The Gap Most Localization Programs Leave
A video localized into ten languages with subtitles in each is often described as accessible. It is not, or at least not fully.
Subtitles serve viewers who cannot hear the audio only if they contain the non-speech information a hearing viewer receives. Most translated subtitle tracks contain dialogue and nothing else.
Nothing in that ten-language localization serves blind and low-vision viewers at all. Visual information — what is shown, demonstrated, or displayed on screen — is conveyed only visually, and translating the audio does not change that.
The result is a program that expanded language access substantially while leaving disability access roughly where it started, multiplied across every new language. Each localization decision that omits accessibility compounds the omission.
Building accessibility into the localization workflow rather than treating it as a separate project is both cheaper and more effective, because the two share most of their production steps.
What Audio Description Is
Audio description is a narration track that describes visual information that is not conveyed by the existing audio. It is delivered as an additional audio track that a viewer can select.
The description covers what a viewer who cannot see the screen would otherwise miss: actions, scene changes, on-screen text, graphics, facial expressions where they carry meaning, and the appearance of people and settings where relevant.
It is not a description of everything. Good description is selective, prioritizing what matters for understanding and skipping what does not. Over-description is a real failure mode: filling every gap with detail produces a track that is exhausting to listen to and obscures the content it is meant to support.
Two variants exist in practice. Standard audio description fits within the natural gaps in the existing audio, adding nothing to the runtime. Extended audio description pauses the video where necessary to allow longer description, changing the runtime — appropriate for dense instructional content where fitting description into gaps would mean omitting essential information.
Writing Description
Description writing is a distinct skill with established conventions, and applying them makes the difference between a track that helps and one that intrudes.
Describe what is shown, not what it means. The description should give the listener the visual information and let them draw conclusions, in the same way a sighted viewer does. Interpreting on the viewer's behalf removes their agency and often imposes a reading the visuals do not support.
Prioritize ruthlessly. Gaps are short. Decide what the listener needs to follow the content and omit the rest. In instructional content, the priority is usually actions and on-screen text. In narrative content, it is usually actions, expressions, and setting changes.
Use present tense and active voice. Convention favors describing action as it happens.
Be concise. Description competes for limited time. Economical phrasing fits more information into the available gap.
Read on-screen text. Text that appears visually — titles, labels, captions, slide content, data in charts — is inaccessible without description. This is frequently the highest-value description in instructional and corporate content.
Do not talk over dialogue. Description belongs in gaps. Overlapping the primary audio makes both harder to follow.
Identify speakers when it is not obvious. In content with multiple speakers who are visually distinguished but not aurally identified, description supplies what the visuals were doing.
The Gap Problem in Translated Content
Audio description depends on gaps in the primary audio, and translation changes where those gaps are.
This is the core interaction between description and localization. A description script written to fit the gaps in English audio will not fit the gaps in the German version, because German dialogue expands and fills space that was previously silent. Languages that contract, like Korean or Japanese, open new gaps in different places.
The practical consequences:
Description cannot simply be translated and dropped in. The translated description must be re-timed to the gaps in the translated audio, and in expanded languages there may not be enough room for the same content.
Description may need condensing per language. Where a translated primary audio track leaves less room, the description for that language has to say less. Deciding what to cut is an editorial decision that should follow the same priority logic used in the original.
Extended description may be necessary in some languages and not others. If a language expands enough that essential description no longer fits, pausing the video becomes the only way to include it.
Order matters. Translate and generate the primary audio first, then write or adapt description against the actual resulting gaps. Writing description against the source timing and then discovering it does not fit is the common and avoidable failure.
Producing Description at Scale
Description has historically been expensive because it required a writer to watch content, identify gaps, and write to precise timings, followed by studio narration.
Synthesized voice changes the production economics substantially. Once a description script exists with timings, generating the audio is inexpensive and repeatable, which makes description viable for content volumes that would never have justified studio narration.
The writing remains the substantive work. Automated description generation from video analysis exists and is improving, but the editorial judgment about what matters — the prioritization that separates useful description from exhaustive narration — is where quality is determined, and it benefits from human review.
A workable production model: generate a first-pass description draft, have a writer edit it for priority and phrasing, time it to actual gaps, and generate the audio synthetically. This is far cheaper than fully manual production while preserving the editorial control that quality requires.
For localization, the same model applies per language, with the description re-timed against that language's audio.
Voice Selection for Description
The description voice should be clearly distinguishable from the content's primary voices so listeners can tell description from dialogue without effort.
Convention favors a neutral, measured delivery that does not compete with the content emotionally. Description is informational, and an expressive read draws attention to itself.
Where the primary audio uses a particular voice, choose a description voice with different characteristics — different pitch range, different gender, or noticeably different timbre. In localized content, this needs checking per language, since the primary voice differs.
