Linly-Talker is an open-source digital avatar conversational system on GitHub, with about 3,474 stars under the MIT license and most recently active in 2026. Instead of chatting with a plain text box, you talk face to face with a virtual persona: the system listens to your voice, thinks with a large language model, answers in a synthesized voice, and animates a talking head that speaks the reply back to you.
The project combines several AI components into one pipeline. Speech recognition is handled by Whisper and Alibaba’s FunASR, dialogue is powered by large language models such as GPT, Gemini, Qwen, or the Linly model itself, speech synthesis uses Microsoft Speech Services with a wide choice of voices, and the SadTalker talking-head system generates the animated face. You can upload any photo to create a digital persona, or capture one directly from your camera inside the interface. A Gradio WebUI makes the system easy to try, a FastAPI deployment option serves programmatic use, and generated videos can include subtitles for clarity. The modular design means each stage, recognition, dialogue, speech, and animation, can be swapped for alternatives as better models appear, which keeps the system useful as the field moves quickly.
Linly-Talker is completely free and open source under the MIT license. You run it on your own hardware, though some connected cloud speech or model services may require your own accounts.
Linly-Talker suits creators building virtual presenters, researchers experimenting with digital humans, and developers who want a complete, hackable reference stack for voice-driven avatar conversation.