A simple latency-dependent spiking-neuron model of cricket phonotaxis.

A simple hypothesis regarding the recognition behaviour of crickets for conspecific songs is implemented in a dynamic simulation of spiking neurons and tested on a robot base. The model draws on data from cricket neurophysiology but requires only four neurons to reproduce a wide range of the observed behaviour. The directional response depends on relative latencies in firing onset, and the 'recognition' emerges from the implicit filtering properties of leaky-integrate-and-fire neurons. Experimental conditions reproduced include tests of syllable rate preference, song from above with sound from one side, and choice between songs. The robot produces behaviour closely comparable to the cricket in all but a 'split-song' condition. A number of properties can be observed in the neural circuit that correspond to cricket neurophysiology including apparent 'recognition neurons'. Limitations of the model, extensions and alternative models are discussed.