Remove Subtitle Formatting

There are four separate things sitting on top of the words in a subtitle file, and they are four switches rather than one clean it up button because removing them is four different decisions with four different consequences. Strip formatting from a track and you have a plainer file; strip the accessibility layer with it and you have a different file altogether.

Markup. <i>, <b>, <u>, <font color=…> and WebVTT’s <c.classname>. Removing italics is the safest thing on this page, and the usual reason to do it is a player that renders the tags as text rather than honouring them — a subtitle reading <i>whispering</i> on screen is worse than one with no italics at all.

Positioning tags. {\an8} and the rest of the ASS override syntax, which turns up inside SubRip files constantly because that is what falls out of a careless conversion. They mean nothing to a SubRip player and are printed literally by most of them.

Sound descriptions. [DOOR SLAMS], (sighs), and the music note that marks a lyric. Speaker labels. JOHN: at the start of a line.

Those last two are not formatting. They are the content of a subtitle track for the deaf and hard of hearing, and removing them is what turns an SDH file into an ordinary subtitle file. That is exactly right when you need a clean script to translate from or to read as prose, and exactly wrong when the file being cleaned is the one that ships — so they are off by default and warn when switched on.

Why the speaker rule looks over-careful

A label is only removed when it is set in capitals. JOHN: goes; Note: stays. There is no way to tell those apart by shape — both are a word followed by a colon at the start of a line — and a rule that removes both eats the first word of every sentence that happens to begin with one. Capitals are the convention every subtitle house uses precisely so a reader can tell a label from a word, and it is the only signal in the file.

The dash in front of a label survives it. In a two-hander, - JOHN: Get down. becomes - Get down., because once the name has gone the dash is the only thing left saying the speaker changed.

Press Load a sample. Five cues, and with the default switches — markup and positioning on, the accessibility layer off — the status line reads cues 5 / 5 and changed 1. Only the first cue had anything to remove.

Now switch on sound descriptions. The count changes to cues 4 / 5 and a new figure appears: emptied 1. Cue 2 was [DOOR SLAMS] and nothing else, so it has no text left. An empty cue is a gap on screen rather than nothing, so it is dropped — and it is counted, because a tool that quietly loses a cue is a tool you find out about while watching.

Cue 5 keeps its line. ♪ and the music plays on ♪ loses the notes and stays as words, because the lyric is dialogue and the notes are the decoration around it.

Switch on speaker labels as well. Cue 3 becomes - Get down. / - Where? — the two names gone, the two dashes kept. And cue 4 still reads "Note: this line keeps its first word.", which is the whole reason the capitals rule exists.

Once the track is clean it is usually going one of two places: into a transcript, or into a translation pass — in which case the reading speed is worth measuring on the stripped version, since the tags were never on screen and should never have been counted.