Speech Emotion Recognition: A Multimodal Approach Using Multi-Feature Fusion and Self-Attention

As human-computer interaction continues to develop, Speech Emotion Recognition (SER) is a vital component for machines to gain insight into human emotions for applications in fields such as medical services, educational purposes, and social computing. Variations in speech factors (for example, intonation, intensity, and culture) can create notable challenges in emotion recognition across speech. This study proposes a robust deep learning framework that includes steps for feature extraction, sequence modeling, and attention mechanisms, which can improve accuracy and generalization. The proposed use of timeand frequency-domain features, as well as salience approaches, outperforms traditional methods using large datasets, and these findings will increase the use of adaptive and emotional Artificial Intelligence (AI) to recognize emotions in a more effective way across several contexts.