Skip to main content
. 2023 Mar 13;153(3):1662–1673. doi: 10.1121/10.0017546

TABLE III.

Effect sizes (Cohen's d) for the statistically significant differences between the HASPI w2 and HASPI v2 accuracy for scores averaged over listeners and repetitions. The RMS error and correlation differences are the average of 10 000 bootstrap replications of the differences in the model outputs, and significance was computed from the bootstrapped confidence intervals. Differences at the 5% level are indicated using one asterisk while differences at the 1% level are indicated by two asterisks. The effect sizes were computed from the bootstrapped means and standard deviations.

Experiment Subject group RMS error Pearson Spearman Kendall
Noise and distortion NH 0.4283 0.4184 0.9605 1.1230
HI 0.7144 0.1844 0.4384 0.5077
NH + HI 0.6104 0.2361 1.4115** 1.3789*
Frequency compression NH 0.5410 0.2934 0.2551 0.2787
HI 0.8749 0.8440 0.7399 0.6399
NH + HI 0.1337 1.0437 0.1356 0.1658
IBM noise suppress NH 0.0172 0.2587 0.2035 0.4527
HI 0.3307 0.0434 1.0789 1.1466
NH+HI 0.1390 0.4780 0.7839 0.6909
Noise vocoder NH 10.041** 2.0603* 1.4733* 1.6740*
HI 8.706** 0.6705 0.4081 0.5040
NH + HI 12.091** 1.5311* 1.4922 1.5251
Reverb envelope modify NH 2.7374** 0.9848 0.3729 0.3893
HI 0.4432 0.6316 0.0987 0.0376
NH + HI 1.5569 1.3915 0.1140 0.2410