Event Details

Advances in AI-generated speech make it nearly indistinguishable from real voices, enabling new cybercrimes where AI mimics a target's voice to deceive listeners. This talk explores if familiarity with the speaker (like a loved one or celebrity) helps listeners detect fake voices and how other factors influence detection.

Recent technological advancements mean that AI-generated speech has rapidly advanced in quality, to the point where it can be virtually indistinguishable from genuine human speech. This opens the door to a new breed of cybercrime: one which utilises AI to create fraudulent representations of a target’s speech in an attempt to deceive listeners into believing that they are genuine. Presently, curbing this deception is difficult, since the conditions dictating accurate discrimination between genuine and synthetic speech are little understood. This talk examines one factor: a potential victim’s familiarity with the speaker’s voice. When the speaker’s voice is well-known to a listener (i.e. a celebrity or loved one), does this impact the listener’s ability to recognise an AI-generated sample of that speaker’s voice? If so, why? What other factors may also be at play, and how might those factors interact with familiarity to influence the potential victim’s performance? With the insights gained by addressing these questions, we begin to give ourselves the tools to mitigate against AI-mediated cybercrime.