Closed Captions & Subtitles
A team files captions under accessibility, gives them to whoever has capacity, and ships whatever the automatic tool produced.
Captions open your product to millions. Learn the rules hands-free, on the go.
Listen to the full Closed Captions & Subtitles course on DesignPath
Ofcom counted 7.5 million people in the UK using subtitles, about 18% of the population. Only 1.5 million of them were deaf or hard of hearing, so four out of five could hear the audio and had the captions on anyway.
They are on in noisy rooms, on muted phones in public, in a second language, and by people who retain more when they read along. So caption quality is not a compliance item affecting a minority. It is the reading experience of most of the audience for the video.
Two things follow from taking it seriously. The words captions and subtitles describe different tracks for different people, and one file cannot do both jobs. And a caption is a designed object with a position, a duration and a line length, none of which an automatic tool decides well.
Anatomy of a Caption Track
Every decision in captioning lands on one of two objects: a single cue, or the track it belongs to.
The Cue
The unit on screen at any moment, carrying six properties.
The text, on one or two lines. The in and out times, which set how long it is readable. The speaker label, where it is not obvious who is talking. The sound cue, describing audio that carries meaning. The position on the frame. And the styling that marks emphasis, whispering or a change of voice.
The Track
A set of cues in one language, produced for one purpose.
A track carries a language code, a kind (captions, subtitles or SDH) and a label somebody picks from a menu. A video with an international audience and a deaf audience needs at least two, and they are chosen separately rather than derived from each other. Article 3 covers why that separation is not a technicality.
Two Lines, and the Ceiling Behind Them
A cue holds two lines because a third one starts covering the picture.
That single constraint produces most of the formatting rules in article 4. Around 40 characters per line, split into two balanced parts, is what fits without shrinking the type or eating the frame. Everything about breaking, timing and reading speed follows from having that much room and no more.
Where a Track Comes From
Three routes, with different costs and different quality.
A script, where one exists, which is the cheapest and most accurate. A human transcription, which is the most accurate and the slowest. An automatic track, which is instant and always needs a pass over it, for the reasons in article 7. Whichever route, somebody watches the video with the track on before it ships.
Pro tip: Open the caption settings on any streaming service and count the properties you can change. That list is the anatomy, written by a team that has had this argument already.