Caption standards: WCAG, Netflix and BBC in context

Start with the purpose of the captions

Captions make the information in an audio track available through text. That includes dialogue, but it can also include speaker identity and meaningful sounds. A speech recognition model mainly produces words. The missing information has to be added during a caption review rather than assumed to exist because a subtitle file was generated.

W3C’s explanation of prerecorded captions describes synchronized text for audio content in video and identifies the role of non-speech information. Its guidance should not be reduced to a character count. If a sound explains what happened in a scene, a viewer who cannot hear it needs a useful textual equivalent.

Do not combine unrelated delivery rules

Netflix’s English USA guide provides language-specific instructions for its delivery context. BBC guidance discusses subtitle editing and presentation for its own output. These are useful references, but they are not identical universal laws for every web video. Choose the requirements that apply to the project and keep their scope visible.

Even a single publisher may have separate documents for finished subtitles, templates or another language. The Netflix English USA adult SDH speed shown here is not interchangeable with a number from a separate template guide. A search snippet can lose that distinction. Read the section heading and intended deliverable before adopting a limit.

Use counts as review prompts

The checker defaults to the Netflix English USA guidance represented in the table. It flags long lines, fast cues and excess lines. It counts characters in the cue text and divides by the cue’s visible duration. That makes it useful for locating likely trouble spots, not for declaring a file finished.

A short line can still be difficult because it contains unfamiliar names or dense terminology. A line that fits a character limit can still be too wide in a particular font. Conversely, a delivery system may define width by a screen region rather than a simple text count. Preview the actual rendering and follow the destination’s rules.

Timing must follow the media

A caption should appear with the relevant speech or sound and remain readable for its intended duration. Check the beginning and end of a cue against the actual video. Do not extend text into a later speaker’s turn merely to satisfy a speed figure if doing so creates confusion about who is speaking.

A fixed timing shift is appropriate when the entire file is early or late by the same amount. It cannot fix a video that has been shortened in the middle or plays at a different speed. In those cases, the relationship changes across the timeline and the cues need more specific editing.

Always review against the final cut. An added introduction or changed advertisement can make otherwise correct subtitle text appear at the wrong moment. Keep the subtitle export paired with its source version.

Read the broader quality criteria

The FCC’s published caption-quality order describes accuracy, synchronicity, completeness and placement. Those dimensions are useful for understanding why a technically valid file can still be poor captioning. This guide cites that document as background; it does not determine the legal duties of an individual publisher.

Accuracy concerns the audio information being represented. Synchronicity concerns its relationship to time. Completeness concerns the material that is covered. Placement concerns the viewer’s ability to see both captions and relevant picture content. A basic SRT file does not settle every placement issue because the player and delivery format also matter.

If the project has contractual, regulatory or accessibility requirements, use the applicable current rules and qualified review. MurmurScribe does not certify captions or establish conformance by running an automated checker.

Build a human review pass into delivery

First correct the words against the recording. Then add meaningful sounds and speaker cues. Next review timing and line breaks in the destination player. Treat these as related but distinct checks: a spelling correction can change line length, and a timing adjustment can change reading speed.

Keep a review copy and a clean delivery copy so unresolved notes are not accidentally published as spoken text. Mark unclear passages for follow-up rather than replacing them with guesses. If a reviewer cannot hear a phrase clearly, a fluent-looking automatic sentence should not resolve the uncertainty by itself.

Test the export after any conversion. A plain SRT-to-VTT conversion can preserve cue text and timestamps while dropping styling or positioning. A platform may support only part of a format. The final player is the place to verify what the audience will actually receive.

Worked example

An illustrative cue contains 60 characters, including spaces and punctuation, and is displayed for 3 seconds. Its calculated speed is 60 ÷ 3 = 20 characters per second. That reaches the adult English SDH setting used by the checker. It does not tell you whether the cue is split well, whether a speaker name is missing, or whether it covers important on-screen text. If the cue is shortened to 2 seconds, the calculated rate becomes 30 characters per second. Review timing and wording under the destination’s rules rather than extending it blindly into the next exchange.

Open the local transcriber to test an authorized recording and review the draft against the sound.

Review the viewing experience after the file passes

Play the video with captions enabled and listen for any information that the text fails to represent. Then consider the experience without relying on sound. Does the viewer know when an off-screen person speaks, why a character reacts, or what a meaningful alert indicates? The answer can reveal omissions that a text-only checker cannot see.

Inspect visual conflicts too. A caption may cover a name label, a diagram annotation or an action needed to understand the scene. The available controls depend on the player and delivery format. If the project needs precise placement, use a format and authoring workflow that can preserve it rather than assuming plain cue conversion will do so.

Review a converted file again after import. A platform may normalize line breaks, ignore unsupported styling or apply viewer-controlled font settings. The original file can be valid while the displayed result differs. Test the actual published presentation when possible, especially around busy scenes and dense text.

Keep a record of the applicable guide and the decisions made for exceptions. If a cue cannot satisfy every constraint without losing essential meaning, an editor needs to make a contextual choice under the project’s requirements. The automatic tool should help locate the problem and expose its timing, not silently claim to resolve the tradeoff. A finished caption deliverable combines faithful content, readable timing and appropriate presentation. It is not simply a transcript with a subtitle extension.