Musical Brainwaves: Decoding Electroencephalography Signals Elicited by Musical Stimuli

The influence of music on the human brain has been recognized for centuries, but recent advancements in brain-computer interfaces (BCIs) offer unparalleled opportunities to probe its complexities. This study investigates the effectiveness of music as a stimulus for driving BCIs using electroencephalography (EEG) signals. We analyze the neural correlates of music perception and imagination in the OpenMIIR dataset, employing advanced machine learning algorithms for EEG decoding. Exploiting various feature engineering techniques in time domain and frequency domain, we leverage convolutional neural networks and speech-inspired architectures like VLAAI to reconstruct the original audio stimuli from EEG data. Our results demonstrate a significant improvement in reconstructing music stimuli from EEG signals in the time domain, achieving a Pearson correlation of 0.35 compared to the current state-of-the-art performance of 0.11. Similarly, an increase in accuracy in classifying lyrical and non-lyrical musical excerpts based on participants' EEG responses was observed, rising from 49.6% to 57%.