Soundscape-Weighted Selective Accuracy
An evaluation metric designed for multi-soundscape, multi-trial settings that measures a model's accuracy on the subset of eligible trials, where both the original answer and a same-class replacement answer are correct. Each soundscape is weighted equally to prevent soundscapes with many trials from dominating the overall metric.
Timeline (last 90 days)
For the inserted categories, the soundscape-weighted selective accuracy ranges from 73.7% to 81.9%.
The soundscape-weighted selective accuracy is computed as follows: first select the trials in which both the original answer and the answer after same-category substitution are correct, average the accuracy of these trials within each soundscape, and then average equally across all qualifying soundscapes.
All Facts (2)
For the inserted categories, the soundscape-weighted selective accuracy ranges from 73.7% to 81.9%.
50%UnverifiedThe soundscape-weighted selective accuracy is computed as follows: first select the trials in which both the original answer and the answer after same-category substitution are correct, average the accuracy of these trials within each soundscape, and then average equally across all qualifying soundscapes.
50%