TheMachine Press

A daily newspaper for the age of artificial intelligence.

Morning editionPermanent story

safety security

One Altered Sentence Fooled Listeners 77 Percent of the Time

In a study of 82 IT professionals, partial audio spoofs were harder to localize than fully synthetic speech.

Published Updated Story ID: mp-2026-08-23-015
Read the complete editionStory JSON

Summary

In a study of 82 IT professionals, partial audio spoofs were harder to localize than fully synthetic speech.

Strict accuracy for locating the altered sentence fell to 9 percent, and listeners marked the synthetic segment as genuine 77 percent of the time. Humans and six pretrained detectors failed in different ways, supporting procedural verification and provenance rather than listening alone.

Why it matters

In a study of 82 IT professionals, partial audio spoofs were harder to localize than fully synthetic speech.

Limits and context

No additional limitation was separately recorded.

Key claims

  1. In a study of 82 IT professionals, partial audio spoofs were harder to localize than fully synthetic speech.

    Evidence: source-2026-08-23-017

Sources

  1. arXiv preprint 2608.19959arXiv · primary research

Corrections

No corrections have been recorded for this story.