13 June 2026, publication

New preprint on audio-visual deepfake detection across domains

EAV-DFD uses a teacher-student structure to adapt an ensemble audio-visual deepfake detector to unseen domains with only a small sample of target data.

Deepfake detectors usually lose much of their accuracy on media that does not resemble their training set. EAV-DFD recovers a large part of that gap using a teacher-student adaptation step, and reports which modality — audio or video — was manipulated.

Related publication

Teacher-Student Structure for Domain Adaptation in Ensemble Audio-Visual Video Deepfake Detection

Elham Abolhasani, Maryam Ramezani, Hamid R. Rabiee

arXiv preprint, 2026Preprint

arXiv

BibTeX

@misc{abolhasani2026teacher,
  title = {{Teacher-Student Structure for Domain Adaptation in Ensemble Audio-Visual Video Deepfake Detection}},
  author = {Elham Abolhasani and Maryam Ramezani and Hamid R. Rabiee},
  year = {2026},
  eprint = {2606.15117},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2606.15117}
}