Die Verwendung von hierachrischen linearen Modellen zur Audio-Feature-basierten Vorhersage von kontinuierlichen emotionalen Reaktionen auf Musik

Christoph Graefe, Hauke Egermann

Research output: Chapter in Book/Report/Conference proceedingConference contribution


Previous research indicates that prediction of emotion is possible based on linear regression models and few interval-scaled audio feature time series. We presents an analysis of a dataset collected during a live concert experiment with solo flute using a new type of linear autoregressive mixed effects models. Time series from (a) continuous ratings of subjective components (valence, arousal ), from (b) continuous measurements of peripheral arousal (skin conductance) and from (c) continuous measurements of expressive components (EMG zygomaticus and EMG corrugator) are predictable by a group of seven audio features (Tempo, RMS, Spectral Flux, Brightness, Roughness, Spectral Centroid and Melodic Contour) and the group mean of unexpectedness ratings as further predictor. These predictions can be improved significantly considering individual differences of listeners concerning their musical background (like instrumental performance or music theory experience in years) and socio-demographics (like gender). The results are expected to contribute to a deeper exploration of the mechanisms of musical emotions and could be of interest for music therapy or to raise musical fit in audio branding and to improve the use of music for advertising purposes.
Original languageGerman
Title of host publicationFortschritte der Akustik
Subtitle of host publicationDAGA 2014
PublisherDeutsche Gesellschaft für Akustik
Number of pages2
ISBN (Print)9783939296065
Publication statusPublished - 2014

Cite this