Large-scale multi-dialect Arabic speech dataset — 3,000+ hours across 13 Arabic dialects — for training and evaluating dialectal Arabic ASR systems. Part of the Lahgtna (لهجتنا) project for dialect-aware Arabic speech AI.

Dataset Summary

~611K utterances / 3,000+ hours of transcribed dialectal Arabic speech **13 Arabic dialects **, labeled per utterance 16 kHz mono audio Transcripts written in authentic dialectal orthography (not… See the full description on the dataset page: https://huggingface.co/datasets/oddadmix/dialectal-arabic-lahgtna-v2.