DAVID-Net | Multimodal Deepfake Forensic Detection

Media Ingestion & Player

Multimodal Demux

Select or drop audio-visual media

MP4, WebM, AVI, MOV, WAV, MP3

Direct-to-Inference stream•Max 50 MB

Benchmark Test Presets1-Click Evaluation

Extract Cross-Modal Attention Saliency (Av→a, Aa→v)

Pipeline ArchitectureDAVID-Net 2.0

Visual BackboneVideoMAE (ViT-Base)

Audio BackboneWavLM (Base-Plus)

Cross-AttentionDisentangled Pre-LN

Attribution4-Quadrant Softmax

Forensic Intelligence Dossier Standby

Select an audio-visual media clip on the left, or tap one of the benchmark presets (RVRA, FVRA, RVFA, FVFA) to execute instant multimodal verification.

Feature 012D Cartesian$p_a \times p_v$ coordinate radar

Feature 02DisentanglementIndependent per-stream logits

Feature 03Dual MLP BoundsFrame-dense interval cuts

Feature 04Sync Curve ($s_t$)Phoneme-viseme InfoNCE