
A minimal DIY loop is: record with arecord, transcribe with whisper-cli -m models/ggml-base.en.bin -f note.wav -nt, then pass the text to xdotool, wtype or wl-copy. Bind it to a key in your desktop's shortcut settings. It works, but you maintain every piece, and on Wayland a global shortcut goes through your desktop's own settings or the GlobalShortcuts portal.
Whisper on Linux vs Whispek for Linux
Whispek for Linux does all three steps for you: it records, recognizes your speech in the cloud (it doesn't use Whisper), and types the text where you were typing. No model, no GPU build, no uinput daemon. It does need an internet connection.
| DIY Whisper pipeline | Whispek for Linux (beta) | |
|---|---|---|
| Install | pip or a whisper.cpp build, a model, SDL2/ALSA tools, plus a typing tool | .deb for Debian/Ubuntu, or the AppImage (mark it executable and run). x86_64 only, built against an Ubuntu 22.04 baseline. Version 0.1.16 beta |
| Shortcut | Your desktop's keybinding or a portal | Left Ctrl + Left Alt + Space for Voice input by default; change it on the Home screen |
| X11 | xdotool or ydotool | Registers an ordinary key combination and places text in the focused app |
| Wayland | wtype, ydotool or clipboard | Not yet fully tested on Wayland. Whispek runs in a basic mode there: the shortcut goes through your desktop's global-shortcut portal, so use the combination your desktop assigns. If the text does not appear in the other app, copy and paste it from Whispek. |
| Writing tools | Whatever you script | Auto cleanup, Dictionary, Transforms, Snippets, History, Whispek Assistant |
| Offline | Yes, once set up | No. Internet required; History stays on your computer |
| Cost | Free (MIT) plus your time | Free with an allowance every 7 days; Pro $12/month or $99/year ($8.25/month equivalent, billed yearly), before tax; all current Pro features unlimited |
On X11 the hotkey is an ordinary key combination, Left Ctrl + Left Alt + Space by default. On Wayland Whispek runs in a basic mode: the hotkey goes through your desktop's shortcut portal, and if text doesn't land, copy it from Whispek. x86_64 only for now.
Linux is in beta and not yet fully tested on Wayland. If something fails on your distribution or compositor, tell us which one.
Which Linux route fits you
- Audio must stay on your machine, or you have no network → a local Whisper pipeline. Whispek is not the right fit.
- You enjoy wiring tools together and you're on X11 → whisper.cpp + arecord + xdotool is a reasonable weekend project.
- You want to talk into a browser, editor or chat app today with no scripts to maintain → Whispek. Install the .deb or AppImage, press Left Ctrl + Left Alt + Space and talk. Free every 7 days.
- Working in a terminal (Claude Code, Codex CLI) → dictate in Whispek, review, then copy and paste into the terminal.