Captions or audio description: when do you need both?

Start with the information a viewer would miss

Captions make speech and meaningful sounds available as timed text. Audio description makes essential visual information available through spoken words. A video may need both: captions do not automatically explain a silent action, and audio description does not replace the dialogue or sound information in a caption track.

For a prerecorded video you own or are authorized to adapt, ask two separate questions. What becomes unclear when the sound is unavailable? What becomes unclear when the picture is unavailable? List the missing information before choosing an export format or commissioning a service. This guide helps define that brief; it is not a certification that a particular video meets every accessibility requirement.

Distinguish three deliverables

W3C’s captions guidance describes synchronized text covering speech and relevant non-speech audio. A caption might communicate a doorbell or identify a voice that the picture does not make clear. Names such as captions, subtitles and SDH vary across regions and services, so inspect what the track actually contains. A dialogue-only translation can omit important sound information.

Audio description communicates information carried by the picture: a silent action, an onscreen label or a visual result that the existing soundtrack does not explain. W3C describes several ways to deliver it, including narration integrated into the original script, a described version of the video, or a separate track supported by the player.

A descriptive transcript is a reading document combining relevant audio and visual information in sequence. W3C identifies it as useful for people who are Deaf-blind and others who prefer or need text. It can complement playback, but a transcript link does not make an uncaptioned video captioned or create an audible description track.

Compare one scene in all three forms

Imagine an original, fictional film about sorting donated books. A presenter says, “That one goes here.” They place a book in a tray labeled REPAIR. A bell rings offscreen and they turn toward the door. The line, label, action and bell each contribute different information.

Captions should preserve the spoken line and the meaningful bell sound. Description should make the unspoken tray label and placement available without guessing why the book needs repair. A descriptive transcript can bring those elements together for reading. These examples are editorial demonstrations, not a transcription test or timed production script.

Do not silently rewrite the spoken caption as “Put this book in the repair tray.” That might improve a future script, but it is not what this recording says. If you can revise and rerecord the original narration, making the referent explicit may serve everyone. Otherwise plan a separate description solution that fits the final scene.

Fictional scene

Captions:
That one goes here.
[bell rings]

Description draft:
She places the book
in the tray marked REPAIR.

Descriptive transcript:
Presenter: That one goes here.
[Places the book
in the REPAIR tray.]
[A bell rings; she turns
toward the door.]

Choose description delivery before making files

For a new recording, review the script for visual references such as this, there and as shown. Say the essential action or result when it happens, where doing so fits the content. For an existing recording, first identify what is already clear from the soundtrack; repeating it can crowd out the information that is missing.

Ask the publishing team which description method the destination actually supports. A separate audio track, a separate described video and a timed description file require different production work. A WebVTT file used for captions does not become spoken description merely because its extension is supported. Text-based description needs a player that can deliver that function.

Check whether there is time to hear the additional words without obscuring dialogue or meaningful sounds. When the pauses are too short, discuss an extended described version or another supported approach with the production team. A script that looks concise on paper is not proof it will fit. Confirm the final playback behavior before approving the delivery method.

Keep the brief specific to one video version

Create a small external delivery note naming the final video, its language and revision. Record which moments need captions, which need description, who reviews each, and how viewers will find the outputs. For the book film, the repair-tray scene is a useful acceptance checkpoint because it exposes an unclear visual reference as well as a meaningful sound.

If the described version adds pauses or changes the edit, do not assume the original caption timing still fits it. Check the captions against that actual version and account for newly spoken description in the text alternative. Keep original and described deliverables clearly labeled so the publishing team cannot accidentally pair a revised video with an older track.

Delivery note
Video revision: __________
Caption language: ________
Description method: ______
Described version: _______
Transcript location: _____
Reviewers: _______________
Destination check: _______

Use Polyglot for the caption and text stages

On an Apple Silicon Mac running macOS 14.4 or later, Polyglot can create a timed speech draft from authorized media or import an SRT or WebVTT file for local review. You can edit and translate subtitle text, then export a separate timed file or transcript text. Review recognition and translation results against the source.

Polyglot does not promise to inspect the picture and write audio description, narrate a description script or mix a described soundtrack. Speech transcription also does not guarantee complete sound labels or speaker identification. Use suitable authoring and audio/video tools for those tasks; exporting transcript text is a starting point for a separately edited descriptive document.

Keep visual-description notes outside the speech track unless you are deliberately preparing the appropriate text alternative for a described version. This makes it easier to verify what was spoken, what was added from the picture and which deliverable each addition belongs to.

Review each route to the information

Check the captioned video with sound muted: can you follow speech, important sounds and speaker changes? Then listen to the intended described playback without relying on the picture: can you understand the essential actions, labels and results? Finally read the descriptive transcript independently. These passes help expose gaps; return to the full source to verify accuracy, and involve intended users where possible.

At the actual destination, confirm that captions can be selected and description can be found and heard through the promised method. Open the transcript link and verify it matches the same revision. In the book-film example, a working caption toggle still leaves a gap if the repair-tray label remains available only in the picture. Approve the information delivered, not simply the presence of three files.

Keep exploring.