Thaka at KSAA-2026 Task 2: Regularized Fine-Tuning for Arabic Speech Diacritization
For low-resource Arabic speech diacritization, this work provides a practical regularization recipe that achieves state-of-the-art results under strict data constraints.
The authors won Task 2 of the KSAA-2026 Shared Task by fine-tuning CATT-Whisper with regularization techniques (R-Drop, Optuna-tuned hyperparameters, Focal Loss) and Monte Carlo Dropout averaging, achieving 23.26% WER and 1st place.
We describe the winning system for Task 2 of the KSAA-2026 Shared Task on Arabic Speech Dictation with Automatic Diacritization. The task requires producing fully diacritized Arabic text from speech audio and undiacritized transcripts, with only 2,327 training samples available and no external data permitted. Our system fine-tunes CATT-Whisper, a character-level multimodal model combining a pretrained CATT text encoder with a frozen Whisper speech encoder. The key to our approach is training regularization: R-Drop consistency regularization, Optuna-optimized hyperparameters with high weight decay, and Focal Loss. At inference, we average 200 stochastic forward passes across four model checkpoints using Monte Carlo Dropout at the softmax probability level. The system achieves 23.26% WER on the primary leaderboard metric (with case endings, including no-diacritic positions), placing 1st among all participants.