Automatic speech recognition and dialect identification for Arabic and Saudi dialects: a systematic literature review

This paper presents a systematic literature review (SLR) of research on automatic speech recognition (ASR) and dialect identification (DID) for Arabic and Saudi dialects published between 2019 and 2025. The review aims to provide a structured synthesis of methodological practices, datasets, dialectal coverage, and evaluation trends reported in recent studies. Following a multistage SLR protocol, related papers were retrieved from major scientific databases, screened using predefined inclusion and exclusion criteria, and analyzed among multiple dimensions, including modeling approaches, linguistic resources, acoustic features, evaluation metrics, and reported challenges. In total, 82 peer-reviewed studies met the eligibility criteria and were categorized into four methodological approaches: machine learning (ML), deep learning (DL), hybrid, and transfer learning. The reviewed literature indicates a clear shift from traditional ML techniques toward DL and self-supervised architectures, coinciding with the emergence of new Saudi-specific speech and text corpora. However, the analysis also highlights persistent challenges such as dialectal imbalance, limited availability of large-scale Saudi speech datasets, inconsistent evaluation protocols, and restricted generalizability across dialects. Emerging research directions—including multilingual transformer models, self-supervised pretraining, data augmentation strategies, and multimodal integration—represent promising opportunities for addressing these challenges. Overall, this review provides a structured and up-to-date synthesis of recent studies, identifies methodological and empirical gaps, and provides evidence-based insights to inform future dataset development, model design, and benchmarking efforts in Arabic and Saudi dialect ASR.