ForReal
20241st runner-up — Code4Change
Tells real audio and images apart from AI fakes at 93%+ — because a spoofed call from "your dad" shouldn't work.
Voice and video deepfakes have moved well past memes — it's not hard to spoof a phone call from someone's parent and phish sensitive details out of them. ForReal. is a check against that.
It's a cross-modal CNN trained to catch the artifacts generators leave behind: MFCC features from 80,000+ real and synthetic audio files, plus pixel data from 40,000+ images. Output is a real-vs-deepfaked call at over 93% accuracy. 1st runner-up at Code4Change (IEEE-SSIT × CSED VIT).
PythonTensorFlowKerasCNNs