Captions and subtitles for video

Captions or subtitles?

The words are used interchangeably in casual speech and mean different things in practice.

WebVTT: the format the web uses

Web Video Text Tracks is the W3C format that HTML's <track> element and HLS both use. A file is plain UTF-8 text that begins with the line WEBVTT, a blank line, and then cues. Each cue is an optional identifier, a timing line, and one or more lines of text, separated from the next cue by a blank line:

WEBVTT

1
00:00:01.000 --> 00:00:03.500
Welcome back to the workshop.

2
00:00:03.500 --> 00:00:06.000
[applause]
Today we start with the encoder.

Timestamps use a dot before the milliseconds, and hours are optional. Cues can carry settings after the timing line. line moves text away from the bottom when a lower-third graphic is in the way. position and align set horizontal placement, and size limits width. Inline tags mark voices (<v Priya>), italics, bold and classes that a STYLE block or the page's CSS can address. NOTE blocks hold comments. The full grammar is in the W3C specification listed at the end.

WebVTT versus SRT

SRT is older, simpler and still what most editing tools export. Converting between them is largely a matter of the header and the timestamp separator.

How a track reaches the player

On a plain web page, captions are attached with a <track> child of the video element: <track kind="captions" src="en.vtt" srclang="en" label="English" default>. The kind attribute distinguishes captions, subtitles, descriptions, chapters and metadata tracks; browsers render the first three and expose all of them through the TextTrack API. The track file may sit on a different origin from the page. If so, it must be served with CORS headers, and the video element needs the crossorigin attribute. Otherwise the track loads silently as empty.

In HLS, the format VideoBB streams, subtitles are a separate media playlist of WebVTT segments, declared in the multivariant playlist with an EXT-X-MEDIA tag of type SUBTITLES. The player fetches the text segments in step with the video segments, which is why a caption track can be switched without reloading the stream. Broadcast-style CEA-608 and CEA-708 captions can also be carried inside the video segments themselves, though web delivery has largely standardised on WebVTT.

Writing captions people can read

These are working conventions from broadcast and accessibility practice, not hard limits. Reading speed is the constraint everything else serves.

1. Keep cues to two lines, and prefer breaks at natural phrase boundaries rather than at a character count.
2. Aim for roughly 32 to 42 characters per line as a rule of thumb; longer lines force the eye to travel and shorter ones flicker.
3. Hold each cue on screen for about one to six seconds. A cue that flashes for half a second cannot be read. One that lingers past the next sentence confuses.
4. Sync to within a fraction of a second of the speech. Late captions feel wrong even when the text is perfect.
5. Describe meaningful sound in square brackets and identify speakers when the picture does not make it obvious.
6. Transcribe what was said, including slang and false starts where they carry meaning. Clean up filler that adds nothing.
7. Move cues with the line setting when they would cover on-screen text.

Automatic captions: a draft, not a deliverable

Speech recognition is now good enough that an automatic transcript is a sensible starting point. It is still weak at the things that matter, so it should never be the finished product. Names, product terms, homophones, numbers and anything said over music are where it fails, and it punctuates badly, which makes even a correct transcript hard to read. The efficient workflow is to generate, then edit against the audio in a caption editor, then export WebVTT. Budget roughly the running time of the video for the edit on clear speech, and more for interviews and accents the recogniser has not heard.

Captions, search and reuse

A caption track is not crawled as page content, but the transcript it contains is the most complete description of a video you will ever write. Publish it on the page, in full or as a summary with timestamps. That gives search engines text to index and viewers something to skim. VideoBB does this for you: published captions appear as a transcript under the video, and the player features guide covers the captions menu. The same file translates into subtitle tracks for other languages, feeds chapter markers, and becomes the script for a blog post. Of all the assets around a video, the caption file is the one that keeps paying back. Once your source file is ready, the preparation guide covers the rest of the export, and upload accepts files up to 4 GB with resumable transfers.