Figure 6.

Model simulated digit-in-noise performance scaled by mean behavioral response. DNN model under the normal-hearing (NH) model produced the expected sigmoidal shape of measured human behavior (black curve), leaving a small man-machine gap of 1.3 dB SNR at 50% digit recognition accuracy. To facilitate visual, qualitative comparison between between the various models of cochlear degradation studied, we scaled the DNN output accuracies on all DNN outputs using an SNR dependent scaling factor. We computed the scaling factor used to remap all subsequent model accuracies once using the sigmoid fits of the DNN NH-model and NH human mean performance.