Patent · US Active

Generative adversarial networks

US12354202B1 · kind B1 · utility

0Cited by
0References
18Claims
0Family size

Assignee

Inventors

Key dates

Filing dateOct 5, 2022
Grant dateJul 8, 2025
Priority date
Expiry dateJun 2, 2043

Classification

  • Technology area (CPC G)Physics
  • CPC primaryG10L2021/105
  • WIPO fieldComputer technology
  • WIPO sectorElectrical engineering

Abstract

An audio-driven lip reanimation GAN network and method of reanimating lips based on an input audio using a GAN network. The GAN network includes a 1st stage GAN generator configured to receive 1st stage audio inputs and 1st stage input frames. The 1st stage GAN generator is trained to produce 1st stage synthetic output frames in which a pair of lips in a target face in the 1st stage synthetic output frames has been reanimated in reference to the 1st stage input frames based on the 1st stage audio inputs. The GAN network also includes a 2nd stage GAN generator configured to receive the 1st stage synthetic output frames as inputs. The 2nd stage GAN generator is trained to generate 2nd stage output frames that improve on the realism of the 1st stage output frames.

Source: USPTO / EPO open patent data. Objective bibliographic and citation counts.