Duix-Mobile

Why we included this project

Duix Mobile is an SDK rather than a ready-made app: you plug in your own LLM, speech recognition, and text-to-speech services, and it handles the hard part, rendering a digital human with lip sync, facial expressions, and streaming audio on phones, tablets, and embedded screens. Everything runs on-device with no cloud dependency, which makes it a reasonable fit for finance, government, and in-car use where keeping data local matters. Response latency stays under 120 ms on a Snapdragon 8 Gen 2, and the audio supports interruption and barge-in, so the avatar can stop talking and respond, which makes exchanges feel less scripted. Teams that want a talking avatar under their own control, rather than renting a hosted service, should find this a solid base to build on.