Berkeley researchers test AI voice cloning detection

Researchers and tests indicate that distinguishing AI-cloned voices from human speech has become increasingly difficult as cloners now intentionally add breaths, pauses, and disfluencies. While traditional human listeners struggle to spot the fakes, audio forensics models analyze subtle waveform features such as jitter and shimmer to detect AI generation. Additionally, some technology companies are embedding inaudible watermarks directly into audio files to help verify their authenticity. Open article →