Open access
Sep 2026
DynaID-VAE for Speech-Driven Virtual Anchor Generation: Identity-Disentangled Temporal Memory Variational Modeling
DynaID-VAE outperforms the one-reference baselines Wav2Lip, PC-AVS, SadTalker, and DiffTalk under both protocols and on unseen VoxCeleb2 identities under both protocols and on unseen VoxCeleb2 identities.
Run-Duo Yang, Liang Chen
· Electronics · 0 citations