Bi-CALM for RAVDESS: Convolutional Attention Bidirectional LSTM Model for Speech Emotion Recognition

Speech emotion recognition (SER) poses significant challenges within human-computer interaction, primarily because of complex emotional expressions and signal variability. This paper introduces Bi-CALM for RAVDESS, a novel architecture that integrates convolutional neural networks for feature extraction, bidirectional LSTMs for temporal context processing, and an attention mechanism to highlight emotion-salient features. Our approach addresses key limitations in existing SER systems by optimizing the network structure to balance performance and generalization while mitigating overfitting. Experimental validation on the RAVDESS dataset demonstrates the effectiveness of our proposed model, achieving 91.5% test accuracy—significantly outperforming conventional approaches. The Bi-CALM architecture provides a comprehensive framework designed to identify emotional states from Speech, presenting promising use-cases in areas such as psychological health assessment, customer service, and affective computing applications.