JMS1717/ming-voice

Low-latency local speech-to-speech voice assistant with Ming-3.0-audio-tiny on AMD and Nvidia GPUs. Easy setup on Windows and Linux

What it solves

Ming Voice provides a fully local, real-time voice assistant that runs entirely on your own hardware. It eliminates the need for cloud-based voice services, ensuring that your audio and personal data never leave your machine while offering low-latency interaction for home automation and daily tasks.

How it works

The project uses a speech-to-speech model that processes audio locally on your GPU. It listens for your voice, determines when you have finished speaking using smart turn detection, and generates a spoken response. It can be used via a browser, a dedicated desktop client, or integrated into Home Assistant. It supports tools for web searching, weather, timers, lists, and controlling smart home devices like lights, thermostats, and media players.

Who it’s for

It is designed for users with a powerful GPU (AMD or NVIDIA) who want a private, always-on voice assistant. It is ideal for those who want to control their smart home, manage personal reminders, or perform web searches without relying on external cloud providers.

Highlights

  • Fast response time: Typically begins speaking about a quarter of a second after you finish your sentence.
  • Natural conversation: Supports interrupting the assistant while it is speaking and handles follow-up questions without needing to repeat wake words.
  • Local-first privacy: No audio leaves your machine; only search terms are sent to external services if web tools are enabled.
  • Extensive integrations: Controls Home Assistant devices, Jellyfin media servers, and manages personal lists, timers, and memory.
  • Flexible deployment: Works on Linux and Windows (via WSL2), supports voice cloning, and can be accessed from mobile devices over your home network.

Related

  • Project
    kstonekuan/tambourine-voice

    An open-source universal voice-to-text interface that uses AI to transcribe and format speech into clean text, typing it directly into any active application.

  • Project
    livekit/python-sdks

    LiveKit’s Python SDK provides two packages (`livekit` for real‑time media participation and `livekit-api` for server‑side token and room management). It enables Python apps to join LiveKit video/audio rooms, publish/consume media, run RPC between participants, and integrate with LiveKit’s voice‑AI Agents (STT, LLM, TTS). Install via pip, generate a JWT token, connect to a room, and use the `MediaDevices` helper for microphone/speaker handling. The SDK is async, works with Python 3.9+, and is MIT‑licensed.

  • Project
    kouhxp/yapsnap

    A CPU-based command-line tool that transcribes video URLs and audio files into plaintext locally, featuring automatic language detection and speaker diarization.

  • Project
    OHF-Voice/wyoming-faster-whisper

    Wyoming Faster Whisper is a server that implements the Wyoming protocol for Home Assistant, using the faster‑whisper speech‑to‑text engine. It offers fast, local transcription, optional GPU support, VAD‑based endpointing, and can bias results with Home Assistant entity names.