Figure 6.

The proportion of variance explained by single-cue and multiple-cue models of the average listener for the (a) 2900-Hz bandwidth, (b) 100-Hz bandwidth, and (c) 50-Hz bandwidth waveforms. The x axis shows the names of different models (CB: energy cue, ES: envelope cue, PO: fine-structure cue, LC: linear combination of three cues, LRT: nonlinear logarithmic likelihood ratio test combination of three cues). The stars indicate that multiple-cue model predictions were significantly improved compared with predictions from any single-cue model (p < 0.05, n = 25 for 2900- and 100-Hz waveforms, n = 50 for 50-Hz equal-energy waveforms). The y axis shows the proportion of variance explained by different models. The length of the error bar shows the standard deviation of the predicted proportion of variance across individual listeners. The dotted lines indicate the predictable variance for hit and FA rates.