Resonances and Spectral Characteristics of a Neural Network for the Song Motor Pathway in Birds

Abstract

Birdsong is a complex learned behavior regulated in an intricate way. Neuromuscular coordination of different muscular sets is necessary for producing high quality and consistent songs. We developed a realistic neural network that emulates neurons from the High Vocal centre (HVc) and the robust nucleus of the archistriatum (RA) neurons that drive the muscles to generate birdsong sounds. We used modern computational tools and neural architecture to simulate the entire motor pathway up to the physical oscillator muscle system and its spectral characteristics. Several network parameter dependences were analysed and elucidated. An optimal network size within 10 to 25 neurons within which minimal and smooth frequency variations occur, was found. Beyond that range we observe, instead, strong frequency dependence. Moreover, response frequency is influenced by the pathway input current; also in this case frequency response keeps smooth within a certain range of current, but shows interesting resonant values where negative peaks are observed. Resonant values are found in respect to the non-linear dissipation constant of the equation of motion for the bird's labial oscillation and also in respect of the network background noise level (stochastic resonance). This work demonstrate that is possible to achieve a realistic computational model for the motor pathway leading to generation of sounds in birds, which contributes to the understanding of its spectral and dynamical fundamental properties and characteristics.