No per words pronunciation on "video file" part or LR

No sound for the word pronunciation in the video file player since the recent upgrade,
it works on Netflix and YouTube though.

My bad, I kept LR main target language selector in Russian while using a Hindi-speaking video… I thought it would auto select like with Netflix and Youtube though…
Bah, at least I know now…

1 Like

I made a couple of fixes on video file function, but, it’s pretty neglected at the moment, sorry, glad you found a way to get it working.

2 Likes

Yeah, I understand you guys have your work cut out for ya, it’s not the only module you all have to work at. Could you guys please implement being able to upload TWO subtitle files in the video module? So you could have target language and proper human translation if you have both subs on your gear… because machine translation is useful, but boy is it flawed x’3 BTW, is it possible to get ASR on Youtube and the video module? :slight_smile: ASR isn’t perfect either, but boy that’s so useful.

2 Likes

Yes! I second this request!:slight_smile:

Two tracks, yeah that’s faily easy. What languages are you using for machine translation btw? Maybe I can do something there.

ASR on video file mode - yes, it’s possible, a few things to set up but doable.

On youtube, the problem is the huge volume of different videos that would need to be transcribed… for pro users maybe, once I’ve got the second machine set up right (gpus overheat), and there’s new code that does batching, to increase the througput… but, Youtube has ASR already for ~10 languages. It’s not broken into nice subtitles, that can be improved… although there are word timing issues which are harder to fix. Anyway, how would you want to use language reactor ASR on Youtube?

Thanks for feedback

3 Likes

One of the big things I’ve found is that many Korean videos have burned-in subtitles, and YouTube’s auto-generated subtitles are usually pretty bad —in my experience— for most of those.

Some have also been uploaded without marking them as Korean (or whatever the proper native language is), so CCs don’t always show up there.

I would use this feature for Korean vlogs with more talking (and potentially less background noise) than, say, a random K-drama clip from a show online (as an example).

And I think the machine translation that Mr_FrO.6 is referring to is the machine translation of the video file feature uses to translate the one subtitle file (the target language subtitle).

ASR on video file mode might be a middle ground if someone downloads the YouTube video from YouTube — that would have more steps to immersing into the content, but it makes sense if doing ASR on YouTube produces a lot of strain on the gpus.

I definitely think making it a pro feature makes more sense, given the potential resource drain.

Just some thoughts, anyway!

I use English and sometimes French for machine translation (both languages I’m fluent in), I’m basically learning Russian and Hindi. Youtube already has ASR for 10 languages? Well, maybe but if the creator of the video didn’t set it there is no subtitles at all. By the way, how does that ASR function work? Do it redo all the subtitle file each time or does it do it once and save the file (to save on GPU crunching power… it would only be needed to redo it when asr has been updated to get a better file) ? Also, can those subtitles file created be downloaded to be humanly improved? Could the asr use context from the original subtitle file and/or translated file to avoid translation mistakes? I once tried to put both Russian and Russian ASR to have more chance to have accurate text but then the dictionary and all is deactivated and of course you can’t select or nothing with the second subtitle line (it would be nice to be able to have both the original language sub + it’s asr on screen as main sub if that’s possible somewhat , one day… I even think making possible to have two different subs for translation could help for learning. I mean you understand that easily when you’re already bilingual. For instance since I’m learning Russian some words will resemble English a lot and therefore been easier for an English speaker to learn, but some other words will resemble French and will be easier for a French speaker to learn… so if you have both on screen it pops right out and is easier to notice and retain I think… the problem is that then you have to do the same for the dictionary , having it display in both language… (same in text module now that I think about it… it would rock though) but shit that would rock, having to toggle between both to learn better is a drag. I dunno, tell me watchu think about it.

1 Like

I’ve been thinking about this video with burned in subtitles… it’s a real shame because those subs can be done really fine… I generally use an ocr on my PC or Google lens on my phone… so I can slap the text in LR Text module and learn from it. I even went so far as to looked into how some kind of “Lens” could be done on video, and it’s a bit tricky… Since I’m learning Russian I wanted to use those videos of Atomic Heart full game to learn from it but the youtube auto subs are crap… the people playing activated the subs from the game so subs are there but basically burned in (and awfully tiny…).

1 Like