Oscillations create groove: A reservoir model for learning complex drumming performances

Abstract Musical performances, particularly drumming, intricately balance rhythmic patterns and creative expression, all heavily influenced by the concept of “groove.” This study explored how the brain learns and internalizes complex rhythms and grooves from a computational perspective. The cerebellum and basal ganglia play key roles in rhythm perception, production, and timing. We utilized an oscillation-driven reservoir computing model, which is a simplified recurrent neural network designed for timing learning, to acquire complex rhythms. The model was trained to replicate the drumming style of Jeff Porcaro by learning his hi-hat patterns. Analyses revealed that the model’s outputs, including fluctuations in interbeat timings and amplitudes, were similar to the original drumming. Next, it was trained to generate multidimensional drum kit performances for various genres (funk, jazz, samba, and rock). Model outputs had timing deviation and audio features related to a sense of groove, similar to the original performances. By incorporating the oscillations of multiple frequencies into the reservoir, the model reproduced fluctuations of timing intervals (microtiming) and amplitudes, as well as audio features of each musical genre. Our results indicated that oscillation-driven reservoir computing can successfully replicate the timing and rhythmic complexity of professional drumming, suggesting that it is a common computational principle for motor timing and rhythm generation. Furthermore, this approach offers insights into the neural underpinnings of musical groove, shedding light on how the brain processes and reproduces intricate rhythmic patterns. Author summary Drumming is a sophisticated art form combining rhythmic precision and creative flair, encapsulated by the elusive quality of “groove.” Understanding how the brain generates these complex rhythms can provide insights into both musical and neural functions. In this study, we employed oscillation-driven reservoir computing to model the principal neural processes involved in learning and generating complex drum rhythms in the cerebellum and basal ganglia. We trained the model using the renowned drummer Jeff Porcaro’s hi-hat patterns and expanded its capabilities to produce multi-instrument drum performances. By introducing oscillations of different frequencies to reservoir computing, we enhanced the reservoir’s complex dynamics to create dynamic, non-repetitive, fluctuating time intervals and amplitudes of skilled drumming perfromances. Our findings demonstrate that this computational approach can emulate the nuanced microtiming and audio features essential for skilled drumming, shedding light on the potential neural mechanisms underlying skilled musical performances.