Captions and audio description.
On every piece of video on this site, without exception. This is a commitment rather than a feature, which means it is a reason not to publish something rather than something to publish alongside it.
What a commitment means here.
If a video cannot be captioned, it does not go up. That is the difference between a commitment and a feature, and it is the only sentence on this page that matters.
What is on what.
Live broadcasts
- Live captions
- Every oneMachine-generated, with the error rate published below.
- Corrected captions
- Within 24 hoursBy a person, against the transcript.
- Audio description
- On the archive copyAdded before the recording is published.
- Transcript
- Published with the archive copySearchable, and the source for write-ups.
- Audio-only stream
- Every broadcast64 kbps, one button.
Podcast
- Transcript
- Every episodePublished with it, linked in the feed.
- Chapter marks
- Every episodeFour, so a listener can skip to the part they want.
- Speaker labels
- YesIn the transcript, which matters more than people assume.
Everything else on the site
- Camera feeds
- Captioned where there is speechA silent field camera has nothing to caption.
- Scene plates and diagrams
- Alt text, alwaysWritten by a person, not generated.
- Motion
- Respects reduced-motionEvery animated block has a still equivalent.
How good the captions actually are.
Why the live number is published even though it is the bad one.
Ninety-four per cent sounds respectable and is not. At ninety-four per cent, roughly one word in seventeen is wrong, and in a broadcast full of agricultural vocabulary the wrong words cluster exactly where the meaning is.
Machine captioning on a live feed from a grain store, with a combine running, is genuinely hard. Publishing the number is the only way to be honest that a live caption is an aid rather than an equivalent, and the reason the corrected version exists at all.
The commitment that matters is the second number and the twenty-four hours. The first number is published because hiding it would imply the live captions are better than they are, and somebody relying on them deserves to know.
At ninety-four per cent, one word in seventeen is wrong, and the wrong ones cluster where the meaning is.
What is committed, and what is not.
Committed
- Captions on every video, live and archived
- Human-corrected within 24 hours
- Audio description on every archive copy
- A searchable transcript with speaker labels
- An audio-only stream at 64 kbps
- Alt text written by a person
Not yet, and said so
- A promise that live captions are perfect
- Sign language interpretation, yet
- Audio description on the live stream
- Captions in languages other than English
The questions this page gets.
Because doing it badly is worse than not doing it, and doing it properly for seven broadcasts a year is a commitment we have not yet made. It is the largest gap on this page and it is listed as one rather than left off.
A person, for every scene plate and diagram on this site. Generated alt text describes what is in an image; a person describes what the image is for.
Transcript with every episode, speaker-labelled, linked in the feed and published on the page.
The wider accessibility statement.
Everything else about how this site behaves, including where it falls short.
