Fig. S4.

Signal distortion of synthetic IRs. (A) Box plots of the distribution of MSE distortion introduced by synthetic IRs to the stimuli used in the source discrimination (Exp 2, Left) and IR discrimination (Exp 3, Right) experiments. The boxes outline the 25th, 50^th, and 75th percentiles. The whiskers delineate the minimum and maximum distortion values. (B) MSE distortion is robust to computational details of the cochleagram. Box plots show the distribution of MSE distortion in the cochleagram across the different IR types (as in A) for a range of changes to the cochleagram, from Left to Right: fewer subbands, more subbands, lower threshold, higher threshold, L1 norm (rather than L2), L-infinity norm, and exponential compression ($x^{0.3}$) rather than logarithmic. (C) Modulation transfer functions for the synthetic IRs used in experiment 2. These were obtained by subtracting the modulation spectrum of the dry source signal from that of the corresponding reverberant stimulus presented on an experimental trial and then averaging this difference over all stimuli. The dry and reverberant signals were first normalized to have the same rms level and hence the transfer function is symmetric around 0 dB.