Below is a comparison table with audio samples for different stimuli using four audio processing stages: Neural Network Model (NN), Dynamic Range Compression (DRC), Ground Truth (GT), and Raw Mixture (Mix).
Click the play button to listen to the corresponding audio files.
| Trigger | Non-Trigger | NN | DRC | GT | Mix |
|---|---|---|---|---|---|
| tap | cricket | ||||
| squeak | waves | ||||
| sniff | camera noise | ||||
| finger-snapping | raining | ||||
| slam | river flowing | ||||
| cutlery-silverware | crows | ||||
| chewing-mastication | speech | ||||
| breathing | footsteps | ||||
| alarm | humming/singing | ||||
| bark | birds chirping |
T-bars represent bootstrapped 95% confidence intervals. Stars indicate the significance level of an algorithm having a higher perceived value than the algorithm of comparison (*p < 0.05, **p < 0.01, ***p < 0.001). The color of the star indicates the algorithm of comparison. For the comparison, a Friedman test was conducted, followed by a paired Wilcoxon post hoc test. P-values were adjusted with Bonferroni. The different stimuli versions are abbreviated the following way: Ground Truth (GT), Noise Cancelling (ANC), Dynamic Range Compression (DRC), Auto-Encoder Neural Network (NN), Raw Mixture (Mix).
