Facial emotion recognition is pivotal for advancing human–computer interaction, with applications in psychology, adaptive learning systems, and personalized AI experiences. This paper presents a novel deep learning framework for real-time facial emotion recognition, integrating convolutional neural networks (CNNs) for spatial feature extraction and recurrent layers (LSTMs/GRUs) to capture temporal dynamics in facial expression sequences. Implemented using TensorFlow, our model is trained on the FER-2013 dataset with preprocessing techniques such as normalization and data augmentation, and optimized via gradient-based methods like Adam. To ensure robustness, we incorporate hyperparameter tuning and dropout layers to mitigate overfitting. The proposed framework achieves real-time performance with a processing speed of 144 frames per second, while attaining an accuracy of 92.14% and an F1-score of 0.89 on FER-2013, demonstrating competitive performance against existing methods. Our main contributions include (1) a scalable deep learning architecture that balances real-time processing with high accuracy, (2) the integration of spatial and temporal features for enhanced emotion recognition, and (3) a comprehensive evaluation demonstrating robustness across diverse datasets. These advancements bridge the gap between human emotions and computational understanding, paving the way for more empathetic and responsive AI systems.