An intelligent interface for drum pattern variation and comparative evaluation of algorithms

Abstract

Drum tracks of electronic dance music pieces are a central and style-defining element. Yet,creating them can be a cumbersome task, mostly due to lack of appropriate tools and inputdevices. In this work we use a UI prototype that aims at supporting musicians to compose therhythmic patterns for drum tracks, to compare different algorithms for drum pattern variation.Starting with a basic pattern (seed pattern), which is provided by the user, a list of variationswith varying degrees of similarity to the seed pattern is generated. The variations are createdusing one of the three algorithms compared: (i) a similarity-based lookup method using arhythm pattern database, (ii) a generative approach based on a stochastic neural network, and(iii) a genetic algorithm using similarity measures as a target function. The interface visualizesthe patterns and provides an intuitive way to browse through them. User test sessions withexperts in electronic music production were conducted to evaluate aspects of the prototype andalgorithms. Additionally a web-based survey was performed to assess perceptual properties ofthe variations in comparison to baseline patterns created by a human expert. The web surveyshows that the algorithms produce musical and interesting variations and that the differentalgorithms have their strengths in different areas. These findings are further supported by theresults of the expert interviews.