MultiTalk

Why we included this project

MultiTalk turns separate audio streams into realistic talking-head video of two or more people conversing. Instead of animating one speaker at a time, it coordinates a whole group, deciding who is listening and how they react while someone talks. It also handles singing, cartoon characters, and explicit interaction control, so it is not limited to plain dialogue. Developers and small teams exploring automated podcast production, dubbing pipelines, or interactive avatar demos will find the included Hugging Face checkpoint lets them move from the NeurIPS paper to something runnable quickly. Teams already working on single-speaker talking-head systems will get a useful reference for the harder multi-person coordination problem.