What Audio Description Actually Is
Audio description is a spoken narration track, inserted into the natural pauses of a video's existing audio, describing visual information a blind or low-vision viewer would otherwise miss entirely. It is a required element of full accessibility compliance at higher conformance levels, as covered in more detail in the discussion of WCAG video compliance elsewhere in this series, and it is also one of the most commonly under-resourced accessibility deliverables, largely because writing it well is a genuinely distinct skill from ordinary scriptwriting or captioning, with its own specific constraints and craft.
The fundamental constraint that shapes everything else about audio description writing is that it has to fit into gaps that already exist in the original recording — the pauses between lines of dialogue, the quiet moments in a scene — without extending the video's runtime or talking over existing dialogue. This is a genuinely different writing discipline from most other content work: you are not deciding how much space your content needs and then finding room for it; you are measuring the room that already exists and writing content precisely sized to fit inside it.
What to Describe
Describe visual information that is necessary to understand the content and is not already conveyed through dialogue or other audio, which is the core selection principle and the one most often misapplied by inexperienced describers who default to narrating everything visible rather than everything necessary. A shot of a character nodding in agreement while saying "yes" needs no description, since the dialogue already conveys the meaning; a shot of a character shaking their head while saying nothing at all needs description, since without it the moment is lost entirely to a listener.
Prioritize actions, expressions, and visual information that affects plot or meaning over purely decorative or atmospheric visual detail, particularly under real time constraints, since available pause time is finite and a describer choosing what to include when there is not room for everything should favor content the audience actually needs to follow the story or understand the point being made, over incidental visual texture that would be nice to include but is not load-bearing.
Describe on-screen text that carries information not otherwise spoken aloud — a title card, a caption, a sign, a piece of correspondence shown on screen — since a blind or low-vision viewer has no other way to access this information, and it is easy for a sighted describer, who absorbed this text visually without conscious effort, to simply forget it needs describing at all.
Identify speakers when it is not otherwise clear who is speaking, similar to the speaker identification requirement covered in the discussion of SDH captions elsewhere in this series, though audio description handles this through natural narrative phrasing — "Sarah frowns" rather than a bracketed speaker label — since it is a spoken narration track rather than a text caption, and the stylistic conventions differ accordingly even where the underlying informational need is closely related.
Avoid describing what dialogue or other audio already makes clear, resisting the instinct to add description as a form of general enrichment or explanation rather than as a targeted fill for a genuine information gap — over-description wastes scarce pause time on redundant information and can crowd out something that was genuinely needed elsewhere in the same scene.
Fitting Description Into Available Time
Identify and measure every usable pause in the original audio before writing a single word of description, treating this as a distinct, sequential first step rather than writing description first and then discovering it does not fit — a pause inventory, timestamped and measured in advance, is the actual constraint the entire script has to work within, and writing against it from the start avoids the wasted effort of drafting description that then has to be cut down after the fact to fit a gap that was never actually measured.
Write to the available time, not to the ideal level of detail, which means accepting that some visual information genuinely cannot be described given the pauses available in a specific scene, and prioritizing accordingly rather than attempting to cram more content into a gap than natural, comprehensible speaking pace actually allows — description delivered too quickly to fit an inadequate gap is a worse outcome than shorter, clearer description that respects a natural speaking pace within the time genuinely available.
Consider extended audio description for content where the available pauses are genuinely insufficient for the visual complexity involved, a distinct technique that pauses the video itself to insert additional description time beyond what the original recording's natural gaps provide, appropriate for visually dense content — an instructional video demonstrating a complex physical process, for instance — where standard audio description's fit-into-existing-gaps constraint would force unacceptable compromises in what can actually be conveyed.
Time description precisely against the actual video, not against an estimate, since even a small timing misalignment between spoken description and the visual moment it refers to undermines the description's usefulness, and this is a case where the same kind of precise timing verification discussed for subtitle work elsewhere in this series applies with equal or greater force to audio description specifically.
Objectivity and Tone
Describe what is visible, not what it means or how the describer interprets it, which is a foundational principle of audio description craft: "she clenches her fists" is objective description of a visible action, while "she is furious" is an interpretation that the audience should be allowed to draw themselves from the same visual information a sighted viewer would use to reach that same conclusion independently, rather than having the interpretation handed to them pre-digested.
This objectivity principle has real limits and requires judgment rather than being an absolute rule applied mechanically in every case, since some visual information is genuinely ambiguous even to a sighted viewer, and in these specific cases a describer may need to convey a reasonable interpretation rather than an impossible, purely mechanical description of raw visual data with no framing at all — the skill is knowing where description can and should remain objective, and where some interpretive framing is genuinely necessary to convey what a sighted viewer would actually understand from the same shot.
Match the tone and register of the description to the tone of the content itself, since a flat, clinical narration style appropriate for a technical demonstration video would feel jarring and tonally wrong applied to an emotionally charged dramatic scene, and the describer's voice and word choice should support the content's own intended emotional register rather than working against it or flattening it uniformly regardless of context.
Use present tense and active, concrete language, generally the established convention in professional audio description practice, since it reads and sounds more immediate and natural than a more distant, passive, or overly formal narration style would, and it more closely matches the real-time nature of the visual experience being described as it actually happens.
Translating Audio Description
Audio description scripts need translation like any other spoken content, and the same timing constraint that shaped the original description applies again, independently, in every target language, meaning a well-fitted English audio description script translated into a language with meaningfully different text expansion characteristics, as covered throughout this series for different language pairs, may no longer fit the same pause it was originally written for, and this needs to be checked and potentially rewritten rather than assumed to transfer automatically.
