jms1717/ming-voice is a local voice conversation assistant that receives voice input on AMD Radeon and replies with voice. The latest release in the GitHub repository is v1.6.0. However, the provided FACTS do not list what changed in this version, so the changes cannot be confirmed. The repository is written in Python and uses the MIT license.
This tool runs Ming-3.0-audio-tiny using ROCm and WSL2 in AMD Radeon environments. It presents a figure of starting the voice response roughly 0.25 seconds after the user stops speaking, and it also includes fused MoE and fastlinear related modules. Besides conversation features, it guides you through search, weather, timers and memory storage, and the README states that Home Assistant and Jellyfin integration are also supported. It provides a web interface and a compatible API, and it can run on Linux and Windows.
The README explains that risky actions such as unlocking require confirmation. The response start time is a self-measurement on the specific hardware presented in the README, and the FACTS contain no results compared with other implementations. So rather than judging general performance by the roughly 0.25 second figure alone, it is more appropriate to view this as a case study of an AMD Radeon based local voice conversation setup.