Humans can detect artificially generated audio known as ‘deepfakes’ around 3 out of 4 times, according to new research published in PLOS ONE.
Deepfakes use artificial intelligence (AI) to create highly convincing – but false – images, audio or video. In this case, researchers from University College London investigated participants’ ability to detect AI generated audio.
Deep fake audio algorithms are widely available, and are able to recreate a person’s voice from a 3-second clip of them speaking. The technology can then be used to mimic that person’s voice to convey new information.
In the study, researchers used publicly available datasets to obtain 50 genuine samples of audio in both English and Mandarin. They also generated 50 deepfake speech samples.
The 529 participants in the study were played 20 randomly chosen and ordered samples of both real and artificial audio and asked to identify the fake clips. Participants were able to correctly identify fake audio 73% of the time. Further training in detection improved their accuracy slightly.
Kimberly Mai, computer scientist and author of the study says: “Our findings confirm that humans are unable to reliably detect deepfake speech, whether or not they have received training to help them spot artificial content.”
Deepfakes in English and Mandarin were equally challenging to identify, and spending more time on the task or listening to clips more frequently didn’t help people to detect the artificially generated audio, according to the paper.
As algorithms become more advanced the task of identifying artificial speech will become even more challenging, the researchers write. Given these findings they argue further research should focus on improving automated detection tools.
The paper cites examples of people already using deepfake audio technology to commit fraud or criminal activity.
“With generative artificial intelligence technology getting more sophisticated and many of these tools openly available, we’re on the verge of seeing numerous benefits as well as risks,” says co-author Professor Lewis Griffin.
SCINEMA runs from August 1 to August 31 every year. Register now to be part of the festival and watch the films for free. REGISTER NOW.
Originally published by Cosmos as Detecting deepfake audio can be hit and miss for humans
Petra Stock
Petra Stock is a journalist and engineer. She has previously worked in climate change, renewable energy, environmental planning and Aboriginal heritage policy.
Read science facts, not fiction...
There’s never been a more important time to explain the facts, cherish evidence-based knowledge and to showcase the latest scientific, technological and engineering breakthroughs. Cosmos is published by The Royal Institution of Australia, a charity dedicated to connecting people with the world of science. Financial contributions, however big or small, help us provide access to trusted science information at a time when the world needs it most. Please support us by making a donation or purchasing a subscription today.
