The E-Drum A Case Study for Machine Learning in New Musical Controllers

This paper describes a system of drum gesture acquisition via machine learning methodologies. Discussions of techniques, advantages, and previous work are discussed in the context of evaluating a new controller for musical expression. An evaluation of the machine learning mechanism is provided in the form of statistical results of classification tests. An informal evaluation of the effectiveness of the system for percussionists is provided in the form of performance reports of the first author’s use of the system. 1 Background in Music Performance The rate of advance in Music Technology has increased enormously, yet commercial electronic percussion has been stagnant for approximately 20 years. There is not even an entry for it in the standard book Percussion Instruments and their Histories Blades (1997). Commercial electronic percussion hardware has not improved and the sounds have only evolved to imitate sounds that have already become popular. Current percussion controllers only provide data on the velocity of the impact, forcing a single dimension of data to represent something as complex as musical gestures.