Closed captions are a time-synchronized text version of the spoken dialogue and meaningful sounds in a video, delivered as a separate track that viewers can turn on or off. In video accessibility, they serve primarily people who are deaf or hard of hearing, so they include not only speech but also speaker identification, sound effects and music cues such as a door slamming or tense music.
The word closed distinguishes them from open captions, which are burned permanently into the picture. Closed captions are required by broadcast and accessibility regulations in many countries and are expected on most online video platforms, where they also help viewers in noisy or quiet environments and people watching in a second language.
How closed captions work
A closed caption track is a sequence of caption cues, each holding a piece of text, a start time and an end time, and often position and styling information. The player or television decoder renders the text over the picture only when the viewer enables captions.
In broadcast television, captions travel inside the video signal or stream: the North American CEA-608 standard originally used line 21 of the analog picture, and its digital successor CEA-708 carries captions in the transport stream. Online video usually delivers captions as separate sidecar files or as text tracks referenced by the streaming manifest.
The following table lists the caption formats that practitioners meet most often.
| Format | Typical context | Notes |
|---|---|---|
| CEA-608 | Legacy North American broadcast | Limited characters and styling |
| CEA-708 | Digital television in North America | More languages, fonts and positioning |
| WebVTT | Web players and HLS | Plain text with cue settings |
| TTML and IMSC | Streaming and broadcast delivery | XML-based, rich styling |
| SRT (SubRip) | General exchange and social platforms | Simple, minimal styling |
| SCC | Exchange of 608 data | Common in professional delivery |
Closed captions compared with subtitles and related features
Closed captions differ from subtitles mainly in audience and content: captions assume the viewer cannot hear the audio, while standard subtitles assume the viewer can hear but does not understand the language. In everyday usage the terms are often mixed, and some regions use subtitles for both.
| Feature | Primary audience | Includes sound cues | Can be turned off |
|---|---|---|---|
| Closed captions | Deaf and hard-of-hearing viewers | Yes | Yes |
| Open captions | All viewers | Yes | No, burned in |
| Subtitles | Viewers who do not know the language | Usually no | Usually yes |
| SDH subtitles | Deaf viewers, in subtitle format | Yes | Yes |
| Audio description | Blind and low-vision viewers | Narrates visuals instead | Yes |
Subtitles for the deaf and hard of hearing (SDH) combine the subtitle file format with caption content, which is common on disc and streaming releases. For full accessible video, captions are usually paired with audio description, which serves a different audience.
Quality standards for closed captions
Good captions are judged on accuracy, synchronization, completeness and placement, the four quality criteria adopted by broadcast regulators in the United States. Captions that are technically present but inaccurate or badly timed can fail viewers as badly as missing captions.
| Parameter | Common guideline | Why it matters |
|---|---|---|
| Accuracy | Close to 99 percent of words correct | Errors change meaning |
| Line length | About 32 to 42 characters | Readable at a glance |
| Lines per caption | Usually one or two | Limits obscured picture area |
| Minimum duration | About 1 to 1.5 seconds | Short cues are unreadable |
| Reading speed | About 15 to 20 characters per second | Viewers must also watch the picture |
| Placement | Bottom center, moved when needed | Avoids covering faces and on-screen text |
Sound cues follow conventions such as square brackets for non-speech sounds and a music note symbol for songs, and speaker labels are added when the speaker is off screen. Exact style rules differ between broadcasters and platforms.
Producing captions: offline and live workflows
Captions for recorded content are created offline, with time to transcribe, edit and time every cue precisely. Live programs require real-time captioning, which accepts some delay and errors in exchange for immediacy.
| Method | Typical use | Delay | Accuracy |
|---|---|---|---|
| Offline transcription and timing | Films, series, recorded courses | None at playback | Highest |
| Stenography | Live news and sports | A few seconds | High with skilled writers |
| Respeaking | Live broadcasts, events | A few seconds | High after training |
| Automatic speech recognition | Webinars, social video, drafts | Low | Variable, needs review |
Automatic speech recognition is a useful first draft, but it struggles with names, technical terms, accents and overlapping speech, and it rarely adds sound cues on its own. Human review remains the norm for published content.
Common mistakes with closed captions
- Publishing raw automatic captions without correcting names, numbers and terminology.
- Omitting sound cues, which turns captions into plain subtitles and removes context for deaf viewers.
- Covering on-screen text such as lower thirds or scoreboards instead of moving the caption.
- Losing captions in the delivery chain, for example when a transcode or streaming packager drops the caption track.
- Captions drifting out of sync after an edit changes the program timing.
The idea in brief
Closed captions are the words that appear at the bottom of the screen when you switch them on. They show what people say and what can be heard, such as laughter, a phone ringing or music playing.
They exist mainly so that people who cannot hear can follow the whole story, not just the conversation. Many other viewers use them too, for example on a train without headphones or late at night when the sound must stay low.
A practical scenario
A university records a 50-minute engineering lecture and publishes it on its video platform. An automatic speech recognition draft is produced within minutes, but it misspells formulas, technical terms and the lecturer’s name dozens of times and contains no sound cues.
An accessibility coordinator edits the transcript, splits it into two-line captions of no more than 42 characters per line, adds labels for student questions from off camera and exports a WebVTT file. A deaf student in the course can now follow the lecture fully, and search across the transcript helps every student find the exact minute where a topic is explained.