The Reality of YouTube Autocaptions

[jesse_the_k]

(x-posted from my personal journal)

You may have seen Google/YouTube announce the magic of auto-captioning last November.

Gee whiz, they even had a deaf programmer write the blog entry. Things are good, right?

Watch this Bill Moyers interview with David Simon on YouTube. It's got captions. They're automagically generated with voice recognition. Compare the audio tracks and the caption track and be stunned at the high level of errors. Notice that White speakers' words are around 80% correct and Black speakers' words more like 30% correct.

Yes, it takes time to make good on technology's promise. In the meantime, disabled people put up with sub-standard services—and often at premium prices. When they're perfected, they'll be generally available.

These bad captions are particularly frustrating because the original sources were already captioned! Since the 1980s all network PBS (US public television) has been captioned; the same has been true for all HBO (paid US cable network) productions since 1995.

Arghhh.

Comments

[codeman38]

Yeah... I toyed with the auto-captioning a while back, and it was like watching a really bad bootleg DVD.

I'm tied about what my favorite mistranscription is. In one University of California video, "we're doing rock climbing as well as table tennis" was captioned as "for doing what I mean as well as democrats". But that's nothing compared to what happened to the alphabet song.


These bad captions are particularly frustrating because the original sources were already captioned! Since the 1980s all network PBS (US public television) has been captioned; the same has been true for all HBO (paid US cable network) productions since 1995.

This. This is what gets me the most, not just about YouTube but about online content in general.

I can understand when something isn't captioned because it's a web original thing, but when it's something that's already been captioned for TV? There are ways to rip it and convert it for online use. I've done this from DVD recordings using some open-source tools. And WGBH, the PBS station in Boston, invented one of the ripping tools for caption data, so PBS in particular has no excuse.

[exor674]

Not that I'm defending PBS, but I wonder if it's a licensing thing or something. $company_who_captioned is all "waah, we don't you to put our captions on YouTube"

[codeman38]

I've actually considered that; a lot more of their programming seems to be outsourced to external captioning companies lately. But a good deal of PBS' programming is still captioned in-house at WGBH, for which that excuse hardly seems to hold water.

[treeowl]

Has anyone contacted PBS about this?

Subtitle removal

[roserodent.myopenid.com]

A similar thing which drives me to distraction is when a series I was watching one a mainstream channel is repeated on one of the less popular channels. I think 'excellent' because I missed episode 2, only to find that although the original version was subtitled, the repeat is not! How is it within the terms of their service contract (which actually requires a minimum percentage of subtitled programming) to REMOVE subtitles from a program which has them? OK, it's expensive to add subtitles, but how expensive is it to show the subtitles which have already been done?