Investigation of Familiarity Effects in Music-Emotion Recognition Based on EEG

Familiarity is a crucial subjectivity issue in music perception that is often overlooked in neural correlate studies and emotion recognition research. We investigated the effects of familiarity to brain activity based on electroencephalogram (EEG). In our research, we focused on self-reporting and continuous annotation based on the hypothesis that the emotional state in music experiencing is subjective and changes over time. Our methodology allowed subjects to select 16 MIDI songs, comprised of 8 familiar and 8 unfamiliar songs. We found evidence that music familiarity induces changes in power spectral density and brain functional connectivity. Furthermore, the empirical results suggest that using songs with low familiarity could slightly enhance EEG-based emotion classification performance with fractal dimension or power spectral density feature extraction algorithms and support vector machine, multilayer perceptron or C4.5 classifiers. Therefore, unfamiliar songs would be most appropriate for emotion recognition system construction.