Workshop Room: To be announced
Workshop Room: To be announced
Embedded Files
The workshop program is now online!
08:00 – 09:00
08:00 – 09:00
Registration, Coffee, Networking
Registration, Coffee, Networking
Workshop Session # 1
Workshop Session # 1
09:00 – 09:10
09:00 – 09:10
Welcome by Workshop Organisers
Welcome by Workshop Organisers
09:10 – 10:00
09:10 – 10:00
Keynote I
Keynote I
40 mins + 10 mins (Q/A)
40 mins + 10 mins (Q/A)
Title: On the Evaluation of Speech Anti-Spoofing and Deepfake Detection: Lessons I Learned from the ASVspoof Challenges
Speaker: Dr. Xin Wang (Associate Professor & JST PRESTO researcher, National Institute of Informatics, Japan Science and Technology Agency, Japan)
Speaker’s Bio
Speaker’s Bio
Xin Wang is a Project Associate Professor and a JST PRESTO researcher at the National Institute of Informatics, Japan. He is one of the organizers of the last three ASVspoof Challenges. He is also a member of the organizing team for the last three VoicePrivacy Challenges, an international joint effort on speech privacy.
Abstract
Abstract
Detecting synthetic speech is important. The community has long evaluated detectors using various types of error rates, including the equal error rate in past ASVspoof Challenges. These metrics measure how well a detector discriminates synthetic from real speech, but they assume an oracle threshold that users do not have, and they say nothing about whether the scores can be trusted at a threshold set in advance. This property is known as calibration. This talk reviews the evaluation metrics, links them to existing theoretical research, and shares findings from the latest ASVspoof 5 challenge. It then extends the discussion to speaker anonymization, where the equal error rate remains the dominant metric, and asks whether it is adequate there.
10:00 – 10:30
10:00 – 10:30
Coffee Break
Coffee Break
Workshop Session # 2
Workshop Session # 2
Paper Presentations
Paper Presentations
Deepfake Detection, Generalization, and Forensic Analysis
Deepfake Detection, Generalization, and Forensic Analysis
Each Paper: 15 mins + 5 mins (Q/A)
Each Paper: 15 mins + 5 mins (Q/A)
10:30 – 10:50
10:30 – 10:50
CertAV: Manifold-Aware Certified Robustness for Audio-Visual Deepfake Detection
Palak Parmar (Pandit Deendayal Energy University – PDEU; LDRP Institute of Technology and Research); Chintan Bhatt (AI-EEECS, Queen’s University Belfast); SantoshKumar Bharti (Pandit Deendayal Energy University – PDEU); Tanvish Desai (LDRP Institute of Technology and Research)
10:50 – 11:10
10:50 – 11:10
STONE: Semantic Transformation-Oriented Neural Editing Dataset for Benchmarking Deepfake Detector Generalization
Eunbi Hwang, Hyunseok Lee, Yunseo Lee, Giwon Kang, Taekyoung Kwon (Yonsei University)
11:10 – 11:30
11:10 – 11:30
Training-Free Detection of AI-Generated Images Based on Coherence Analysis of Image Representations
Ji Li, Kai Wang (GIPSA-lab, CNRS, Grenoble INP, Univ. Grenoble Alpes)
11:30 – 11:50
11:30 – 11:50
Beyond Known Speakers: Enhancing Spoofing Detection Generalization with VIVED
Wenbin Wang, Sanjay Jha (University of New South Wales); Xing Zi, Mukesh Prasad, Justin Lipman (University of Technology Sydney)
11:50 – 12:10
11:50 – 12:10
Charting Limitations and Research Routes in Forensic Synthetic Speech Examination
Carolina Lins Machado, Tina Cambier-Langeveld, Willemijn Heeren (Netherlands Forensic Institute)
Moderated Discussion
Moderated Discussion
Beyond Detection: Generalization, Robustness, and Forensic Evidence
Beyond Detection: Generalization, Robustness, and Forensic Evidence
12:10 – 12:30
12:10 – 12:30
Discussion topics:
Discussion topics:
How can deepfake detection systems generalize to unseen generative models?
Are existing robustness benchmarks representative of realistic adversarial conditions?
What evidence is needed to move from detection results to defensible forensic conclusions?
