goodroot/hyprwhspr

hyprwhspr – Fast, private speech‑to‑text for Linux

What it is

• A native Linux application that turns spoken words into text anywhere you can type. It runs as a system‑wide service and can paste the transcription directly into the active window.
• Designed for performance: it can use on‑device models (Whisper, Cohere, Parakeet, Qwen3‑ASR, etc.) that run in memory, or optionally call cloud APIs.
• Works on both Wayland and X11 desktops (Hyprland, Sway, KDE, GNOME, etc.) and integrates with status‑bars like Waybar or Noctalia.

Key features

Feature	What it does
Instant, accurate transcription	Loads models into RAM (or ONNX for CPU) for near‑real‑time dictation.
Offline‑first / privacy	All local models keep audio on your machine; cloud services are optional.
Multiple back‑ends	Choose from Cohere Transcribe, Parakeet TDT V3, Whisper, Qwen3‑ASR, or a generic REST/WebSocket API.
GPU‑efficient	CUDA (NVIDIA) or Vulkan (AMD/Intel) acceleration, with options to limit or drop GPU memory usage.
CJK support & translation	Qwen3‑ASR handles Chinese/Japanese/Korean; built‑in translation can convert non‑English speech to English.
Live visual feedback	An on‑screen visualiser that matches your desktop theme (Omarchy, Noctalia, etc.).
Hot‑key control & auto‑paste	Default Super+Alt+D toggles dictation; results can be auto‑pasted or copied to the clipboard.
Audio ducking	Optionally lowers other audio while recording.
REST API / WebSocket	Expose the service to other programs or cloud providers like Gemini or ElevenLabs.
Long‑form mode	Pause and resume dictation, saving intermediate results.

How to get it

• Arch Linux – install from the AUR (hyprwhspr or hyprwhspr‑git).
• Other distros – run the one‑line installer: curl -fsSL https://hyprwhspr.com/install.sh | bash.
• Manual install is also supported via the repository’s scripts.

Typical workflow

1. Run hyprwhspr setup – the interactive wizard walks you through selecting a transcription backend, downloading the model, and configuring visualiser / bar integration.
2. Log out/in to apply group permissions.
3. Press the hot‑key (Super+Alt+D) to start dictation; speak naturally.
4. Press the hot‑key again to stop – the transcribed text appears in the currently focused window, or is auto‑pasted.

Command‑line utilities

• hyprwhspr record – start/stop dictation from a terminal.
• hyprwhspr transcribe <file> – transcribe an audio file.
• hyprwhspr model – download, list, unload or reload models.
• hyprwhspr config – view or edit the configuration file.
• Service management commands (hyprwhspr systemd, hyprwhspr uninstall, etc.)

Documentation & help

• Full docs at https://hyprwhspr.com/docs – includes configuration, recording modes, backend details, and troubleshooting.
• Logs are available via journalctl --user -u hyprwhspr.service.
• Issues can be opened on GitHub.

License

• MIT License.

Bottom line hyprwhspr is a genuine, actively maintained open‑source speech‑to‑text tool for Linux, targeting low latency, privacy‑preserving dictation with flexible backend choices and desktop‑integration features.