apply-media-captions
Use when publishing video, audio, or any time-based media — to provide captions, transcripts, and audio descriptions for users who cannot hear or see the media content.
SKILL.mdskills/design/ui-ux/skills/apply-media-captions/SKILL.md
GitHubApply Media Captions
Provide synchronized captions for video, transcripts for audio, and audio descriptions for visual-only content.
Why This Is Best Practice
Adopted by: WCAG 2.1 SCs 1.2.1–1.2.5 are Level A/AA — legally required by Section 508 (US), ADA, EU EN 301 549, and the UK Equality Act. Netflix was sued and settled for $755,000 in 2012 for missing captions (NAD v. Netflix) — the landmark case that accelerated industry adoption. All major streaming platforms (YouTube, Vimeo, AWS Elemental MediaConvert) now provide auto-captioning as baseline infrastructure. Impact: WHO estimates 1.5 billion people have hearing loss. Captions also benefit users in noisy environments, non-native language speakers, and users with cognitive disabilities. Facebook data showed 85% of videos are watched without sound — captions improve engagement for all users, not just those with disabilities. Why best: Auto-generated captions alone are insufficient — accuracy rates of 80–90% produce too many errors for compliance. Human-verified captions or caption editing is required for WCAG conformance.
Sources: W3C WCAG 2.1 SC 1.2.1–1.2.5 (2018); NAD v. Netflix settlement (2012); Facebook Captions research (2016); W3C WebVTT specification
Steps
Step 1: Identify what each media type requires
| Media type | Level A requirement | Level AA addition |
|---|---|---|
| Pre-recorded audio-only | Transcript | — |
| Pre-recorded video-only | Audio description OR transcript | — |
| Pre-recorded video+audio | Captions | Audio description |
| Live video+audio | Live captions | — |
Step 2: Add synchronized captions to video using WebVTT
<video controls>
<source src="product-demo.mp4" type="video/mp4">
<track kind="captions"
src="product-demo.en.vtt"
srclang="en"
label="English"
default>
</video>
WebVTT format:
WEBVTT
00:00:01.000 --> 00:00:04.000
Welcome to the product demo.
00:00:04.500 --> 00:00:08.000
Today we'll cover the three main features.
NOTE Speaker change
00:00:08.500 --> 00:00:12.000
[Jane] First, let's look at the dashboard.
Caption quality requirements:
- Synchronize within 2 seconds of speech
- Identify speakers when more than one person speaks:
[Speaker Name] - Describe significant non-speech audio:
[applause],[alarm beeping] - Do not exceed 42 characters per line, 2 lines per caption
Step 3: Write transcripts for audio-only content
<audio controls src="podcast-ep-42.mp3"></audio>
<details>
<summary>Transcript — Episode 42</summary>
<p><strong>Host:</strong> Welcome to episode 42...</p>
<p><strong>Guest:</strong> Thank you for having me...</p>
</details>
Transcripts must include all speech, speaker identification, and meaningful non-speech sounds. A transcript linked adjacent to the player satisfies WCAG 1.2.1.
Step 4: Add audio descriptions for visual content in video
When video conveys information through visuals not described in dialogue, add an audio description track:
<video controls>
<source src="tutorial.mp4" type="video/mp4">
<track kind="captions" src="tutorial.en.vtt" srclang="en" label="English captions" default>
<track kind="descriptions" src="tutorial.en.desc.vtt" srclang="en" label="Audio descriptions">
</video>
Or provide a separate audio-described version as a link adjacent to the player.
Step 5: Verify caption quality before publishing
Checklist:
- Captions are synchronized (within 2s)
- All speech is captured — no paraphrasing
- Speakers identified where more than one voice
- Non-speech audio described where meaningful
- Captions do not overlap or crowd the screen
- Auto-captions reviewed and corrected (do not ship raw auto-captions)
When NOT to Use
- Purely decorative background video (no meaningful audio or visual content) — mark with
aria-hidden="true"andmuted autoplay. No captions required. - Live video where real-time captions are technically infeasible — document the exception and provide a text alternative as soon as practically possible after the live event.
Common Mistakes
Shipping auto-generated captions without review. Auto-caption accuracy of 80% means 1 in 5 words is wrong. This fails WCAG 1.2.2 and produces an unreliable experience.
Captions that only cover speech. A video where an alarm sounds and the character reacts — but the captions say nothing about the alarm — fails to convey equivalent information.
Transcript behind a paywall or login. If the media itself is accessible without login, so must the transcript be.