arXiv:2606.10912v1 Announce Type: cross Abstract: Deepfake speech detectors often output a single score without explaining why an audio sample is flagged, where in the signal the evidence lies, or what cues drive the decision. We propose an audio-native explainability pipeline using Integrated Gradients on time-aligned self-supervised representations to localize decision evidence over time. We apply the proposed method to three WavLM-based detectors (AASIST, CA-MHFA, SLS) on ASVspoof 5 and manually annotate the highest-attribution regions to provide a semantic meaning of the most important cue
The proliferation of sophisticated deepfake speech generation demands equally advanced and interpretable detection methods to maintain trust in digital communication.
Understanding *why* deepfake detectors flag certain audio is critical for improving their robustness, preventing adversarial attacks, and building systems that can explain their decisions to human users.
The ability to localize and explain deepfake detection evidence shifts the paradigm from black-box scores to transparent, actionable insights into model behavior and vulnerabilities.
- · AI ethicists
- · Cybersecurity firms
- · Forensic audio analysts
- · Platforms combating misinformation
- · Deepfake creators
- · Generative AI model developers (without explainability)
- · Unexplainable AI detection systems
Improved deepfake detection systems will be more effective at identifying synthetic audio, enhancing platform security.
The explainability of detection methods will lead to a deeper understanding of deepfake generation techniques, guiding defensive and offensive AI research.
Increased public trust in audio content through transparent detection, while simultaneously raising awareness of the sophistication of synthetic media.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG