Project · In development
Real-time Avatar Generation
An experimental pipeline that turns speech into a live stream of talking-avatar frames.
Generate frames while the audio plays
The goal is to take incoming speech, generate matching mouth movements and facial animation, and stream frames to the client as they’re produced. The client plays the frames in sync with the audio. This lets the avatar start speaking before the full clip is ready.
Measure the wait and the flow
I’m working on two latency problems. The first is how soon the first frame appears. The second is whether frames arrive steadily enough for smooth playback. A playback buffer can absorb brief stalls, but it also adds delay. I’m testing that trade-off alongside lip-sync drift and the transition from idle to speaking.