Peer-reviewed research
Warning: Humans cannot reliably detect speech deepfakes
Kimberly T. Mai and Lewis Griffin, PLOS ONE, 2023 · DOI 10.1371/journal.pone.0285333
Read the originalRead on October 17, 2026
- What it is
- A listening experiment from University College London in which 529 participants judged whether English and Mandarin speech clips were real or deepfakes. Listeners identified the deepfakes correctly only about 73% of the time.
- How much weight it carries
- Peer-reviewed in PLOS ONE. The fakes were made with older speech synthesis algorithms, so newer clones could fool listeners more often. The author list here is partial, taken from the page.
Facts taken from this source
- The PLOS ONE study found that giving listeners examples of speech deepfakes improves their results only slightly.
“Increasing listener awareness by providing examples of speech deepfakes only improves results slightly.” (Abstract)
- In a 2023 PLOS ONE study, listeners correctly spotted speech deepfakes only 73% of the time.
“Listeners only correctly spotted the deepfakes 73% of the time” (Abstract)