Human Perception of Audio Deepfakes

Authors:
Nicolas M. Müller, Karla Pizzi, Jennifer Williams

Where published:
Proceedings of the 1st International Workshop on Deepfake Detection for Audio Multimedia (DDAM ’22)

Published:

10 October, 2022

 

Dataset names (used for):

  • ASVspoof 2019: A large-scale public database of synthesized, converted, and replayed speech

 

Some description of the approach:
The dataset is used in a gamified online experiment where participants distinguish between real and fake audio samples. It includes both bona-fide and deepfake audio samples, with users’ classifications and AI model predictions recorded.

 

Some description of the data:
The dataset includes 13,229 game rounds played by 410 unique users. It captures user demographics (IT experience, age, native language) and AI model predictions.

 

Keywords:
Deepfake, Human perception, Audio spoofing

Instance Represent:
Audio samples classified by users and AI for authenticity

Dataset Characteristics:
13,229 game rounds, user demographics, and AI model predictions. Audio recordings are used, with user classifications and AI assessments for authenticity.

Subject Area:
Cybersecurity, Human-Computer Interaction

Associated Tools:
Deepfake detection, human vs. machine comparison

Feature Type:
Audio recordings, user demographics

Main Paper Link


License: Not specified


Last Accessed: 7/7/2024

NSF Award #2346473