Missing and invalid files
Some of the files in this dataset are either missing (the mp3 column is empty) or invalid (the mp3 column contains HTML or other data, not MP3 compressed audio).
I've filtered through everything and made lists of all problematic files.
Quite a few of the missing files seem to be related to misdecoding of the URLs (latin1 instead of utf8) which can be fixed by: url.encode('latin1').decode('utf8') (see also https://stackoverflow.com/a/26492671). The invalid files seem to all be directory listings (the URLs end with /) for valid audiobooks.
https://gist.github.com/jpc/24726773ac0d73c096f0905f334e2ad0 – here are the URL lists with parquet file names where they occur
The character issue is now fixed, I'll start the re-downloading of missed recordings