Translate for meaning and timing fit together, not for literal word-for-word accuracy alone, since a literal translation that accurately conveys every word of the source description but no longer fits the available pause in the target language has failed at the actual job, in the same way an overly literal dubbing translation that ignores lip-sync timing fails at its actual job even while being technically accurate to the source text.
Consider that some visual or cultural references in the original description may need adaptation for a different market, similar in kind to the broader cultural adaptation considerations covered elsewhere in this series, since a description referencing a culturally specific visual detail that would be unfamiliar or require additional unpacking for a different audience may need to be reframed rather than translated literally, to actually achieve the same communicative effect for its new audience that the original achieved for its own.
Where audio description exists for a piece of content that is also being dubbed into other languages, plan the audio description translation and the dubbing work together rather than as entirely separate, uncoordinated workstreams, since both need to fit around the same underlying pauses and pacing in the source video, and coordinating them avoids each workstream independently discovering and separately solving the same fundamental timing constraint problem without benefiting from what the other has already learned about it.
Working With Automation
Automated tools can assist with identifying candidate pauses in the audio track for a describer to review, using the same kind of voice activity detection covered elsewhere in this series to flag silence and low-activity segments as candidate description windows, which speeds up the otherwise time-consuming manual process of scanning through a full video to build the pause inventory that good description writing depends on.
Automated scene and object detection can assist with flagging moments that likely warrant description consideration — a scene change, a new character entering frame, an on-screen text element appearing — as candidates for a human describer to evaluate and decide whether description is actually warranted, though the actual judgment about what specifically to describe, how to phrase it, and whether it is genuinely necessary given the surrounding dialogue remains a human writing and editorial task that current automation does not reliably perform well on its own.
Text-to-speech generation of the description narration itself is increasingly viable for audio description delivery, similar to voice generation used elsewhere in this series, provided the underlying script was well-written by a human with genuine audio description craft and the generated voice is clear, appropriately paced, and does not itself introduce any new timing problems relative to the carefully measured pauses the script was written to fit.
A Working Checklist
- Describe visual information necessary to understanding, not everything visible.
- Prioritize plot- and meaning-relevant visual content over decorative detail under time pressure.
- Describe on-screen text that carries information not otherwise spoken.
- Identify unclear speakers through natural narrative phrasing rather than bracketed labels.
- Avoid describing what dialogue or other audio already makes clear.
- Build a timestamped pause inventory before writing any description.
- Write to the available time rather than to an ideal level of detail.
- Consider extended audio description for visually dense content where natural pauses are insufficient.
- Time description precisely against the actual video rather than an estimate.
- Describe objectively where possible, using interpretive framing only where genuinely necessary for clarity.
- Match description tone and register to the content's own tone.
- Use present tense and active, concrete language as the general convention.
- Translate audio description for meaning and timing fit together, not literal accuracy alone.
- Re-verify that translated description still fits its original pause given target-language text expansion.
- Adapt culturally specific visual references where a literal translation would not convey the same effect.
- Coordinate audio description translation with dubbing work on the same content rather than as separate workstreams.
- Use automated pause and scene detection to assist a human describer, not to replace editorial judgment.
Frequently Asked Questions
What is the difference between audio description and a full audio description script for someone who cannot see at all?
They are usually the same thing described from two angles — audio description is written specifically to convey what a sighted viewer sees, fitted into existing pauses in the audio, so that a blind or low-vision listener receives equivalent information through sound alone. Extended audio description is a distinct technique that pauses the video to insert additional description time when the natural pauses are not sufficient for genuinely visually dense content, used as an exception rather than the default approach.
How do I decide what to describe when there isn't enough time to describe everything visible?
Prioritize what is necessary to understand the plot or meaning of the content over decorative or atmospheric visual detail. Build a timestamped inventory of actual available pauses before writing anything, and write to that measured time rather than to an ideal level of completeness. Description delivered too quickly to fit an inadequate gap is worse than shorter, clearer description that respects natural speaking pace within the time genuinely available.
Should audio description ever include interpretation rather than pure objective description?
Mostly not, but with real limits rather than as an absolute rule. The standard principle is describing what is visible rather than what it means, letting the audience draw their own conclusions the way a sighted viewer would from the same visual information. Some visual information is genuinely ambiguous even to a sighted viewer, and in those specific cases some interpretive framing may be necessary to convey what a sighted viewer would actually understand, which requires judgment rather than mechanical rule-following.
Does a translated audio description script need to be re-timed?
Yes, and this is commonly overlooked. The original script was fitted precisely to available pauses in the source language, but the same text expansion considerations that affect subtitles and dubbing across languages, covered throughout this series, apply to audio description translation too. A literal translation may no longer fit the pause it was written for, and translating for meaning and timing fit together, rather than word-for-word accuracy alone, is what actually solves this.
Can audio description be generated automatically?
Partially. Automated tools can identify candidate pauses and flag moments likely to warrant description, using techniques like voice activity detection and scene change detection, which speeds up the pause-inventory and candidate-identification work. The actual judgment about what to describe, how to phrase it objectively and concisely, and whether it is genuinely necessary given surrounding dialogue remains a human writing and editorial skill that current automation does not reliably replace.
Should audio description translation be coordinated with dubbing work on the same video?
Yes, when both exist for the same content. Both workstreams need to fit around the exact same underlying pauses and pacing in the source video, and coordinating them avoids each one independently rediscovering and separately solving the same fundamental timing constraint without benefiting from what the other workstream has already learned about the video's specific pause structure.
Related reading: Video Accessibility Guide | WCAG Video Compliance | SDH Captions Explained
