ALADAN Low-Resource Arabic Dialectal Speech Translation

This research article presents the ALADAN team's submission for the IWSLT 2025 conference, focusing on low-resource North Levantine Arabic to English speech translation. Building on previous work, the study introduces a multi-dialect ASR model using Zipformer-Large and a crowd-sourced parallel corpus covering five Arabic dialects. The findings demonstrate the feasibility of using crowd-sourced data in low-resource scenarios, achieving significant improvements in translation metrics.