Abstract:
A computational model of the early stages of dolphin hearing was developed to model dolphin echolocation performance. The first stage consists of a bank of 49 gammatone filters with center frequencies from about 2.5 to 147 kHz. This stage computes the basilar membrane motion that results from the echo time series used as input. The second stage of the model computes a neural activation pattern that rectifies and log transforms the basilar membrane motion and computes lateral inhibition in both frequency and time. The third stage integrates and aligns the neural activation pattern from multiple echoes, a process that has been demonstrated empirically in dolphin echolocation. The result of this processing is a stable auditory ``image'' or time-frequency map of the signal. The parameters of this model are tuned for the anatomy of the dolphin cochlea. The model has been applied to echo returns from a synthesized dolphin signal reflecting from cylinders in clutter. The cylinders were made of different materials and had an outer diameter of 3.8 cm and a length of 10.2 cm.