Seems like they have enough resources to train their translation NN and speech-to-text NN. I don't think google has introduced those features for the users in the first place but they were a means to get training data.
Their automated CC sucks, always has, and I check multiple times a year only to see that it still sucks, most recently last month. I’ve never had the best hearing, so I usually turn CC on when I can, particularly with ESL speakers with heavy accents; but if the only option on a YouTube video is the automated captions, then the only option is crap and I’ll go without. On the other hand channels with a large dedicated fan base will often have fans contributing captions out of their own volition.
If they really think their NN is good enough to replace community contributed captions, I have yet to see evidence of this, and there’s no evidence in the announcement that this is a vote of confidence in their tech. If anything, it’s a vote against their ability to manage spam and abuse, they just don’t want to deal with it anymore.
Exactly same experience. Hearing disabled and using subtitles whenever I can. The generated ones are not good at all. The more unusual wording (i.e. full of technical terms or custom names) the worse it is.
A lifetime of thinking that companies can reimplement cool tech they have in use somewhere else in a place that it would be even more useful has left me disappointed in the possibilities. I’m not familiar with Pixels because I don’t use them, but even if I took you at your word that it is automated captioning that works, I wouldn’t hold my breath that it will show up on YouTube just because Google has the tech on a Pixel.
It’s not like companies like Google have never had technical regressions either. Their core service, Google Search, is less useful to me than it was 10 years ago.
Search became less useful because it's optimizing for ad revenue, not user experience. I don't see that conflict of interest here. If anything, better subtitles would lead to more watch minutes (from non-natives) and hence more ad revenue.
but the tl;dr amounts to just because a company has the technology does not mean it is broadly applied across all of its services where it would be most useful. That said, if YouTube ever implements automated CC that works I’ll happily scarf down that humble pie.
Earlier today I watched a video of two people talking. Youtube decided it was in french and the transcript had less than one word per minute, all of them wrong.
If they really think their NN is good enough to replace community contributed captions, I have yet to see evidence of this, and there’s no evidence in the announcement that this is a vote of confidence in their tech. If anything, it’s a vote against their ability to manage spam and abuse, they just don’t want to deal with it anymore.