Credit: University of Washington
A neural network first converts the sounds from an audio file into basic mouth shapes. Then the system grafts and blends those mouth shapes onto an existing target video and adjusts the timing to create a new realistic, lip-synced video.