FIG. 1.
Accuracy as a function of TMR for the human-trained ESR model variant and for human listeners. Thin solid lines show performance of each fold-specific model (five-fold cross-validation), with all five models evaluated on the same external test set of 10 000 mixtures whose TMR values were not used in the human-subjects experiment or for model training. The thick solid line shows the mean accuracy across folds, and the shaded region indicates ±1 SD across folds at each TMR. The dashed line with circular markers shows mean human accuracy at each tested TMR.