PunithVT/ai-avatar-system ? reverse-engineered prompt

Build me a self hosted web app where someone can upload a face photo, record a short sample to clone a voice, and then have a live conversation with that avatar. I want the avatar to listen through the microphone or typed text, turn speech into text, answer with an AI model, and show a talking head video with lip sync in real time as the reply comes in. It should feel fast and interactive, with the option to stop or interrupt the avatar while it is still talking.

Please make it work as a normal product, not just a demo, with login, saved conversations, and a clean dashboard for managing avatars and voices. It should run locally with Docker if possible, but also be easy to deploy to a GPU server. If you need current setup details for any AI model or deployment steps, look them up online and use the latest docs.