A product demo can be perfectly edited and still leave someone out of the story. A chart animates. A customer clicks the one control that matters. A lower-third quietly names the speaker. If none of that reaches the soundtrack, a listener who cannot see the video has to guess.
Audio description closes that gap by adding concise, timed narration for visual information that the original audio does not already communicate. Vois gives an editor a local workspace for short description cues, voice auditions, and replacement takes while the picture edit evolves. It is not a second voiceover that recaps the whole film. It is a layer of useful evidence, placed where it belongs.
Start with the visual information that changes the meaning
The W3C guidance on description of visual information gives a useful test: imagine explaining the video to someone who cannot see it. What would they need to know to understand the point, not merely to picture every pixel?
That usually includes:
- actions that move the story or instruction forward
- people, objects, locations, and scene changes when they matter
- on-screen text, labels, links, names, prices, or data that carry information
- a chart's conclusion or a visual comparison that is not spoken aloud
- a reaction or gesture when it changes how dialogue should be understood
It usually does not include wallpaper, clothing, camera angles, or every facial expression. Detail has a cost: it takes time, can mask the original soundtrack, and can turn a focused description into noise.
Watch the video once without writing. On a second pass, make a cue sheet with the time range, the visible event, what the main audio already says, and the smallest useful description. A cue such as "00:42-00:45, warning banner appears, no spoken equivalent, read the banner" is much easier to script than a memory of "something important happened around the settings screen."
A talking-head interview may need very little if names, actions, and key visuals are already spoken. A tutorial, documentary, silent montage, or data-heavy presentation often needs more. The decision comes from what is missing from the audio, not from the genre label.
Write what is observable, not what you think it means
Good description is specific without pretending to know a person's inner state. W3C recommends objective, present-tense, active-voice wording. That keeps the narrator from becoming a commentator.
Compare these two cues:
The frustrated engineer realizes the dashboard is broken.
A red error banner covers the dashboard. The engineer closes the laptop.
The second version gives the listener the evidence. It leaves room for the listener to interpret the reaction. The same rule matters for identity, disability, race, gender expression, relationships, and any attribute that a video does not clearly establish. Describe what is relevant and observable. Do not fill gaps with assumptions.
Use the language the video gives you. Once the host says "Maya," say Maya rather than repeatedly describing her appearance. If a diagram contains a title and three numbers that matter, say the title and the conclusion rather than narrating its decorative colors. If a button label launches the next task, read the label.
Keep each cue short enough to say cleanly. Read the description beside the scene, out loud, before generating it. A sentence that looks tidy on a page can become a tongue twister when it has to fit between two lines of dialogue.
Make timing serve the original soundtrack
For standard description, the narration belongs in existing pauses around dialogue and other essential audio. W3C calls out actions, characters, scene changes, and on-screen text as common candidates, and warns against missing usable pauses for information that matters. The description should arrive with the visual event or just before it, not after the moment has passed.
Do not fill every quiet beat. A pause may contain music, a sound effect, room tone, or the breath that gives a scene its shape. Mark these as protected audio in your cue sheet. If a description overlaps a nonessential flourish, that is a creative and editorial decision. If it overlaps a line that carries the plot, it is a problem.
Sometimes there simply is not enough space. For a new training video, you can weave key visual facts into the presenter's script so they are part of the main audio. For an existing, densely narrated video, a separate described version or an extended-description treatment may be more honest than cramming narration over the top. The W3C media guide explains these options and the player support they require.
Generate a description take, then listen in context
A clear, steady voice is usually a better fit than a highly characterful performance. In Vois, start with a library voice that feels distinct from the program speakers, generate the description in short cues, and bring those takes into the video edit. The desktop-local workflow lets you keep the cue script and generation work on your machine while you iterate.
Do not judge a take in isolation. Place it against the actual dialogue, music, and effects. Check that the beginning and end of every cue fit its marked gap. If the description is hard to hear, solve the mix rather than making the narrator rush.
A useful mix pass asks:
- Can the listener understand every description word without losing essential original audio?
