Detection of Auditory Periodicity: Comparing Behavioral Data and a Decision Algorithm Based on a Neural Net

We present psychophysical data on fast auditory periodicity detection in the millisecond range (temporal pitch) that rule out a simple first-order inter-spike interval model. We then present a neural sequence learner that examines fast spike patterns as they are supposed to occur with periodic auditory signals and measures the degree of their regularity. The output of the neural net is evaluated by a decision algorithm that is able to decide between more and less regular sequences. The data of this virtual decider prove to be compatible with the psychophysical data. A periodic sound with the fundamental frequency f0 consists of a series of harmonics f0, 2 f0, 3 f0, etc. or only of some of the harmonics (e.g. in the case of the missing fundamental). There are two complementary mechanisms which accomplish the perception of the periodic sound's pitch: • Spectral pattern recognition acts on the neural excitation pattern produced in the inner ear (cochlea) by the lower, resolvable harmonics (≤15 f0). • Temporal periodicity analysis interprets the temporal structure of the excitation of the cochlea. This is the only operative mechanism for harmonics above 15 f0, which can no longer be resolved by the cochlea and do not produce any interpretable spectral pattern. Our research concentrates on the second mechanism.