A shared task on five speech-processing challenges for the Arab world’s living dialects , including recognition, identification, synthesis, translation, and spoken understanding.
Arabic speech systems still struggle: Many current systems perform reasonably well on Modern Standard Arabic (MSA) and clean speech from dominant dialects, but degrade under realistic conditions such as background noise, low-bandwidth audio, mixed dialects, and sub-country regional variation.
Building on NADI 2025’s spoken dialect work, NADI 2026 broadens the focus with three main speech-focused families of tasks: robust dialectal ASR for real-world conditions, spoken dialect identification under cross-domain conditions, and dialectal Arabic text-to-speech as a new generative component — alongside speech translation and spoken language understanding.
Together, these tasks benchmark discriminative and generative Arabic speech systems to advance robust, inclusive technologies reflecting the Arab world’s linguistic diversity.
Each task ships with new blind test data. Baselines, evaluation scripts, and submission links released with the data on June 16, 2026.
Rankings appear after the final results release on July 30, 2026. Until then, follow individual task pages on CodaBench and Hugging Face Spaces for ongoing submissions.
Final results will be released July 30, 2026.
Combined NADI shared-task milestones and ArabicNLP 2026 conference deadlines. Source: arabicnlp2026.sigarab.org.
| Date | Milestone | Status |
|---|
Fill out the registration form to receive access to training and development data, baseline systems, and submission links.
Registration form →Submit via CodaBench for ASR & SID where suitable, and Hugging Face Spaces for large TTS audio submissions. SID hidden test runs through a private platform.
View baselines →Submit a system description paper by August 22. Document external data, pretrained models, preprocessing, and decoding settings clearly.
Paper guidelines →For questions contact the organizers at nadisharedtask@gmail.com.
Registration opens May 16, 2026. Training and development data, baseline systems, and evaluation scripts land June 16. Blind test data ships July 20.