Enhancing Arabic Speech Therapy with AI: A Specialized AI-Based System for Arabic Stuttering Classification

The phenomenon of stuttering and its associated speech disorders disrupt fluency through repetition, prolongation, and delay, affecting millions of people. Although considerable progress has been made in artificial intelligence-based Automatic Speech Recognition (ASR) technology, most of the current models remain mainly designed for high resource and dominant lingua franca languages, e.g., English, and underperform for Arabic. This paper presents additional insights into stuttering disorder classification in Arabic using Whisper ASR from OpenAI and painstakingly trained to classify phonemic patterns as fluent or disfluent. The most salient aspect of this study is the construction of a Marked Stuttering Speech Database within which quota speech segments of Fluent and Disfluent speech were collected from real clinical cases. The systematic comparative framework is done between the full Whisper family (Tiny-Large) vs. Wav2Vec2.0 family (Base-XLarge) under identical conditions, which is the first benchmark for Arabic stuttering. We found that Whisper beats Wav2Vec2.0 at every scale, including the smallest variants, and it is reliable even for low-resource deployment. This Confirms that Whisper encoder is suitable for clinical Arabic speech-disorder workflows.