.

Voice deepfakes have become an increasingly serious threat to digital society. Potential risks range from impersonation, privacy violations, vishing and financial fraud to misinformation, social engineering, and attacks against voice-based authentication systems.

At the same time, advances in generative AI are making synthetic speech increasingly sophisticated. Modern systems can clone voices using only small amounts of reference audio and generate speech that can be difficult to distinguish from genuine human voices. As these technologies continue to evolve, reliable deepfake detection is becoming an increasingly challenging research problem.

In his keynote, Prof. Sadeghi presented ongoing research by the System Security Lab on voice-deepfake detection, including efforts to address new generations of synthetic speech and increasingly advanced attack scenarios.

The keynote also highlighted the emerging arms race between powerful generative AI systems and reliable detection mechanisms. Maintaining trust in digital audio will require new approaches that go beyond detecting artifacts specific to individual generations of voice-synthesis models.

The talk was followed by engaging discussions with researchers from the security, artificial intelligence, and signal-processing communities.

We thank Prof. Ali Ghorbani for the invitation and the organizers and chairs, including Prof. Paria Shirani and Prof. Suryadipta Majumdar, for organizing and hosting the event.