How does YouTube's auto closed caption feature work so well?
How does YouTube's auto closed caption feature work so well?
Kevin Jackson
Ayden Morales
Because captions have to be added by the author?
Austin Roberts
The automatically generated ones are near perfect too
Austin Nguyen
It uses the same voice recognition of google translate, which is pretty good too
It's that good because Google stores the voice files of every android user using "ok google" by default.
They then use the billions of voice recordings and process them with machine learning.
Landon Gray
Where the FUCK were you when Google Assistant made an entire phone call using robotics
Ryder Thompson
Because they know.
Liam Reed
>near perfect
I lazily googled for an image to illustrate, but it's hardly perfect. American english is fine, any other dialect can be really sketchy. Other languages, especially asian or indio-pakistani ones are just awful. Sure it's a technical marvel that they're where they are now, but it'll be a while.
Jeremiah Harris
youtube.com
My computer is way too slow for me to provide sources on this n shit, but I distinctly remember this 12-second video with currently 2.4m views having a funny maymay auto closed caption transcription. Youtube comments are always pointing this shit out.
Easton Davis
>ancient interface from about 2014
it improved greatly since then, for English videos anyway
foreign language recognition still sucks and the results are either mismatched language (like Polish on a Portuguese video) or everything is just misinterpreted
Jack Watson
Hence the reason I wrote
>I lazily googled for an image to illustrate
Isaac Brooks
Not him but his filename has a timestamp of mid 2015. You couldn't just cite that instead of strawmanning it's a year older?
Luis Clark
>didn't see filename
>said "about 2014" for a reason
>2015 still leaves a 3 year gap
>hurr durr le strawman
Luke Green
>saving screenshots as jpg
Get it together
Jonathan Torres
What is, Jow Forums X's post from URL feature
Carter Morris
cute css!
Lincoln Watson
Just a slightly tweaked Blackboard, it's in the Jow Forums X list
Sebastian Bennett
Like 80% of the image is a camera video, so jpg makes more sense.
Save it as png and the filesize will be 6 times as high.
James Rogers
It uses machine learning to generate them, you can edit the auto generted ones to make them accurate which in turn fine tunes the model making it more accurate. Do that process over a few years and you get this result
Brayden Moore
What is this illiterate excuse of a reply. The only valid information contained here is you concede that you strawmanned when you think it'd still be fair criticism if you didn't, so it was unnecessary. But you say this while being angry you got called out on your strawman. Why can't you just say "my bad, you're right" without pretending you have any high horse to speak from.
Julian Powell
I legit don't know and I also couldn't care less.
Asher Fisher
Lurk more
Austin Mitchell
Can't find it, link please? ;_;
Daniel Taylor
A jstoddler attempting to tell me to lurk moar?
Kayden Moore
Bentley Sullivan
Oh sorry, I meant Oneechan.
So github.com
and
github.com
Oliver Sanders
Oneechan moved off github when Microsoft bought the site
Adam Thomas
Sad
Could try adding this manually, I've just pasted it from the js. pastebin.com
Camden Collins
Zachary Ward
Can't actually click the .js though
Juan Cox
fuck it, joined your botnet, thanks!
Thomas Murphy
Lel. As lulzy as it would have been to install a miner script, it tends to be very obvious as your fans go apeshit.
Ethan Fisher
I don't care desu, just want the cool look, owo
Cooper Gomez
That seems like a very nitpicked example.
In my experience auto generated captions are near perfect. Of course problems start when you watch people with heavy accents, adoredtv is a good example of that. But even his videos are very much watchable with auto-CC turned on.
Hudson Scott
install gentoo
Carter Smith
This. Jesus, dude. It made, like, three mistakes in a minute of commentary.
Dylan Rodriguez
Like the CGP Grey video about bots?
Jaxon Gomez
Most likely it uses some sort of learning algorithm, there're plenty of petabytes of samples available at google. I use it for timing my DIY anime subs, it still fails completely on audio parts with songs (like op/ed), I trim those before uploading though and it really is good for timing.