Skip to content

Instantly share code, notes, and snippets.

@patientzero
Last active January 5, 2025 23:11
Show Gist options
  • Save patientzero/b23943107d8c6f17cbcb5ffc13fd45b9 to your computer and use it in GitHub Desktop.
Save patientzero/b23943107d8c6f17cbcb5ffc13fd45b9 to your computer and use it in GitHub Desktop.
Speaker exclusive splits used on the FluencyBank part of SEP-28k in the paper Detecting Dysfluencies in Stuttering Therapy Using wav2vec 2.0
# Sebastian P. Bayerl
FB_TEST_SPK = ['68m', '26m', '54m', '24fc', '34m', '24ma']
FB_DEV_SPK = ['46ma', '27f', '24fb', '37m', '29mc', '32m', '57m']
FB_TRAIN_SPK = ['24fa', '62m', '50fa', '29ma', '64m', '35ma', '27ma',
'29mb', '28m', '46mb', '43m', '26f', '60m', '35mb',
'62f', '24mb', '25m', '39f', '54f', '50fb']
@marekjg
Copy link

marekjg commented Jul 2, 2024

FB_TEST is entirely contained in FB_TRAIN_SPK. From what I can see, the string corresponds to file name in the url. Is it for sure the split used in the paper or was there some copying mistake in the way?

@patientzero
Copy link
Author

patientzero commented Jul 4, 2024

Thanks for pointing out the mistake. Indeed, there was a mistake in changing it from the previous revision of the splits that only had two sets (train/dev and test). The current version is the one used in all experiments.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment