A team of computer science researchers have made an ex-president "lip-sync" words as they chose, and in a very realistic way.
The scientists used a neural network -- the AI tech everyone is talking about these days, and that recently beat the world Go champion.
They had the neural net study hours of Obama video and understand how his face moves. Then they used a program based on this understanding to make new videos of Obama's lips saying what they wanted them to. They cut and pasted Obama's body into two different scenes to show how realistic their fairly simple project could make anyone say anything on video.
The report, from a University of Washington team, is called "Synthesizing Obama: Learning Lip Sync from Audio."
No comments:
Post a Comment