핵심 개념
Combining human intuition with machine precision enhances deepfake detection capabilities.
초록
Scammers exploit AI voice-cloning for social engineering attacks, worsened by Real-time Deepfakes (RTDFs).
Research introduces challenge-response method for deepfake audio call detection.
Evaluation shows 86% deepfake detection rate and 80% AUC score.
Human-AI collaboration boosts accuracy to 82.9%.
Challenges include vocal distortions, waveform manipulations, and language-specific articulations.
Machine evaluation achieves AUC of 86.7% with 11 effective challenges.
Human evaluation matches machine performance, with potential for improvement through collaboration.
통계
우리의 연구는 딥페이크 오디오 통화 감지를 위한 도전-응답 방법을 소개합니다.
평가 결과, 86%의 딥페이크 감지율과 80%의 AUC 점수를 보여줍니다.
인간-인공지능 협업은 정확도를 82.9%로 향상시킵니다.
인용구
"Combining human intuition with machine precision offers complementary advantages."
"Challenges include vocal distortions, waveform manipulations, and language-specific articulations."