cecilkorik@lemmy.ca 4 days ago
Youtube’s AI generated subtitles are the worst. They’re borderline useless, I have to spend as much time trying to explain to my hard-of-hearing grandmother what the people actually said instead of the confusing grammarless gobbledygook that the subtitles said, as I would if we just had subtitles off and I had to simply tell her the words she couldn’t hear. I hate it.
Every single time I try youtube subtitles because I didn’t understand something, they end up being useless AI subtitles and I am faced with disappointment again.
I could understand how it would be helpful for the hearing impaired watching clearly-enunciated youtube videos, though. Should still hire professionals for movies.
The automated captions have been there for ages, and yeah, they often have a lot of mistakes especially if there are other sounds in the background.
But what sucks even more are those auto-dubbed videos, because often they are made from those subs. So now they speak in AI voices using a broken translation from a faulty transcript. Great, thanks youtube, very helpful.
Years ago, there used to be community subtitles as a feature - people could submit translations and corrections and creators could then allow them on the video. Why they removed that I don’t know, but those would be really fucking handy now that you want to auto-dub the videos eh, youtube?
For those of us used to using multiple cues they can help, but they’re also often absolutely terrible
I’m pretty sure it’s just text recognition. I don’t think it’s AI I think it’s the same garbage text recognition system they’ve had in there home hub speaker things for ages. Still infinitely better than what they’ve got in Siri.
Youtube’s AI generated subtitles are the worst.
I see it has the same success rate and issues that the Closed Captioning system in broadcasting, which existed long before current AI solutions, has. The best way to get subtitles, is to have a hand-written .srt.
Most of those closed captions are, or at least were, done by humans. Even live TV.
I don’t know if it’s changed recently, but it used to be all humans with a stenotype. Professionals are very good at it.
It’s been moving toward automation since the 80’s. Almost all of it today has no human hands involved.
Treczoks@lemmy.world 4 days ago
Yep. They “see” music and applause at every corner.
cecilkorik@lemmy.ca 4 days ago
I also find it loves to caption background noise “Heat” for some reason.
Treczoks@lemmy.world 4 days ago
That seems to be a recent addition, yes, I’ve seen it in a few videos, but it is rather new.