Ming Voice

What Ming Voice does

Ming Voice runs the Ming-3.0-audio-tiny model on your own GPU. It listens, decides when you have finished, thinks and speaks back, with the README citing about 0.23 seconds from the end of a sentence to the first sound. It supports hands-free, push-to-talk and wake-word modes, and you can interrupt it, even on a speakerphone.

Built-in tools cover web search, weather, calculator and clock, timers and reminders that ring until answered, shopping and to-do lists, and remembering things you say. It controls Home Assistant devices and Jellyfin, confirms risky actions like opening a lock first, and offers a Jarvis butler persona. The conversation is saved after every turn. No audio leaves your machine. With tools on, only search terms and place names go to DuckDuckGo or Brave Search and Open-Meteo.

Key features

• Speech-to-speech on a local GPU
• About 0.23 s response start
• Smart turn detection and interruption
• Wake word, push-to-talk or hands-free
• Home Assistant and Jellyfin control
• Hand-off to a Hermes Agent for bigger jobs

When to use it

• Running a private voice assistant at home
• Controlling lights and music by voice
• Passing a larger task to Hermes by voice

Who it is for: Home-lab users with an AMD Radeon GPU who want a private voice assistant.

How it fits with Hermes Agent

Hermes Agent is an optional back end, so the assistant can hand larger jobs to it while handling quick requests itself.

Requirements: An AMD Radeon GPU (NVIDIA experimental) on Linux or Windows

Note: NVIDIA GPU support is described as experimental in the README.

FAQ