- Does the description begin at the right visual beat, or slightly before it when context is needed?
- Is the narration audibly distinct from dialogue but calm enough not to dramatize the scene?
- Have you lowered competing background audio only where the description needs room, then returned it to normal?
- Does a transition, sound effect, or music cue still communicate what it needs to communicate?
That last question catches a common mistake. Description is part of the soundtrack, not an overlay that gets a free pass over everything else.
Prompt your agent: "From this locked video, cue sheet, and approved descriptive transcript, prepare a Vois description plan. Use only observable visual facts that the original soundtrack does not already convey. Keep each cue within its marked gap, flag collisions with protected dialogue or sound, and return the cue list and proposed descriptions for editorial and accessibility review before audio generation."
Then make the release decision with people, not the agent:
- An editor checks every cue against the picture and original soundtrack.
- Blind or low-vision reviewers assess a near-final described cut and identify missing, late, or unnecessary information.
- The producer approves only the revised cue script and the selected description voice.
- The mix owner verifies the final described version, captions, and descriptive transcript against the locked picture.
Ask blind and low-vision reviewers to challenge the script
A cue sheet is a hypothesis, not proof that the description works. Invite blind or low-vision reviewers who reflect the video's intended audience, compensate them for their time, and let them experience a near-final cut without being coached through it.
Ask concrete questions: "What was unclear?" "Did any description arrive too late?" "What did you need to know but never hear?" "Which details felt unnecessary?" Their answers can expose the gap between an editor's visual memory and a listener's actual experience.
Keep a revision log that ties feedback to cue IDs and timecodes. That makes changes traceable when the picture edit shifts later. It also prevents a vague note such as "make accessibility better" from becoming a pile of untestable rewrites.
Review by people with relevant lived experience improves the work, but it is not a blanket compliance guarantee. The WCAG 2.2 explanation for prerecorded audio description is a technical reference; delivery requirements can also depend on the player, platform, contract, and jurisdiction. Confirm those requirements with the people responsible for the release.
Captions, transcripts, and description have different jobs
Do not treat the description track as a substitute for other media alternatives. Each answers a different access need.
| Deliverable | What it carries | How it is used |
|---|---|---|
| Audio description | Meaningful visual information not present in the main soundtrack | Heard with the video through an integrated track, a separate track, or a described version |
| Captions | Spoken dialogue, speaker identification, and meaningful non-speech audio | Read in sync with the video |
| Descriptive transcript | A text account of audio and meaningful visual information | Read independently, searched, downloaded, or used with assistive technology |
W3C's media accessibility overview treats captions, description, and transcripts as distinct components. Build the cue sheet so it can feed both the description script and the visual portions of the transcript. That reduces duplicate research while preserving the right output for each format.
For a broader look at audio alternatives beyond video, see our guide to accessible audio content. The goal is not one magic file. It is giving people more than one workable way into the material.
Hand off a package an editor can actually use
An audio file named final-final-description.wav does not tell an editor where it belongs. Export a tidy handoff package instead:
- Description master: a version that contains the timed description, named with the video version and language.
- Cue script: timecodes, cue IDs, final text, and any notes about protected audio.
- Individual cue files: when the editor needs to place or replace descriptions independently.
- Caption file and descriptive transcript: separate, reviewed assets with clear version names.
- Mix notes: where background audio was reduced, what was intentionally left uncovered, and the source video version used for review.
Keep the source video's edit version in every filename and request a final picture lock check before release. If the cut changes, timecodes can drift even when the words stay correct.
The first description you make does not need to solve every future video. Pick one scene with meaningful visual information, write a handful of cues, listen to them in the cut, and invite someone who will tell you what you missed. Get started with the Vois voice library, then download Vois when you are ready to make description part of production craft rather than an add-on at the finish line.
Sources
- W3C WAI: Description of visual information
- W3C WAI: Making audio and video media accessible
- W3C WAI: Understanding Success Criterion 1.2.5, Audio Description (Prerecorded)
The Vois Team