Levels matter. Description should sit clearly above the background audio without being jarringly louder than dialogue. Where description overlaps music or effects, ducking those elements slightly under the description improves intelligibility.
Captions as the Other Half
Audio description addresses one access need; captions address another, and localized programs frequently under-serve the second as well.
A caption track intended for deaf and hard-of-hearing viewers must include more than dialogue: speaker identification, relevant sound effects, music cues where they carry meaning, and indications of tone where it changes interpretation.
Producing this in every localized language, rather than only in the source language, is what full localization means. A Spanish subtitle track containing only dialogue serves hearing Spanish speakers. It does not serve deaf Spanish speakers, who need the same non-speech information their English-language counterparts receive.
The additional work is modest relative to the translation already being done: the non-speech elements are identified once in the source and translated alongside the dialogue.
Reading speed guidance matters here as well. Caption users cannot fall back on the audio, so captions that are too fast to read are a harder failure than subtitles for hearing viewers who can catch some content aurally.
Standards and Requirements
Accessibility requirements for video vary by jurisdiction and by sector, and they are increasingly enforced.
Common threads across regimes include obligations for captions on published video, obligations for audio description in some contexts, requirements for accessible players and controls, and standards specifying quality rather than mere presence.
Public sector bodies, broadcasters, education providers, and organizations serving the public frequently carry specific obligations. Commercial organizations may carry obligations under general accessibility or non-discrimination law even where no video-specific rule exists.
Where requirements apply, they generally apply to all language versions rather than only to the primary one. A program that provides English captions and audio description while providing only dialogue subtitles in other languages may be compliant in one market and not in another.
Determine the applicable requirements for each market you publish into rather than assuming the source market's rules apply everywhere.
Integrating Into the Localization Workflow
The efficient approach builds accessibility into the same pass as translation rather than treating it as follow-on work.
During transcription, identify non-speech audio elements alongside dialogue. This is the source data for caption non-speech entries and costs little at this stage.
During source preparation, identify all on-screen text and visual information requiring description. This is the same inventory needed for graphic localization, so the work serves two purposes.
Write the source description before translating, so that the description script goes through the same translation and review process as the dialogue.
Generate primary translated audio first, then re-time description against the resulting gaps per language.
Produce caption tracks with non-speech information in every language as part of the standard subtitle deliverable rather than as an extra.
Review accessibility tracks with the same rigor as primary content. A description track that was never checked is as likely to contain errors as any other generated output.
Designing Content to Need Less Description
The cheapest description is the description you do not have to write, and content design choices substantially affect how much is needed.
Narrate what you demonstrate. Instructional content where the presenter says what they are doing as they do it is largely self-describing. A tutorial that says "click the settings icon in the sidebar" needs no description of that action; one that shows the click in silence does.
Read on-screen text aloud. Titles, key figures, and slide content spoken in the primary narration serve every viewer and eliminate the highest-priority description need.
Avoid meaningful silence over visual information. Long silent passages showing critical visuals are the hardest case, because there is nothing to describe around and the description must carry the whole load.
Describe data rather than only displaying it. A chart that is discussed in the narration — what it shows, what the trend is, what the notable value is — is accessible without description and is better content for sighted viewers too.
This practice, sometimes called integrated or embedded description, produces content that needs little or no separate description track. It costs nothing at production time and saves description work in every language.
For content programs producing at volume, adopting these habits in scripting is the single highest-leverage accessibility decision available, because it reduces per-language description work across the entire future library.
Getting Started
Audit what exists. Most programs find they have captions in the source language, dialogue-only subtitles in other languages, and no description anywhere.
Close the caption gap first. Adding non-speech information to existing translated subtitle tracks is the cheapest meaningful accessibility improvement available and applies across the whole library.
Add description to the highest-value content next — typically instructional content where visual demonstration carries information that narration does not, since this is where the access gap is widest.
Build both into the standard workflow for new content, so the library stops accumulating inaccessible assets while the backlog is addressed.
Measure by asking. Accessibility quality is difficult to assess from the outside, and feedback from viewers who rely on these tracks is more informative than any internal check.
Set a standard and write it down. Without a documented description standard covering what gets described, in what order of priority, and at what level of detail, output varies by whoever produced it, and that variation is more disorienting to regular users of description than any individual imperfection.
Budget realistically rather than aspirationally. Description writing is the substantive cost and does not disappear because voice generation is cheap. A program that funds audio generation without funding the writing produces description tracks that were never edited for priority, which is a worse outcome than none at all for listeners who then have to sit through them.
Localization and accessibility are the same underlying project: making content usable by people it currently excludes. Programs that treat them as one workflow spend less and reach further than those that run them separately.
