Human Detection of Political Speech Deepfakes across Transcripts, Audio, and Video

TL;DR

The paper investigates how well people can distinguish real from fabricated political speeches across transcripts, audio, and video, revealing how multimedia cues influence discernment. It uses five pre-registered randomized experiments with 2,215 participants and varied stimuli from the Presidential Deepfake Dataset (PDD) and Barari et al. materials to examine modality effects, audio sources (voice actor vs. text-to-speech), base-rate misinformation, and context. Key findings show that audiovisual information substantially improves accuracy, with audio-based deepfakes being harder to detect than voice-acted ones, and that base-rate manipulation generally has a limited impact on overall accuracy. The results offer nuanced insights into the seeing-is-believing heuristic, inform content moderation strategies by highlighting what components are manipulated, and underscore the role of perceptual cues in misinformation discernment in political contexts.

Abstract

Recent advances in technology for hyper-realistic visual and audio effects provoke the concern that deepfake videos of political speeches will soon be indistinguishable from authentic video recordings. The conventional wisdom in communication theory predicts people will fall for fake news more often when the same version of a story is presented as a video versus text. We conduct 5 pre-registered randomized experiments with 2,215 participants to evaluate how accurately humans distinguish real political speeches from fabrications across base rates of misinformation, audio sources, question framings, and media modalities. We find base rates of misinformation minimally influence discernment and deepfakes with audio produced by the state-of-the-art text-to-speech algorithms are harder to discern than the same deepfakes with voice actor audio. Moreover across all experiments, we find audio and visual information enables more accurate discernment than text alone: human discernment relies more on how something is said, the audio-visual cues, than what is said, the speech content.

Figures (11)

• Figure 1: Screenshots of the Experiments' Stimuli Frames from the 10th second in the 32 videos from the Presidential Deepfakes Dataset (PDD) and the 12 other videos used in Barari et al 2021. The left column shows the real videos, and the right column shows deepfakes. The top 4 rows show frames from the PDD videos and the bottom 2 rows show frames from the other videos.
• Figure 2: Accuracy Distinguishing Real and Fabricated Speeches across Video Stimuli Accuracy across the original Presidential Deepfakes Dataset (PDD) video stimuli in Experiment 1a, the enhanced PDD video stimuli in Experiment 2, and the non-PDD video stimuli in Experiment 2. The error bars represent 95% confidence intervals.
• Figure 3: Accuracy Distinguishing Real and Fabricated Speeches across Media Modalities in Experiment 1 A. Accuracy across all permutations of text, audio, and video in Experiment 1a with 501 recruited participants. B. Accuracy across all permutations of text, audio, and video in Experiment 1b with 41,313 non-recruited participants. The error bars represent 95% confidence intervals.
• Figure 4: Average Accuracy and Confidence for All Video Stimuli in Experiment 2Scatter plot showing participants' mean accuracy and confidence on each of the 60 videos in experiment 2. "PDD" indicates videos from the enhanced Presidential Deepfake Dataset and "Other" indicates videos are drawn from the same sample used in Barari et al 2021. "VA" indicates voice actor deepfakes and "TTS" indicates text-to-speech deepfakes.
• Figure 5: Accuracy Distinguishing Real and Fabricated Speeches across Media Modalities and Base Rates in Experiment 3 A. Accuracy across all permutations of text, audio, and video and high and low-base rate conditions in Experiment 3. The error bars represent 95% confidence intervals. B. Low base rate condition in Experiment 3: Accuracy across on the four modalities. C. High base rate condition in Experiment 3: Accuracy across on the four modalities.