Study: Most people can't distinguish AI voices from humans

New study reveals AI voice clones are indistinguishable from human voices, raising concerns about misuse.

Why it matters

  • AI-generated voices are becoming increasingly realistic, making it hard for listeners to distinguish them from human voices.
  • This has implications for both beneficial uses and potential misuse.

By the numbers

  • 50 participants in each of two experiments.
  • Voice clones were labeled as "human" as often as human voices (58% vs. 62%).
  • Generic AI-generated voices were labeled as "human" 41% of the time.

The big picture

  • AI-generated voices are increasingly realistic, with voice clones being particularly convincing.
  • The study did not find a hyperrealism effect for voices, unlike with AI-generated faces.
  • The findings highlight both the potential benefits and risks of AI-generated voices.

What they're saying

  • Comments mention specific cues that help identify AI voices, such as odd intonation and lack of variation in pronunciation.
  • One comment humorously notes that people might start questioning if they're talking to a robot.

Caveats

  • The study was conducted with a specific set of voices and participants, so results may vary in other contexts.
  • The study did not find a hyperrealism effect for voices, but some individual voice clones were perceived as more real than their human counterparts.

What’s next

  • Future research could explore whether further advancements in AI voice generation could achieve hyperrealism.
  • Studies could also look into how different demographics perceive AI-generated voices and how familiarity with specific voices affects perception.