Validation study demonstrates over 95% emotion recognition accuracy from electroencephalograms in the SEED dataset, suggesting viable brain-computer music generation.
Key Points
To develop and evaluate an EEG-Conformer artificial intelligence model that detects emotional states from brainwave signals and translates them into theory-guided prompts for automated music generation.
Analyzed electroencephalogram (EEG) frequency bands and electrode distributions using a deep learning EEG-Conformer architecture to classify positive, neutral, and negative emotions.
Evaluated model performance on the benchmark SEED electroencephalogram emotion dataset.
Engineered a prototype pipeline converting emotion classification outputs into text prompts to drive an AI-based music-generation system.
Achieved an emotion classification accuracy exceeding 95% on the benchmark SEED dataset.
Visual analyses confirmed the architecture successfully identified neurophysiologically meaningful brain activity patterns corresponding to distinct emotional states.