Summary Murmure is a privacy-first, open-source speech-to-text application offering fast, local, offline transcription. It processes voice data entirely on-device, ensuring zero data collection and supporting 25 European languages. This makes it ideal for privacy-conscious users and offline environments.

How It Works Powered by NVIDIA's Parakeet TDT model, Murmure performs speech recognition locally using a transformer-based architecture. This on-device inference enables low-latency, rapid transcription without requiring an internet connection. The core advantage is its complete data privacy, as no audio or transcription data ever leaves the user's machine.

Quick Start & Requirements

• Installation:
  • Windows: Download and run the unsigned .msi installer from releases.
  • Linux: Download the .AppImage, make executable (chmod +x), and run. Requires ALSA API; Pipewire users may need pipewire-alsa. Wayland is problematic except on Fedora.
  • Arch Linux (AUR): Install via AUR helpers (e.g., yay -S murmure).
• Prerequisites: Sufficient CPU or GPU for local inference.
• Links: GitHub releases page.

Highlighted Details

• Privacy: All processing is local; no data transmission or collection.
• Offline: Fully functional without an internet connection.
• Languages: Supports 25 European languages.
• Technology: NVIDIA's Parakeet TDT for fast, on-device inference.
• Features: Transcription history, clipboard integration, experimental API.

Maintenance & Community Development is community-driven via GitHub. The project relies on NVIDIA's Parakeet and Tauri. A roadmap indicates future macOS support, API enhancements, and potential LLM integration. Issue reporting is handled on GitHub.

Licensing & Compatibility Released under GNU GPL v3, requiring derivative works to remain open source (copyleft). This may restrict integration into closed-source commercial products without compliance.

Limitations & Caveats Wayland Linux distributions (except Fedora) have compatibility issues. The Windows installer is unsigned. macOS support is planned but not yet implemented. The experimental API may require further stability refinement.