Hyprwhspr

Hyprwhspr is a system wide speech to text tool for Linux that runs on Wayland and X11. It captures speech from the microphone, transcribes it, and types the result into whichever application has focus, which makes it useful for dictating prompts, commit messages and documents without leaving the keyboard driven desktop. Transcription runs locally by default. It supports Parakeet TDT V3, Cohere Transcribe and the full Whisper family, and detects NVIDIA CUDA, AMD or Intel Vulkan, or falls back to the CPU. Users who prefer cloud transcription can connect providers such as Google Gemini, OpenAI and ElevenLabs, including near real time streaming over WebSockets. There are five recording modes covering toggle, push to talk, auto, continuous and long form, plus audio ducking, filler word removal, word overrides and an optional translate to English mode. The tool is aimed at Linux users and developers who want a scriptable setup. Text is injected through wtype or ydotool with per application paste rules, a full command line interface allows control from other tools, and a post transcription hook can pipe every result through a custom shell command. It works across Hyprland, GNOME, KDE Plasma and Sway, with a Waybar status indicator, and installs on Arch Linux through the AUR. Hyprwhspr is MIT licensed and free. The project states there are no subscriptions, no telemetry and no monetization.