Year: 2026
Type: conference-paper
Abstract: We describe Abjad AI's submission to KSAA-2026 Shared Task 2 on automatic diacritization of Arabic speech dictation.The task requires generating fully diacritized text given speech audio and an undiacritized transcript.Because text-only diacritization cannot resolve ambiguities that are recoverable from the acoustic signal, we propose conditioning a character-level encoder-only Transformer (CATT) on speech representations.We introduce grouped speech conditioning, which downsamples speech encoder features into a small set of pooled tokens concatenated to the text input, enabling efficient fusion without architectural changes to CATT.We train with a two-phase schedule that first freezes the text encoder, then fine-tunes the full model.Our best system, using Whisper-small features with grouping size 5, achieves a Diacritization Error Rate (DER) of 5.68 and a Word Error Rate (WER) of 11.85 under the official primary test setting.Notably, we find that Whisper-small consistently outperforms Whisper-large-v3, suggesting that compact speech representations better suit this fusion setting.This is an extended and revised version of our previous work (Ghannam et al., 2025).
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7166563192