Audio description: what it is and how to add it to a video

What is audio description?

Audio description, also called described video or video description, is extra narration that carries what the picture tells and the soundtrack does not: who has walked in, what a face does, where the scene now is, what a sign on screen says. The W3C's Web Content Accessibility Guidelines define it as narration added to the soundtrack to describe important visual details, placed in the existing pauses in dialogue.

It is not captions. Captions turn sound into text for people who cannot hear; audio description turns pictures into speech for people who cannot see. A video that serves everyone has both. Closed captions, subtitles and SDH covers the captions side.

Who needs audio description?

Blind and partially sighted people first. Ofcom's code on television access services defines audio description as a commentary that describes what is happening on screen, primarily for people with sight loss. It also helps people with some cognitive and learning disabilities, and anyone listening without watching, such as a viewer cooking with the television in the next room.

Whether you must provide it depends on where the video goes. UK broadcasters have quotas set by Ofcom; streamers such as Netflix ask for it on many deliveries; and public bodies and companies that follow WCAG need it for prerecorded video on their websites.

What do the guidelines ask for?

Different bodies set different parts of the job:

The craft rules agree more than they differ: describe what matters to following the story, in the order it appears, in plain words, without interpreting, and never step on the dialogue.

How is audio description made?

In four stages, whoever does it:

1. Find the gaps. Mark every stretch with no dialogue and no important sound, and how long each is.
2. Write the script. For each gap, the most important visual change since the last description, short enough to say in the time. Present tense, plain words: "She opens the door. Snow blows in."
3. Voice it. A clear, neutral voice that is distinct from the people on screen. Netflix's guide warns against a monotonous or sing-song delivery.
4. Mix it. Dip the programme under each line, bring it back afterwards, and deliver the result as a separate described track or a described version.

The writing takes the longest. A good describer chooses what to leave out: a viewer can only take in so much between two lines of dialogue.

How do you add audio description in Sharpy?

In two steps, both on your Mac: Sharpy writes the description, then voices and mixes it.

The script. Sharpy watches the picture and writes what it shows in the gaps between the words, as a WebVTT descriptions file beside the video. Its rules are the describer's: present tense, what is seen and not felt, nothing the sound already says, and only what changed since the last description. Each cue is at least three words, and the pace stays at or under 160 words a minute. A long silent stretch is broken at the picture's own cuts, so a 19-second sequence of Lisbon became four descriptions: a tram in a narrow street, a woman walking a patterned square, a silhouette against the sunset by the water. A typical line:

A man sits at a desk in a room. He wears glasses and a brown shirt. A keyboard lies before him.

The fit is measured, not estimated: each candidate line is spoken in Sharpy's voice to check it fits its gap. On its maker's own take, over four runs, no description ran into any of the 429 spoken words and none went faster than 160 words a minute. A film with no words at all is treated as one long gap.

Read the draft. Sharpy labels every script a machine's, because it is: its wording is vague where the picture is simply a man talking. Read it against the picture, fix names and anything that matters to the story, and cut what does not.

The voice and the mix. Ask the agent for a described version, and its narrate tool speaks each line in Sharpy's own voice, dips the programme under it by the dip rules of EBU TR 084, and writes a described movie: the original picture and sound untouched, plus a second audio track tagged as audio description so a player lists it, and the narration alone as a stem. Netflix deliveries use Netflix's own dip level, and a description script in the W3C's DAPT format, with its gain instructions, is written beside the narration stem.

What should you check before you publish?

Sharpy checks its own work first. If a line cannot fit its gap at any pace, it is refused with the seconds it needs, so you shorten the description rather than speaking over dialogue. Every voiced line is transcribed back and any lost word reported.

Measured on a 3 minute 41 second reel with three descriptions, every line came back word for word, at 156, 160 and 159 words a minute. The dips were set at −9, −19 and −19 dB and measured in the file at −9.00, −19.00 and −19.07 dB; outside the descriptions the programme was untouched to 0.002 dB, and the finished mix sat at −20.0 LUFS against the programme's −19.9. The whole run took 57 seconds on the base Mac mini M4 with 16 GB.

Then check what no measurement can:

• Listen with your eyes closed. Can you follow the story?
• Check names and on-screen text. Every sign and title that matters should be read out.
• Ask someone who uses audio description, if you can, before a public release. RNIB's 2024 study found listeners more accepting of synthetic voices for factual content than for drama and entertainment.

How do you publish a described video?

As a second track where the platform allows it, or as a second video where it does not. YouTube's help describes adding a descriptive audio track in YouTube Studio under Languages, for creators who have multi-language audio: the file must be audio only, about the length of the video, and the original audio in that language has to be uploaded first. Without that feature, publish a described version as its own video and say so in its title. Players that read a movie's tracks list Sharpy's described track as audio description.

Frequently asked questions

01

What is audio description?

A narration, spoken in the gaps between the dialogue, that describes what is happening on screen for people who cannot see it.

02

Who needs audio description?

Blind and partially sighted people above all, and also some people with cognitive disabilities and anyone listening without watching.

03

How is audio description made?

The gaps in the dialogue are found, a script is written for them, a voice reads it, and the programme is dipped under each line in the mix.

04

What is the difference between audio description and captions?

Captions write down the sound for people who cannot hear. Audio description speaks the picture for people who cannot see.

05

What is extended audio description?

Description that pauses the video to make room, used when the gaps are too short. WCAG asks for it at Level AAA.

06

Is an AI audio description good enough to publish?

As a first draft, yes; unread, no. Sharpy labels its script a machine's for that reason, so a person reads it before it goes out.

Sources

Checked on 18 October 2026.

• W3C: Understanding WCAG 2.2, Audio Description (Prerecorded)
• Netflix: Audio Description Style Guide v2.5
• EBU TR 084: audio description dip levels for non-sound engineers
• Ofcom: TV access services review, code and best practice guidelines
• RNIB: synthetic speech for audio description, study report 2024
• W3C: DAPT, the timed-text profile for dubbing and audio description scripts
• YouTube Help: add audio descriptions