Data Quality Issues
#2
by
harryjulian
- opened
Whilst assessing the data quality I noticed there seems to be quite a lot of misalignment between audio + original transcripts (between 2-5% for some languages). This seems to be a problem with MLS too, and I understand that the transcription + Levenshtein calculation have been included to allow for more straightforward filtering, but I think it could be useful to make this a bit more obvious on the data card.