Can you tell the difference between human and AI-generated speech? This interactive tool lets you listen to short audio clips and decide whether they’re real or deepfaked. It’s part of a broader effort to study human perception and improve deepfake detection in speech.

Hey Guys, I'm building a website where you can upload audio files or youtube links and have them analyzed by our audio deepfake detection model. Give it a go / give us feedback!

Dear All, I set up a game where you listen to a recording of human speech, and have to tell spoofed and authentic audio apart. While you play it, a correpsondingly trained AI also listens to the audio and tries the same. You playing will give us additional insight into how ML models understand authentic and spoofed audio, so it would be much appreciated :) Thanks!

See how well you can detect state-of-the-art audio deepfakes in this interactive game!

In order to better understand audio deepfakes (voice recordings faked with the help of AI), we have developed an online game: You play against an artificial intelligence and try to distinguish deepfakes from real audio recordings.

We find that it is surprisingly hard to tell these apart. You can try it out yourself. Average Human Accuracy is about 77 Percent. A specifically trained AI achieves > 90 percent test accuracy on the data.