Native dictation for Linux that types into whatever you’re using

What it does

hyprwhspr is a system-wide speech-to-text tool for Linux Wayland sessions. You hit a global hotkey, speak, and it transcribes your voice—either with local in-memory models or via cloud APIs—and drops the text directly into the active application buffer. It also handles translation, long-form recording with pauses, and optional audio ducking while you dictate.

The interesting bit

Unlike most dictation tools, it ships a themed visualizer that automatches the Omarchy desktop theme, integrates with Waybar, and lets you manually unload models from GPU memory to free VRAM. That is a surprising amount of desktop-environment bookkeeping for a speech-to-text utility.

Key highlights

• Local-first defaults using Whisper, Parakeet, or Cohere Transcribe models, with optional cloud fallbacks via REST API or WebSocket (Gemini, ElevenLabs).
• System-wide auto-paste into any active buffer, plus toggle, push-to-talk, and long-form recording modes.
• Themed mic overlay on layer-shell compositors and desktop notifications on GNOME/Mutter.
• GPU memory controls to limit or zero VRAM usage, plus CPU-optimized onnx-asr paths for non-GPU hardware.
• Full CLI management for configuration, models, hotkeys, and systemd services.

Caveats

• Wayland and systemd are hard requirements; X11 sessions need not apply.
• Non-Arch Linux support is explicitly flagged as new and potentially rough.
• The visualizer and Waybar integrations require optional dependencies (gtk4, PyCairo, waybar).

Verdict

Ideal for Linux developers and writers who want dictation that feels native to their desktop workflow. If you are on X11, Windows, or macOS, this is not your tool.