The effect of different spectro-temporal representations of an input auditory stimulus on the fitting of a point process model of auditory neurons

Abstract

We compare the effect of the use of three different spectro-temporal representations of an input auditory stimulus on the fitting of a point process model of auditory neuron firing. The three spectro-temporal representations considered are the spectrogram, a gammatone filterbank and the Hilbert spectrum. We firstly investigate how the model fits the recorded neuronal data when using either one of the three representations and secondly how well do the estimated parameters of the model correspond to their experimentally measured counterparts. It is observed that all three representations yield a model that fits well the recorded data. However, the characteristic frequencies obtained with the spectro-temporal parameters of the model using the gammatone filterbank corresponds better to the experimentally measured characteristic frequency than the characteristic frequency obtained with the models using the other two spectro-temporal representations. Therefore, it is concluded that the quality of the fitted parameters can be affected by the choice of the spectro-temporal representation and that, as could have been expected, the gammatone filterbank seems to more accurately extract the relevant spectro-temporal characteristics of the input auditory stimulus.