Datasets:
Bona-fide files in In-the-Wild partition
Hi there, I have a question about the labelling of the In-the-Wild data in this corpus. It seems to me that these are all still generated files, either by TTS or vocoders, but they are organised under "bona-fide" directories rather than dev/test as in the rest of the corpus. Are they called bona-fide because they are derived from the real speech (not fake) portion of the original In-the-Wild corpus?
Hi there, your understanding is correct. We have updated the names of our collection to make it clearer about our generated and real data. The old In-the-Wild collection (now Celebrity subset) is an evaluation only dataset within AUDETER. The audio samples under bona-fide are synthetic audio corresponding to the real speech samples in the original In-the-Wild dataset. This is consistent with other collections in AUDETER to enable matching between real and fake samples.
Cool, thanks for confirming!