AI Voice Clones Can Sound As Real As Human Speech, PLOS One Study Finds

Overview

• Researchers at Queen Mary University of London published the peer-reviewed findings in PLOS One on September 24, 2025.
• Experiments compared real human recordings with two synthetic types: voice clones of specific speakers and voices sampled from a large model’s latent space.
• Listeners frequently struggled to tell cloned voices from human recordings, with overall realism matching human speech.
• AI-generated voices were rated more dominant than human voices, and some were judged more trustworthy.
• The team created convincing clones using commercially available software with minutes of audio, little expertise, and negligible cost, raising concerns over fraud, impersonation, copyright, and security alongside potential benefits for accessibility and education.