Enhanced Egyptian Arabic Speech Emotion Recognition

Speech Emotion Recognition (SER) enhances human-computer interaction but remains underexplored for Arabic dialects. We introduce EmoEgy, a novel Egyptian-Arabic emotional speech dataset recorded under controlled conditions with balanced speaker demographics. Using Emotion2Vec, a transformer-based, self-supervised speech representation model, paired with a lightweight Support Vector Machine (SVM) classifier, we recognize four core emotions—Angry, Happy, Neutral, and Sad (AHNS)—with 91.0% accuracy using 10-fold cross-validation. Our approach outperforms prior Egyptian-Arabic models and contributes to ongoing advances in AI, affective computing, and speech signal processing. This work represents a step toward building equitable AI systems for low-resource languages and preserving linguistic and cultural diversity in the digital era.