ArA-DF 2026 (Arabic Speech Deepfake Detection Challenge)
Yassine El Kheir
Call for Participation: ArA-DF 2026 (Arabic Speech Deepfake Detection Challenge)
We are excited to invite researchers and practitioners to participate in ArA-DF 2026, an official shared task at ArabicNLP 2026 (co-located with EMNLP 2026).
About the Task: While synthetic speech generation has seen massive improvements, robust detection of AI-generated and spoofed speech remains an open challenge, particularly for underrepresented languages. ArA-DF 2026 evaluates the reliability of deepfake detection systems on Arabic speech under challenging, real-world constraints.
Participants can compete in two tracks (Primary Metric: Equal Error Rate):
Dialect Generalization: Evaluating system reliability on zero-shot Arabic dialects and unseen speakers.
Acoustic Robustness: Testing resilience against practical audio degradations (e.g., compression, noise, re-recording).
Challenge Timeline
Data Release: June 16, 2026 (Currently Available)
Evaluation Phase: July 20 - July 25, 2026
Leaderboard Freeze: July 25, 2026
System Papers Due: August 8, 2026
Camera-Ready: August 22, 2026
Links & Registration
Challenge Website: https://arabicspeech-ara-df-2026.static.hf.space/index.html
Registration Form: https://docs.google.com/forms/d/e/1FAIpQLScnJwi998TlAUjLxIAwet8BucyajZGVPjAwvAUA5SWczPPRwg/viewform?usp=send_form
Contact: ara-d...@googlegroups.com
We look forward to your participation!
ArA-DF 2026 Organizing Committee: Vasista Sai Lodagala, Yassine El Kheir, Sara Althubaiti, Pedro Moreno Mengibar, Ahmed Ali