The paper introduces AVT2-DWF, a framework utilizing Audio-Visual dual Transformers with Dynamic Weight Fusion, aimed at enhancing detection capabilities against deepfake methods that adapt from single to multimodal fusions. Summary points:

• Dual transformers capture spatial and temporal dynamics.
• Uses a face transformer with an n-frame-wise tokenization strategy.
• Incorporates audio transformer encoders.
• Employs dynamic weight fusion for information fusion between audio and visual.