Presentation on Attack Phase Descriptors at the Sound and Music Computing Conference
The 14th Sound and Music Computing Conference is organised at Aalto University, Espoo, Finland 5th-8th July 2017. Kristian Nymoen, Anne Danielsen and Justin London presents a paper on describing the attack of short musical events.
Comparing perceptual results for the perceptual attack time of a piano sound to the default estimation of attack time by the MIRtoolbox (red) and the Timbre toolbox (green).
The attack phase of sound events plays an important role in how sounds and music are perceived. Several approaches have been suggested for locating salient time points and critical time spans within the attack portion of a sound, and some have been made widely accessible to the research community in toolboxes for Matlab. While some work exists where proposed audio descriptors are grounded in listening tests, the approaches used in two of the most popular toolboxes for musical analysis have not been thoroughly compared against perceptual results. This article evaluates the calculation of attack phase descriptors in the Timbre toolbox and the MIRtoolbox by comparing their predictions to empirical results from a listening test. The results show that the default parameters in both toolboxes give inaccurate predictions for the sound stimuli in our experiment. We apply a grid search algorithm to obtain alternative parameter settings for these toolboxes that align their estimations with our empirical results.