Figure 1.

A, A bilinear STRF. The STRF (right) is equal to the outer product of two vectors, one describing the temporal response (left) and the other describing the spectral response (middle). B, Schematic of the fully separated input nonlinearity model w^t ⊗ w^f ⊗ w^l. The response to latency, frequency and sound level is assumed to be fully separable. C, Schematic of the input nonlinearity model w^tf ⊗ w^l. The full spectrotemporal component (left) and the sound-level component (middle) produce a response function (right) that is separable in sound-level and time, sound-level and frequency, but not in time and frequency. Note that there are weights (not shown) in the interior of the cube.