Figure 2. Similar reverberation effects were observed in the dereverberation model kernels and neuronal STRFs.

(A) Example model kernels resulting from the dereverberation model. Three example model kernels are shown, after training on the large (top row) or small (bottom row) room reverberation. The frequency channel which the model kernel is trained to estimate is indicated above each kernel. The color scale represents the weights for each frequency (y-axis) and time (x-axis). Red indicates positive weights (i.e. excitation), and blue indicates negative weights (i.e. inhibition; color bar right). (B) Each plot in the top row shows the temporal profile of the excitatory kernel weights for the corresponding example model kernels shown in A. Excitatory temporal profiles were calculated by positively rectifying the kernel and averaging over frequency (the y-axis), and were calculated separately for the small (pink) and large (red) rooms. The center of mass of the excitation, $CO{M}^{+}$, is indicated by the vertical arrows, which follow the same color scheme. The bottom row plots the inhibitory temporal profiles for the small (cyan) and large (blue) rooms. Inhibitory temporal profiles were calculated by negatively rectifying the kernel and averaging over frequency. The $CO{M}^{-}$ is indicated by the colored arrows. (C) Spectrotemporal receptive fields (STRFs) of three example units recorded in ferret auditory cortex, measured for responses to natural sounds in the large room (top row) or small room (bottom row), plotted as for model kernels in A. (D) Temporal profiles of the STRFs for the three example units shown in C, plotted as for the model kernels in B.

Figure 2—figure supplement 1. Model kernels and neuronal STRFs across frequency channels.

(A) Model kernels arranged by the anechoic frequency that they were trained to estimate. For each anechoic frequency, the top row shows the kernel for the large room condition, and the bottom row shows the kernel for the small room condition. In each plot, frequency is on the vertical axis and history on the horizontal. (B) Neuronal STRFs arranged by best frequency, the frequency in the STRF with the largest weight. The STRFs of all cortical units with the same best frequency were averaged to produce these plots. Plots are arranged as in A.

Figure 2—figure supplement 2. Model and neuronal temporal profiles across frequency channels.

(A) Temporal profiles of the excitatory (top rows) and inhibitory (bottom rows) weights of the model kernels shown in Figure 2—figure supplement 1A, plotted as in Figure 2B. The estimated anechoic frequency channel is indicated above each pair of plots. The color code is as in Figure 2: pink = small room excitation; red = large room excitation; cyan = small room inhibition; blue = large room inhibition. The center of mass ($COM$) values for the excitation and the inhibition in each room are indicated by the colored arrows. For each anechoic frequency, each temporal profile was normalized by dividing by the maximum value for the excitatory temporal profile of the same room. (B) Temporal profiles of the excitatory and inhibitory components of the averaged neuronal STRFs shown in Figure 2—figure supplement 1B, plotted and normalized as for the model kernels in A.