goodroot/hyprwhspr

Native speech-to-text for Linux - Fast, accurate and private system-wide dictation

What it solves

It provides a native, system-wide speech-to-text (STT) dictation tool for Linux users, allowing them to convert voice to text instantly and paste it into any active application buffer without needing to write manually.

How it works

The tool captures audio from the microphone and processes it using various transcription backends. It supports local, in-memory models for privacy and speed, as well as cloud-based APIs (like Gemini and ElevenLabs) via REST or WebSockets. It is designed for Wayland sessions and integrates with systemd user services for background operation.

Who it’s for

Linux users (specifically those on Wayland) who want a fast, private, and highly customizable dictation experience, particularly those with high-end NVIDIA GPUs for maximum performance, though it also supports CPU-only setups via onnx-asr.

Highlights

  • Flexible Backends: Supports Cohere Transcribe, Parakeet TDT V3, Whisper, and cloud APIs.
  • Linux Native: Built for Wayland and supports multiple distributions including Arch, Debian, Ubuntu, Fedora, and openSUSE.
  • Automatic Integration: Auto-pastes text into any active buffer and offers optional audio ducking to reduce system volume during recording.
  • Visual Feedback: Includes a themed visualizer and an OSD overlay to indicate recording status.
  • Hardware Optimized: Optimized for CUDA (NVIDIA) and Vulkan (AMD/Intel) acceleration, with high-speed CPU options.
  • Multi-lingual: Supports translation of non-English speech to English and multi-language transcription.