Beyond the Beat: Towards Metre, Rhythm and Melody Modelling with Hybrid Oscillator Networks

In this paper we take a connectionist machine learning approach to the problem of metre perception and learning in musical signals. We present a hybrid network consisting of a nonlinear oscillator network and a recurrent neural network. The oscillator network acts as an entrained resonant filter to the musical signal. It ‘perceives’ metre by resonating to the inherent frequencies within the signal. The neural network learns the long-term temporal structures present in the signal. We show that our hybrid network outperforms previous approaches of a single layer recurrent neural network in melody prediction tasks. By perceiving metrical structure, our system is enabled to model more coherent long-term structures, and can be used in a multitude of analytic and generative scenarios, including live performance applications.