Training a Text-to-Speech System for Dialectal Arabic with a Focus on the Iraqi Dialect

This research introduces a novel approach to Text-to-Speech (TTS) synthesis, focusing on the phonetic complexities of Arabic dialects, with particular emphasis on the Iraqi dialect. While existing Arabic speech corpora provide substantial coverage of Modern Standard Arabic (MSA), they fall short in capturing the phonetic richness of regional dialects. To address this gap, we utilized Nawar Halabi's Arabic Speech Corpus as a base dataset and enriched it with custom-recorded samples of the Iraqi dialect, incorporating distinctive phonemes such as گ ,ڤ ,پ ,چ ,ۆ ,ڵ ,ێ, and using the Tatweel character (ـ) as a vowel. Our approach, powered by the FastPitch model and a customized phonetiser, successfully synthesized the Iraqi dialect while also demonstrating adaptability to other Arabic dialects, including Egyptian, Khaliji, Syrian, and more. The results of this research signify a promising advancement in Arabic TTS technology, expanding its scope to authentically represent the diverse linguistic landscape of the Arabic-speaking world.