Abstract

Rhythmic sight-reading forms a barrier to many musicians' progress. It is difficult to practice in isolation, as it is hard to get feedback on accuracy. Different performers have different starting skills in different styles so it is hard to create a general curriculum for study. It can be boring to rehearse the same rhythms many times. We examine theories of motivation, engagement, and fun, and draw them together to design a novel training system, RhythmTrain. This includes consideration of dynamic difficultly, gamification and juicy design. The system uses machine learning to learn individual performers' strengths, weaknesses, and interests, and optimises the selection of rhythms presented to maximise their engagement. An open source implementation is released as part of this publication.