Auto-captions aren't accessible captions
Auto-generated captions are a starting point, not a finish line. This sample shows the cleanup pass I teach faculty to apply — press the button to transform the raw output, then reveal why each fix matters for WCAG conformance.
Before — raw auto-captions
so today were going to talk
about excess ability in your
course videos um the first
thing you want to
check is weather your captions
are accurate because студents
who are deaf or hard of hearing
rely on them completely and to
sync with the audio ok so lets
look at the three most common
airers that auto captioning
makes in lecture recordings
After — cleaned for WCAG
Press "Apply the cleanup" to see
the corrected captions.
What changed, and the WCAG reasoning
- Misheard words corrected ("excess ability" → "accessibility," "weather" → "whether," "airers" → "errors") — caption accuracy is the substance of WCAG 1.2.2; a caption that says the wrong words fails its entire purpose.
- Filler removed ("um," "ok so") — verbatim isn't the goal for edited educational video; readable meaning is. Cleaned captions reduce cognitive load without changing content.
- Punctuation and capitalization added — sentence boundaries are how readers parse meaning at reading speed; a wall of lowercase text forces viewers to do the parsing themselves.
- Line lengths balanced (roughly 32–42 characters, breaking at natural phrase boundaries) — so captions are readable in the time they're on screen and don't split phrases mid-thought.
- Garbled characters fixed — auto-captioning occasionally inserts wrong-alphabet artifacts; every character has to be real text for screen readers consuming the transcript downstream.