Raw detection scores for every model and every evaluation corpus in
Spoof-SUPERB, a benchmark of self-supervised speech representations for audio
deepfake detection.
Every number in the paper can be recomputed from these files alone, with no
GPU, no model checkpoints and no access to the underlying audio corpora.
