New: the Voice AI Investors list release! Check it out

    murmure

    Git Repo
    Kieirra

    Transcribes speech to text locally using NVIDIA models and LLM post-processing for 25 European languages without internet connectivity.

    About murmure

    Murmure: Local, Private Speech-to-Text

    Murmure is a privacy-first, open-source speech-to-text application designed to run entirely on your local machine. By leveraging NVIDIA’s Parakeet TDT model, it provides fast, multilingual transcription without the need for an internet connection or data collection.

    For the Non-Technical Reader

    Imagine having a personal assistant that listens to your voice and types out your thoughts instantly, but with one crucial difference: they have zero memory and no way to tell anyone else what you said. Murmure turns your computer into a private dictation station. Whether you are drafting emails, taking notes, or talking to an AI like ChatGPT, your voice never leaves your device. It is the perfect solution for professionals who handle sensitive information but want the speed of voice-to-text technology.

    For the Technical Reader

    At its core, Murmure utilizes the NVIDIA Parakeet TDT 0.6B v3 model. This Temporal Dependency Transformer (TDT) architecture is optimized for low-latency, high-accuracy transcription. Key technical highlights include:

    • Architecture: Fully local execution with support for 25 European languages.
    • Post-Processing: Integrates with local LLMs for grammar correction and translation.
    • Privacy: Zero telemetry and no external API dependencies.
    • Compatibility: Cross-platform support (Windows, Linux, MacOS) with specific optimizations for NVIDIA hardware and Intel/Apple Silicon.

    Why It Matters

    The shift toward local-first AI is accelerating. Murmure represents a move away from the "API-as-a-Service" model, offering a zero-cost, high-privacy alternative to proprietary cloud solutions. For industries like legal, healthcare, or R&D, this removes the compliance hurdles associated with sending audio data to third-party servers while maintaining the performance levels expected from modern neural networks.

    The Voice AI Space Lab Idea

    Build a "Local Creative Suite": Combine Murmure with a local LLM (via Ollama) and a local document store. You could dictate a rough project outline while walking around your office; Murmure transcribes it, the local LLM structures it into a project plan, and it’s automatically saved to your private vault—all without a single packet of data hitting the public internet.

    Explore the repository here: https://github.com/Kieirra/murmure