Note: This bibliographic page is archived and will no longer be updated. For an up-to-date list of publications from the Music Technology Group see the Publications list .

Neural and Music Correlates of Music-evoked Emotions

Title Neural and Music Correlates of Music-evoked Emotions
Publication Type Master Thesis
Year of Publication 2016
Authors Patlatzoglou, K.
Abstract One of the basic research interests in cognitive neuroscience of music, comes from the affective phenomena that take place in music. The question of how the human brain represents and organizes conceptual knowledge has been investigated by scientists in different fields and still remains an open problem. Several neuroimaging studies on music-evoked emotions, have shown distinct spatial patterns of activity that emerge from brain structures, already known to be involved in emotions. From the musicological point of view, there has been a strong tendency in the aesthetics of music to emphasize on the importance of the musical structure. Leaving aside factors such as the musical context and properties listener, two questions are addressed in this work: 1) Can we train and test a computational model tha predicts fMRI activity related to music-evoked emotions, based on acoustic features extracted from the music? 2) Which are the features most relevant to the task regarding the basic emotions of joy and fear? Using fMRI data obtained from 17 individuals during a music listening session of 24 tracks (which belong to 3 classes of joy, fear and neutral stimuli), along with the extraction of audio descriptors from music using MIR (music information retreival) tools, a machine learning approach is selected for the creation of the model. By training multiple linear regressions, a predictive relationship is achieved between the extracted musical features and the BOLD activation of fMRI images, that correspond to each stimulus-track. The cross validated accuracies of alternative models seem to depend on the various feature and voxel selection strategies. The results show the possibility of such approach, with high accuracies for specific selection strategies. Nevertheless, what should be predicted and precisely how remains a challenge in the field.
Final publication https://doi.org/10.5281/zenodo.1161287