A client asks for “subtitles” and could mean any of three things.
- The social team wants text in the picture so the cut works on mute.
- The web team needs a caption file because the site goes through an accessibility review.
- The broadcaster’s delivery spec lists closed captions as a separate file with its own rules.
All three put text at the bottom of the frame, but the content and the file you hand over change every time. Working out which one the client means at the briefing saves a re-export at the end, and it decides who has to write what.
In short
- Subtitles carry the dialogue, often translated, for viewers who can hear.
- Closed captions (SDH) add who is speaking and the sounds that matter, and the viewer can turn them off.
- Open captions are burned into the picture, which social feeds and screens without player controls need.
- If the client’s site has to meet WCAG, a dialogue-only SRT doesn’t count as captions.
- Sound descriptions need a human pass after picture lock, quoted as their own line.
Three things people call subtitles
| Type | What’s in it | Can the viewer turn it off? |
|---|---|---|
| Subtitles | The dialogue, often translated | Usually, when delivered as a file |
| Closed captions (SDH) | The dialogue, plus who is speaking and the sounds that matter | Yes |
| Open captions | Subtitles or captions burned into the picture | No |
Subtitles assume the viewer can hear. They carry what’s said, usually translated, and leave the rest to the soundtrack: the music, the laugh, the door slamming off screen.
Closed captions are written for viewers who are deaf or hard of hearing. WCAG, the W3C’s web accessibility guidelines, defines captions as covering speech and also the non-speech audio a viewer needs to follow the content, “including sound effects, music, laughter, speaker identification and location.”
Closed only means the viewer turns them on and off. SDH, subtitles for the deaf and hard of hearing, is the name streaming services use for the same information delivered as a subtitle file.
Open captions are burned into the picture, so everyone sees them and nobody can switch them off. They can be plain dialogue or full captions with sounds. WCAG counts them as captions in the second case.
WCAG also notes that in some countries captions are simply called subtitles, so when a client says subtitles, ask who’s going to read them and where.
What closed captions add to the dialogue
Captions add two things to the dialogue. One is who is speaking, when the picture doesn’t make it clear. The other is the sounds a viewer without audio would miss.
Netflix’s English (USA) style guide puts both in square brackets, in lowercase except for names, and only when the information can’t be seen on screen. It also asks you to describe the sound itself rather than the action that made it. On an event video, part of a caption file could look like this:
12
00:00:41,200 --> 00:00:43,000
[applause]
13
00:00:43,400 --> 00:00:46,100
[announcer] Please welcome
our CEO, Marta Lima.
The announcer gets a label because the voice comes from offstage while the camera holds on an empty stage. When Marta walks on and starts talking in frame, her lines don’t need one. YouTube’s help on caption files suggests the same, with text like [applause] or [thunder].
Music follows the same logic. A cue that sets the mood gets a short description, such as [upbeat music], and Netflix’s guide asks for lyrics to be captioned when they’re audible and don’t clash with the dialogue.
Leaving things out takes as much judgment. The air conditioning hum and a laugh the viewer can already see on a face add reading without adding meaning. Ask whether someone watching without sound would miss something a hearing viewer gets.
Open captions: when the text has to be in the picture
Social feeds are the main case. LinkedIn’s help says some videos play automatically without sound as people scroll, and plenty of people watch on a phone set to silent, in a waiting room or on the train.
A cut that depends on someone talking has to make sense on mute, and the only text you can be sure the viewer sees is text in the picture. Some apps don’t even have a place to upload a subtitle file.
The same goes for screens with no player controls, like the loop on a trade show stand or the video playing in a lobby.
Burning text in has costs. It covers part of the frame for everyone, including people watching with sound, every language becomes a separate export, and a typo means a new render and a new upload.
The captions also compete with lower thirds and on-screen graphics, so their position has to be planned shot by shot, and on vertical video they have to sit above the platform’s buttons. Line length, reading speed and vertical framing are covered in our subtitle guidelines.
What to deliver, destination by destination
| Destination | What to deliver |
|---|---|
| Social feeds | An export with open captions |
| The client’s website | A clean master plus a VTT with captions |
| YouTube | A clean master plus a caption file uploaded to the video |
| Broadcast or streaming | What the delivery spec names, often captions or SDH as a separate file |
| Event screens and loops | An export with open captions |
| Other languages | A subtitle file per language, or an export per language for social |
Which file format goes where, SRT or VTT, is in our post on SRT vs VTT.
Accessibility usually comes in on the website row. If the client’s site has to meet WCAG, captions for prerecorded video are success criterion 1.2.2, at Level A, the most basic level.
By WCAG’s definition, an SRT with dialogue only doesn’t meet it. Ask at the briefing whether the video will go through an accessibility review, because the answer changes the budget.
Who writes the sound descriptions
Automatic transcription has made dialogue cheap. A transcript in minutes, checked once against the audio, turns into a subtitle file the same day. Captions need another pass, because a transcription engine writes down speech and has no way to know which sounds matter to the story.
That pass belongs to someone who knows the cut, usually the editor or an assistant, watching with sound on and the subtitle file open.
They look for voices whose speaker isn’t in frame, sounds that carry meaning (the phone that interrupts the interview, the applause that closes a talk) and music that sets the tone or has lyrics worth reading.
It’s a quick job on a short corporate video and a long one on a documentary, and either way it should be its own line in the quote.
Do it after picture lock, since a trim moves every caption that comes after it. Burned-in text comes last, from the final file, once the client has approved the words.
And put the decision in the delivery list at kickoff: which versions get open captions, which get a file, in which languages, and whether the file needs speaker IDs and sounds. Nobody wants to find out two days before launch that the LinkedIn cut has no text.
How this works in RecReview
In RecReview you transcribe a version in Portuguese, English or Spanish, or let it detect the language from the speech, and subtitles are built from that transcript with up to 42 characters per line and two lines at most.
In the review player they work like closed captions. Once the version has a transcript, the CC button, or Shift+C, turns them on and off, and the choice is remembered on that device. Clients watching on the review link get the same button.
They carry the dialogue only. The transcript tells voices apart as Speaker A and Speaker B, but those labels don’t go into the subtitles, and RecReview doesn’t add sound descriptions.
Workspace owners, admins and editors, and the person who uploaded the version, can click a subtitle in the player and fix its words. The timing stays as it was built, and you can’t add a subtitle where nobody is speaking.
For a caption file, export the SRT or VTT from the transcript tab and add the speaker IDs and sounds in a subtitle editor or your editing software.
RecReview doesn’t burn subtitles into the video either. For open captions, bring the exported SRT into the edit, style it there and upload the captioned export as a new version for the client to review.
When that version goes through AI Review, it reads the text in the picture, burned-in captions included, and points out clear spelling mistakes at their timecode. The transcript also exports as TXT or JSON.
