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 .

Polynomial Extrapolation for Prediction of Surprise Based on Loudness - A Preliminary Study

Title Polynomial Extrapolation for Prediction of Surprise Based on Loudness - A Preliminary Study
Publication Type Conference Paper
Year of Publication 2009
Conference Name Sound and Music Computing Conference
Authors Purwins, H. , Holonowicz P. , & Herrera P.
Conference Start Date 23/07/2009
Conference Location Porto
Abstract The phenomenon of music surprise can be evoked by various musical features, such as intensity, melody, harmony, and rhythm. In this preliminary study we concentrate on the aspect of intensity. We formulate surprise as a critical derivation from the predicted next intensity value, based on the 'immediate' past (around 7 s), slightly longer than the short-term memory. Higher level cognition, processing the long range structure of the piece and general stylistic knowledge, is not considered by the model. The model consists of a intensity calculation step and a prediction function. As a preprocessing method we compare instantaneous energy (root mean square), loudness, and relative specific loudness. This processing stage is followed by a prediction function for which the following alternative implementations are compared with each other: 1) discrete temporal difference of intensity functions, 2) FIR filter, and 3) polynomial extrapolation. In addition, we experimented with different analysis window length, sampling rate and hop size of the intensity curve. Good results are obtained for loudness and polynomial extrapolation based on an analysis frame of 7 s, a sampling rate of the loudness measures of 1.2 s, and a hop size of 0.6 s. In the polynomial extrapolation a polynomial of degree 2 is fitted to the loudness curve in the analysis window. The absolute difference between the extrapolated next loudness value and the actual value is then calculated and divided by the standard deviation within the analysis window. If the result is above a threshold value we predict surprise. The method is preliminarily evaluated with a few classical music examples.
preprint/postprint document files/publications/PurwinsEtAlSuprLoudPredictionSMC09_0.pdf