MUSE: A Music Conducting Recognition System

In this paper, we introduce Music in a Universal Sound Environment(MUSE), a system for gesture recognition in the domain of musical conducting. Our system captures conductors’ musical gestures to drive a MIDI-based music generation system allowing a human user to conduct a fully synthetic orchestra. Moreover, our system also aims to further improve a conductor’s technique in a fun and interactive environment. We describe how our system facilitates learning through a intuitive graphical interface, and describe how we utilized techniques from machine learning and Conga, a finite state machine, to process inputs from a low cost Leap Motion sensor in which estimates the beats patterns that a conductor is suggesting through interpreting hand motions. To explore other beat detection algorithms, we also include a machine learning module that utilizes Hidden Markov Models (HMM) in order to detect the beat patterns of a conductor. An additional experiment was also conducted for future expansion of the machine learning module with Recurrent Neural Networks (rnn) and the results prove to be better than a set of HMMs. MUSE allows users to control the tempo of a virtual orchestra through basic conducting patterns used by conductors in real time. Finally, we discuss a number of ways in which our system can be used for educational and professional purposes.