Enhancing Predictive Models of Music Familiarity With EEG: Insights From Fans and Non-Fans of K-Pop Group NCT127

Predicting a listener's experience of music based solely on audio features has its limitations due to the individual variability in responses to the same music. This study examines the effectiveness of electroencephalogram (EEG) in predicting the subjective experiences while listening to music, including arousal, valence, familiarity, and preference. We collected EEG data alongside subjective ratings of arousal, valence, familiarity, and preference from both fans (N=20) and non-fans (N=34) of the K-pop idol group, NCT127 to investigate response variability to the same NCT127 music. Our analysis focused on determining whether the inclusion of EEG alongside audio features could enhance the predictive power of linear mixed-effect models for these subjective ratings. Specifically, we employed stimulus-response correlation (SRC), a recent approach in neuroscience correlating stimulus features with EEG responses to the ecologically valid stimuli. The results showed that familiarity and preference was significantly higher in the fan group. Furthermore, the inclusion of SRC significantly enhanced the prediction of familiarity compared to models based solely on audio features. However, the impact of SRC on predictions of arousal and valence exhibited variation depending on the correlated audio features, with certain SRCs improving predictions while others diminished them. For preference, only a few SRCs negatively affected model performance. These results suggest that correlations of EEG responses and audio features can provide information of individual listeners' subjective responses, particularly in predicting familiarity.