Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper investigates human detection of synthetic speech in a realistic socio-technical context, manipulating trust cues like instructional framing, affective priming, and provenance labeling. The study found that utterance class (authentic, fully synthetic, partially synthetic) was the primary driver of detection accuracy, with fully synthetic speech detected at below-chance levels. While trust cues didn't have main effects, they did influence detection behavior, suggesting a complex interplay between perception and context.
Humans perform worse than chance at detecting fully synthetic speech, even when primed with trust cues, highlighting a critical vulnerability in our ability to discern AI-generated audio.
Automatic deepfake detection has received considerable research attention, yet the socio-technical environment in which humans actually encounter synthetic speech remains poorly understood. We investigate voice deepfake detection as a perceptual and contextual process, presenting a localization task in which 47 participants marked suspected synthetic segments across authentic, fully synthetic, and partially synthetic utterances under three manipulated trust cues: instructional framing, affective priming, and provenance labeling. Participants provided quality ratings on mechanicalness, expressiveness, intelligibility, clarity, calmness, and confidence of evaluation. Utterance class was the primary determinant of detection accuracy and perceptual quality; trust cues produced no main effects but motivated detection behavior. Fully synthetic speech was detected at below-chance levels. Quality ratings tracked utterance type, indicating implicit discrimination where overt detection failed.