Music Emotion Interaction System Based on Brain-Computer Interface: A Hybrid Deep Learning and Transfer Learning Framework

Emotion recognition from brain signals is a rapidly advancing field within affective computing, particularly in applications involving real-time brain-computer interaction and adaptive music modulation. Accurately interpreting electroencephalogram (EEG) data remains challenging due to its nonlinear and high-dimensional characteristics. This study presents a hybrid deep learning framework combining Temporal Convolutional Networks (TCNs) and Capsule Networks (CapsNets) for real-time EEG-based emotion detection. The proposed architecture captures both temporal dynamics and spatial dependencies of EEG signals through dilated convolutions and dynamic capsule routing, enabling robust feature representation across valence-arousal dimensions. EEG data collected from eight healthy participants listening to emotion-evoking music clips were processed using Short-Time Fourier Transform (STFT) and Power Spectral Density (PSD) for time-frequency feature extraction. Experimental results across three sessions demonstrated a peak classification accuracy of 89.14%, outperforming traditional models such as CNN, TCN, and SVM. The model also exhibited high inter-session and inter-subject generalizability, confirming its stability and reliability. By continuously analyzing neural responses, the system adaptively modifies musical stimuli in real time to reflect the listener's emotional state. This hybrid TCN-CapsNet framework establishes an effective foundation for emotion-aware music interaction, contributing to the advancement of intelligent affective computing and real-time brain-computer interface applications.