Speak and type anywhere on Linux, no cloud attached

What it does

Vocalinux is a background system-tray GTK app that listens to your microphone and transcribes speech into whatever window currently has focus. It bundles three local speech-recognition backends—whisper.cpp by default, plus OpenAI Whisper and VOSK—and works across both X11 and Wayland desktops without sending audio to a server.

The interesting bit

The project’s real work is the plumbing: it injects text via IBus when possible, falls back to Wayland clipboard and input-automation tools when necessary, and uses Vulkan to accelerate whisper.cpp on AMD, Intel, and NVIDIA GPUs alike. That combination of universal GPU support and desktop-environment-agnostic input injection is exactly the kind of tedious integration Linux usually lacks.

Key highlights

• Fully offline operation with three swappable engines (whisper.cpp, Whisper, VOSK)
• GPU acceleration via Vulkan rather than vendor-locked CUDA
• Layout-aware global hotkeys supporting toggle and push-to-talk modes
• Flatpak and AUR packaging for distro-agnostic or Arch-native installs
• Optional voice commands for punctuation and basic editing (“new line”, “delete that”)

Caveats

• Explicitly beta; recent releases have been stability patches fixing IBus reliability and first-dictation focus issues on GNOME Wayland
• Flathub publishing is still in progress, so sandboxed installs currently require building the manifest yourself
• Settings dialog sizing and scroll behavior needed a point-release fix, suggesting the GTK UI is still settling

Verdict

Worth a look if you want private, offline voice typing on Linux and don’t mind beta-grade desktop-integration quirks. If you expect polished, first-party OS integration out of the box, wait for a stable release.