How to caption a video for accessibility

Captions are the most commonly skipped accessibility feature and one of the easiest to get right. They are also, unusually, a feature where doing the accessible thing measurably helps everyone: a large share of video on social platforms is watched with the sound off, and captions are what make that possible.

This guide covers what captions are, how they differ from subtitles, how to produce them, and the editing that separates usable captions from technically-present ones.

Captions, subtitles, and transcripts are three things

These get used interchangeably and they are not the same, which matters when someone asks you for one specifically.

• Captions assume the viewer cannot hear the audio. As well as the speech, they convey non-speech information that carries meaning: who is speaking when it is unclear, and relevant sounds like a door slamming or a phone ringing. This is the accessibility artifact.
• Subtitles assume the viewer can hear but does not understand the language. They translate dialogue and generally leave out sound effects, because the viewer can hear those already.
• Transcripts are the full text as a separate document, with no timing. They are useful for search, for reading instead of watching, and for people using assistive technology who would rather read at their own pace.

Automatic tools produce something closer to subtitles: accurate speech, no sound effects, no speaker identification unless you add it. Turning that into captions is a short editing pass, described below.

Why it matters beyond compliance

The accessibility case is straightforward: without captions, deaf and hard of hearing viewers are excluded from your content entirely.

The standard usually referenced is WCAG, the Web Content Accessibility Guidelines, where captions for prerecorded video sit at Level A, the most basic conformance level. Whether that is a legal requirement for you depends on your jurisdiction, your sector, and your organization, and it is worth checking rather than assuming either way.

The practical case is that captions help far more people than the ones they are designed for: viewers in noisy places, viewers in quiet places who cannot use sound, non-native speakers, anyone watching technical content with unfamiliar terminology, and anyone scrolling a feed with audio muted by default.

Producing the caption file

The mechanical part is now easy. You need timed text in SRT or VTT format, which any modern transcription tool can generate.

1. Get the audio. If you have a video file, export or extract its audio track as MP3 or WAV. Make sure it starts at the same moment the video does, or every caption will be offset.
2. Transcribe it with timestamps on. Our audio to SRT and VTT tool does this in your browser, which matters if the video is not published yet, since otherwise the caption file is the first copy of your content to leave your machine.
3. Turn on speaker separation if more than one person speaks, so you have something to build speaker identification from.
4. Edit before exporting. Corrections made in the editor carry into the downloaded file, which is easier than editing an SRT afterward.
5. Download SRT for editors and platforms, VTT for web video.

The editing pass that actually matters

Auto-generated captions are a first draft. Publishing them unedited is where most people go wrong, and a viewer who depends on captions can tell immediately. Priorities in order:

• Fix the words. Names, technical terms, and numbers first. A caption track full of mangled terminology is worse than useless in instructional content, because the viewer cannot tell which parts to trust.
• Keep lines short. Roughly forty characters per line, two lines maximum. Long lines force the eye across the whole frame and get truncated on phones.
• Break at natural phrase boundaries. Splitting between an article and its noun, or before a preposition, is measurably harder to read than breaking at a clause.
• Give each caption enough time. At least a second on screen. Merge very short captions from rapid exchanges rather than letting them flicker.
• Identify speakers when it is not obvious. Convention is a name followed by a colon, or a name in brackets, used when the speaker changes off screen or when several people are talking. Do not label every line in single-speaker narration, where it just eats space.
• Add meaningful non-speech sounds in brackets, for example a phone ringing, laughter, or music starting. Include what carries meaning, not every ambient noise. If a sound would change how a hearing viewer understands the scene, caption it.
• Caption what was said, including false starts where they matter. Do not silently clean up speech into something more polished than the audio, because captions and audio should match.

Where the file goes

• YouTube.Upload the SRT in the video's subtitle settings, choosing the option for a file with timing. This replaces the automatic captions rather than making you correct them line by line.
• Video editors. Import the SRT as captions and drop it onto a caption track. Keep captions as a separate track rather than burning them into the picture where possible, since burned-in captions cannot be turned off or restyled by the viewer.
• Your own web page. Use VTT with a track element on your video, marked as captions with the correct language code. Browsers require VTT and will ignore an SRT.
• Social platforms. Most accept SRT uploads. Where a platform only offers burned-in captions, remember they cannot be disabled, so keep them clear of the areas the interface covers.

Common mistakes

• Publishing auto-captions unedited, then treating the video as captioned.
• Captions offset from the audio because the exported audio did not start at zero.
• Three or four lines on screen, covering the picture.
• No speaker identification in a multi-person interview, leaving the viewer guessing.
• Omitting a sound that the scene depends on.
• Burning captions in when a separate track was possible, removing the viewer's control.

Also publish the transcript

Captions and a transcript serve different people, and you already have both once you have done the work. A plain transcript on the page lets people read rather than watch, makes the content searchable, and is often easier to use with a screen reader than timed captions. It costs you one extra download and a paragraph of page space.

Ready to start? The subtitle generator runs in your browser and produces both SRT and VTT, and your audio never leaves your device. For the step-by-step version of extracting audio from a video first, see our guide to subtitling from audio.