Matt Prockup

Music, Machine Learning, Interactive Systems

I am currently a scientist at Pandora working on methods and tools for Music Information Retrieval at scale. I received my Ph.D. in Electrical Engineering from Drexel University. My research interests span a wide scope of topics including audio signal processing, machine learning, and human computer interaction. I am also an avid percussionist and composer, having performed in and composed for various ensembles large and small. I’ve also studied floral design and making wheel thrown ceramics.

Percussion Articulation Dataset

Download Here ! 

In domains outside of percussion, there exist large datasets that can be used for expressive performance analysis. A comprehensive, well-labeled set of expressive percussion samples is less common. The presented work makes use of a newly recorded dataset that encompasses a vast array of percussion performance expressions on a standard four piece drum kit. In the context of the presented work so far, only the snare drum, rack tom, and floor tom samples are used.

This subset includes 1804 individual examples across the four articulations over the three drums. Additionally, there are at least 4 examples of each expressive combination. Recordings include samples with the snare wires both touching (snares on) and not touching (snares off) the bottom head of the snare drum. The division of sample variety is not completely uniform across the entire set, but it was designed to allow for the most complete coverage of each instrument’s expressive range. That being said, no one combination of expressive parameters vastly outweighs another and all are adequately represented.

DatasetFig.png

Figure 1. This is a simple outline parameters throughout a portion of the dataset. Stick heights, stroke intensities, articulations, and head position are all varied. 

The full dataset also includes a complete array of expressive bass drum, hi-hat, and cymbal samples as well. Each articulation example has monophonic and stereo versions with multiple mixes using direct (attached) and indirect (room) microphone positioning techniques.