AH-Translit: Arabic to Hindi Transliteration Dataset

AH-Translit is a large-scale Arabic-to-Hindi transliteration dataset comprising 247,654 aligned sentence pairs spanning Quranic Arabic, MSA Daily Use, and MSA Bibliographic domains. The dataset includes complete train, validation, and test splits, domain-specific subsets, balanced and proportional mixed-domain training sets splits. Arabic text covers 212 unique characters with an average sentence length of 11.74 words, while Hindi transliterations span 302 unique characters with an average sentence length of 15.25 words, enabling robust modeling of phonetic variation, domain shifts, and long-sequence transliteration behavior across both classical and modern Arabic sources.