SHAMS Benchmark Enables Audio-Grounded Evaluation for Levantine Arabic Dialects
October 2, 2026
SHAMS provides 1,300 utterances across five Levantine Arabic varieties, featuring four-tier alignment between audio, unvocalized orthography, diacritized text, and phonetic transcription. It enables standardized evaluation for ASR, diacritization, and grapheme-to-phoneme conversion in non-standardized dialects.
HOW THIS AFFECTS YOU
●
builderThis provides a dataset to improve the robustness of speech technologies for Arabic-speaking users.
●
researcherYou can now benchmark model performance on the linguistic diversity of Levantine Arabic.