KSDiff introduces dual-path speech disentanglement and autoregressive keyframe prediction inside a diffusion model to improve lip synchronization and head-pose realism in audio-driven facial animation.
Improved parallel wavegan vocoder with per- ceptually weighted spectrogram loss,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.GR 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
KSDiff: Keyframe-Augmented Speech-Aware Dual-Path Diffusion for Facial Animation
KSDiff introduces dual-path speech disentanglement and autoregressive keyframe prediction inside a diffusion model to improve lip synchronization and head-pose realism in audio-driven facial animation.