Circumventing shortcuts in audio-visual deepfake detection datasets with unsupervised learning

Open in new window