Decomposition of respiratory motor patterns during birdsong production in terms of excitable transients

The production of birdsong involves complex respiratory motor gestures shaped by intricate neural and muscular coordination. In this work, we propose that these gestures can be decomposed into simpler dynamical elements generated by a minimal excitable system. Using air sac pressure recordings from singing canaries (Serinus canaria), we model each syllable as a linear superposition of transients arising from perturbations of a two-dimensional Wilson–Cowan-type excitable system. By fitting these transients to observed pressure patterns using a differential evolution algorithm, we obtain accurate reconstructions of the motor gestures underlying song. We then apply unsupervised dimensionality reduction and clustering to the extracted transients, identifying a compact set of gesture types—motor primitives—shared across individuals with distinct vocal learning histories. Our results suggest that complex motor sequences in birdsong may be constructed from a limited, reusable repertoire of dynamical modules, offering insights into the structure of learned behavior and the neural encoding of motor control. This framework introduces a principled, dynamical decomposition of vocal behavior with potential applications in the study of motor learning and neuroethology.