Missing audio files.
In the website version of data link (and also mentioned in metadata files) there are 53 more files in the German dataset than it is in the HF version. The problem could be with other languages, too; I didn't examine other languages.
Missing file list:
More in tar: 53
More in hf: 0
Differences all: 53
{'10148_11442_002024.flac',
'10148_11442_002054.flac',
'10148_11442_002111.flac',
'10148_11442_002121.flac',
'10148_11442_002128.flac',
'10148_11442_002219.flac',
'10148_11442_002318.flac',
'10148_11442_002322.flac',
'10148_11442_002356.flac',
'10148_11442_002396.flac',
'10148_11442_002405.flac',
'10148_11442_002497.flac',
'10148_11442_002520.flac',
'10148_11442_002543.flac',
'10148_11442_002625.flac',
'10148_11442_002658.flac',
'10148_11442_002688.flac',
'10148_11442_002704.flac',
'10148_11442_002719.flac',
'10148_11442_002744.flac',
'10148_11442_002790.flac',
'10148_11442_002817.flac',
'10148_11442_002835.flac',
'10148_11442_002861.flac',
'10148_11442_002887.flac',
'20_723_000006.flac',
'20_723_000014.flac',
'20_723_000024.flac',
'20_723_000033.flac',
'20_723_000035.flac',
'20_723_000036.flac',
'20_723_000052.flac',
'20_723_000053.flac',
'20_723_000054.flac',
'20_723_000073.flac',
'20_723_000088.flac',
'20_723_000091.flac',
'20_723_000093.flac',
'20_723_000095.flac',
'20_723_000105.flac',
'20_723_000109.flac',
'20_723_000112.flac',
'20_723_000114.flac',
'20_723_000120.flac',
'20_723_000121.flac',
'20_723_000138.flac',
'20_723_000151.flac',
'20_723_000162.flac',
'20_723_000169.flac',
'20_723_000174.flac',
'20_723_000185.flac',
'20_723_000186.flac',
'20_723_000190.flac'}
Hey @grzegorz700 ! Checking the original OpenSLR page, there's a note on additional audio and cleaner transcriptions that were added as a follow-up to the original dataset:
NOTE: We have made few updates to the MLS dataset after our INTERSPEECH paper was submitted to include more #hours and also improve the quality of transcripts. To avoid confusion (by having multiple versions), we are making ONLY one release with all the improvements included.
I'll re-convert the dataset from the latest files, which should add these audio files in.
Thank you for pointing this out.
I've seen that the metafiles didn't change, only the audio files changed. I compared them, and they are the same. So, this hugging-face version has the latest metadata, but users need to download from the source to get all the audio, right?
With @ylacombe the plan is to fix this for the HF Hub version in the coming weeks!