Starting in version 4.0, we attempt to detect the margins, and skip text in them in pdfs.  This is done during import, if a document was imported before 4.0, it’ll have to be re-imported before the margin skipping will work.  The margin detection finds the margins successfully in the vast majority of the time, but there are times where it’ll some clear to human it’s a margin, but the mathematically model we use can not determine it with confidence (especially on short documents).  There’s a switch in Audio settings for toggling the margin skipping off.