fix: audio timeout 20s, cap chunks to 3, cap extraction to 30s — fix stuck at 80%
Deepfake Authenticator
fa1d723
fix: AV_MISMATCH hard override always sets verdict to FAKE
Deepfake Authenticator
bd97ad3
- AV_MISMATCH now bypasses visual threshold entirely
- Confidence boosted to min 72% when AV_MISMATCH fires
- Lowered mismatch trigger: visual >= 0.45 AND audio < 0.55
- Cleaned up broken if/elif syntax in verdict block
fix: audio-visual mismatch detection for face-swap deepfakes
Deepfake Authenticator
03f204c
- Swap audio model to Vansh180/deepfake-audio-wav2vec2 (ASVspoof-trained, 92.8% acc)
- Add AV_MISMATCH detection: when visual says FAKE but voice is human
this is the hallmark of face-swap deepfakes (dubbed movie audio)
- AV_MISMATCH boosts audio fake score based on visual confidence
- New result label: AV_MISMATCH (yellow warning) vs AI_VOICE (red) vs HUMAN_VOICE (green)
- Frontend: yellow warning badge + icon for AV_MISMATCH state
- Fix tooltip text visibility (white on dark, colored border)
- Fix timeline bar heights (px not %, overflow:visible for tooltips)