12:30 – 13:30
12:30 – 13:30
Lunch Break
Lunch Break
Workshop Session # 3
Workshop Session # 3
Session Chair: Dr Nadeem Ahmed, University of New South Wales, Australia
Session Chair: Dr Nadeem Ahmed, University of New South Wales, Australia
13:30 – 14:20
13:30 – 14:20
Keynote II
Keynote II
40 mins + 10 mins (Q/A)
40 mins + 10 mins (Q/A)
Title: Fake Realities? Trust, Bias, and the Limits of AI Disclosure
Speaker: Dr. Abdallah El Ali (Centrum Wiskunde & Informatica, Utrecht University, Netherlands)
Speaker’s Bio
Speaker’s Bio
Abdallah “Abdo” El Ali is a research scientist at Centrum Wiskunde & Informatica (CWI) in Amsterdam, and part-time Assistant Professor at Utrecht University. He leads the research areas on Affective Interactive Systems and Trustworthy Human-AI Interaction, with a focus on human augmentation across physical and digital environments and AI transparency. He is affiliated with the AI, Media, and Democracy Lab, and serves as Chair of CHI Nederland (CHI NL), the ACM SIGCHI Netherlands Chapter.
Abstract
Abstract
How should we disclose our use of AI? As AI-generated and AI-mediated content becomes increasingly indistinguishable from human-produced information, disclosure is becoming central to efforts to address deception and disinformation. Yet knowing how content was produced does not establish whether it is trustworthy or true. Against the backdrop of the EU AI Act’s transparency obligations, I present a series of studies across news and health showing how AI disclosures can bias people’s perceived trust, behavior, and physiological responses. I further show how source labels can bias large language models used as evaluators (LLM-as-a-judge), affecting both trust assessments and factual truth judgments. Together, these findings highlight the limits of static “AI-generated” labels and motivate more nuanced approaches to AI transparency. I conclude by asking what, when, and how we should disclose as AI becomes increasingly integrated into human action and expression.
Paper Presentations
Paper Presentations
AI Voice Security
AI Voice Security
Each Paper: 15 mins + 5 mins (Q/A)
Each Paper: 15 mins + 5 mins (Q/A)
14:20 – 14:40
14:20 – 14:40
Did You Hear That? Detecting AI Voice Agents by Exploiting Their Robustness
Ariel Bitan, Danny Hendler, Yisroel Mirsky (Ben-Gurion University)
14:40 – 15:00
14:40 – 15:00
Active Acoustic Probes for AI Voice Call Detection
Lior Yasur, Yisroel Mirsky (Ben-Gurion University of the Negev)
15:00 – 15:30
15:00 – 15:30
Coffee Break
Coffee Break
Workshop Session # 4
Workshop Session # 4
Paper Presentations
Paper Presentations
Forensics, Attribution, Disinformation, and Mitigation
Forensics, Attribution, Disinformation, and Mitigation
Each Paper: 15 mins + 5 mins (Q/A)
Each Paper: 15 mins + 5 mins (Q/A)
15:30 – 15:50
15:30 – 15:50
Explaining Suspicious Speech Beyond Binary Deepfake Detection: A Five-Module Proposition-Centered Evidence Profile
Thien-Phuc Doan, Woongjae Lee, Inho Kim, An Thien Nguyen, Souhwan Jung (Soongsil University)
15:50 – 16:10
15:50 – 16:10
Beyond Deepfake Detection: MOSAIC, a Confidence-Aware and Counter-Deception Framework for Synthetic-Media Operation Attribution
Yuval Sinay (G. S. Rakovski National Defence College)
16:10 – 16:30
16:10 – 16:30
Countering Disinformation Campaigns and Narratives: An Innovative Graph-Based Solution for Malicious Content Analysis, Detection, and Visualization
Rafał Kozik, Marek Pawlicki, Michał Choraś (IDEAS Research Institute; Bydgoszcz University of Science and Technology)
16:30 – 16:50
16:30 – 16:50
Reverse Prompt Engineering in Generative AI: A Baseline for Deepfake, Deception and Disinformation Mitigation
Jongkil Jay Jeong (TNK2 Pty Ltd); Ashish Nanda (Deakin University; TNK2 Pty Ltd); Sai Gautam Mandapati (TNK2 Pty Ltd; Deakin University); Fokke Heikamp, Robin Doss (Deakin University)
16:50 – 17:00
16:50 – 17:00
Closing Remarks
Closing Remarks
Page updated
Google Sites
Report abuse