Abstract:

This research focuses on integrating Speech Emotion Recognition (SER) with AI chatbots to create a system that is more emotionally intelligent and responsive. Using advanced deep learning techniques such as Convolutional Neural Networks (CNNs), the study enhances the accuracy and robustness of SER models in detecting emotions from speech. The Ryerson Audio-Visual Database of Emotional Speech and Song (RAVDESS) serves as the primary dataset, with data augmentation techniques such as noise injection, speed variation, and pitch shifting applied to improve model performance. Key features such as Mel-Frequency Cepstral Coefficients (MFCC), Mel Spectrogram, and Zero Crossing Rate (ZCR) are extracted to improve the analysis. The study uses regularization techniques, including Batch Normalization and L2 Regularization, to prevent overfitting. Experimental results show significant improvements, with the best test accuracy reaching 87.5%, outperforming previous studies. Visualized training history demonstrates the model’s learning behavior and generalization capabilities. The findings highlight the potential of SER-enhanced chatbots in applications such as customer service and mental health support by enabling empathetic interactions.