AI eller Människa – Kan vi höra skillnaden? : En kvantitativ studie om människors förmåga att skilja påAI‐genererat tal och mänskligt tal

Sammanfattning

As AI technology makes it possible to easily clone voices, the riskof fraud and misinformation increases, placing new demands onpeopleʹs ability to distinguish AI‐generated speech from humanspeech. Based on this, the study investigates people’s ability todifferentiate between AI‐generated speech and human speech. Thestudy was conducted with a quantitative approach where 450participants listened to a number of voice clips and were asked toanswer whether they thought the voice was AI‐generated orhuman. The analysis was conducted using Signal Detection Theory(SDT) to measure sensitivity (d´) and bias (c). The results show thatparticipants found it difficult to make correct judgments, with aclear tendency to perceive voices as human regardless of theiractual origin. Neither gender, age, geographic location, nor thehabit of listening to audiobooks had any statistically significantimpact on participants’ ability to identify AI‐generated speech. Inconclusion, the study shows that people are no better than chanceat determining whether a voice is AI‐generated or human