EXPLORING PERCEPTUAL AND ACOUSTICAL CORRELATES OF POLYPHONIC TIMBRE

POLYPHONIC TIMBRE HAS BEEN DEMONSTRATED TO BE an important element for computational categorization according to genre, style, mood, and emotions, but its perceptual constituents have received less attention. The work presented here comprises two experiments, Experiment 1, to devise a framework of subjective rating scales for quantifying the perceptual qualities of polyphonic timbre and Experiment 2, to rate short excerpts of Indian popular music and correlate them with computationally extracted acoustic features. A factor analysis of the ratings suggested three perceptual dimensions: Activity, Brightness, and Fullness. The present findings imply that there may be regularities and patterns in the way people perceive polyphonic timbre. Furthermore, the perceptual dimensions can be predicted relatively well by the regression models. Spectrotemporal modulations were found to be most relevant, while the well known polyphonic timbre descriptors, the Mel-Frequency Cepstral Coefficients, did not contribute significantly to any of the perceptual dimensions.