Improving Voice-Based Emotion Recognition Using Convolutional Neural Networks: Integrating RAVDESS and JL Corpus Datasets

Voice-based emotion recognition has garnered significant attention in data science, with applications in fields such as psychology, gaming, healthcare, and customer service. This study explores emotion detection using voice recordings from the JL Corpus and RAVDESS datasets, employing Convolutional Neural Networks (CNNs) to classify emotions such as neutral, calm, happy, sad, angry, fearful, disgusted, and surprised. Mel Frequency Cepstral Coefficients (MFCCs) are used for feature extraction. The CNN model achieved 73% accuracy on the RAVDESS dataset, though it underperformed compared to previous studies using multimodal data. However, combining the JL Corpus with RAVDESS improved accuracy to $85 \%$ and significantly enhanced precision and recall. Misclassifications between similar emotions highlight the limitations of amplitude-based features, suggesting the need for more advanced feature extraction techniques. This research contributes to the growing body of work in emotion recognition, emphasizing the potential for dataset integration and feature refinement to improve model performance.