Time-Domain Processing in the Auditory System

delay lines gating neurons Jeffress (1948) model of auditory localization. (a) Elementary delay line and coincidence counting neuron. (b) Array of coincidence counters organized tonotopically and by interaural delay (correlation lag). (c) Pattern evoked by a wideband sound source displaced from the axis. A ridge occurs at the internal delay that compensates for the external interaural delay of 0.2 ms. The topmost curve represents the sum of cross-correlation functions over frequency ("summary crosscorrelation function"). This paper reviews models of auditory signal processing that operate in the . This is in contrast with the traditional view that the cochlea performs a Fourier transformation, and that subsequent processing is carried out in the frequency domain. We classify models into parameter , channel , and within-channel models. Phenomena accounted for are sound localization, pitch perception, vowel timbre identification, and various aspects of sound segregation ("cocktail-party") effects. Most models may be implemented based on a time-domain stochastic neural representation (spike trains) such as found in the auditory nerve. Processing involves "neural filters" based on delay lines and gating neurons (coincidence counters), that select or delete certain spikes, thereby modifying the statistics of the spike train in a way that strengthens or weakens the representation of parts of the sound environment. In particular "cancellation filters", in which the interaction is mainly inhibitory, are suggested as a likely ingredient for models of auditory processing.