context: biden talking about pulling out of elections

Conclusion

• Just based on the Audio data alone, we can confidently fall it an AI-generated audio.
• There seems to be traces of LipSync in the frames of the Video but AI detector is not able to reliably call it so.

Detailed Results

Step 1: AudioSpoof check on Clean section (Audio check)

• Observations:
  • Technique used: AI voice cloning & voice conversion used like RVC (retrieval voice conversion) to produce Biden-like voice.
• Mean score : -2.797 ( below threshold)
• Final classification : Audio Spoof detected with high confidence

Step 2: Frame-only Deepfake check (Video check)

• Observations:
  • Frames seem to have LipSync based manipulation with high quality enhancement in the post-processing.
  • Our current detectors do not cover these latest generation approaches yet
  • Hence, our frame model is producing a False Positive here & is unreliable