SPADE: A Multilingual Dataset for Speech Partial Deepfake Detection and Localization

The repo contains the SPADE dataset, containing partially edited speech deepfakes in 12 languages, generated by up to five systems per language.

Subsets are either named {language}-{model}-edited or {language}-original. Edited and original audio are stored in separate subsets, so they can be downloaded independently. For each edited audio file, the original file corresponding to it can be found via the id column. Following PartialSpoof, we include 0.16-second binary labels under the labels column. Label 0 means original audio and label 1 means generated/edited audio. We also include more finer-grained labels at each model's native resolution under the original_labels column.

The English, Japanese, and Korean subsets are pending review before public release.