17 votes

Forget subtitles: YouTube’s new feature dubs videos with AI-generated voices

17 comments

  1. [8]
    Sodliddesu
    Link
    I mean, subs are generally preferable to dubs and AI generated voices are almost all just shy of nails on a chalkboard and YouTube can barely get subtitles correct in the same language, let alone...

    I mean, subs are generally preferable to dubs and AI generated voices are almost all just shy of nails on a chalkboard and YouTube can barely get subtitles correct in the same language, let alone translating speech as well...

    I just don't see a way this is actually 'good' for anyone. Will my recommendations be flooded by low quality translations with this auto dubbing service that leads me to only look for channel recommendations on external sites?

    29 votes
    1. [3]
      stu2b50
      Link Parent
      Maybe for you, but there's many people and contexts for whom subtitles are not accessible. It's why there's dubbed live translation at any official international event. No, they said in the...

      Maybe for you, but there's many people and contexts for whom subtitles are not accessible. It's why there's dubbed live translation at any official international event.

      No, they said in the article it's a tool for content creators. They'll be able to create additional audio tracks dubbed in other languages.

      It won't be good for you, as you almost certainly speak English, which is the dominant language of Youtube. It will be good for people who do not speak English and will have vastly more content accessible to them.

      13 votes
      1. [2]
        Sodliddesu
        Link Parent
        But if a creator who only speaks English creates an English video and clicks the "Make it Dutch" button, how do they verify that it's the same content after translation? When I turn my video on...

        But if a creator who only speaks English creates an English video and clicks the "Make it Dutch" button, how do they verify that it's the same content after translation? When I turn my video on "How to perform a medical procedure" into Swahili, how will I know if they're properly telling them how to conduct the procedure?

        Yes, the article states the big names hire middlemen who can translate and localize their videos but who will verify the AIs work when everyone on YouTube has the "Farsi" button on their dashboard?

        6 votes
        1. stu2b50
          Link Parent
          In many cases, they won’t. And the translation won’t be perfect, but they don’t need to be perfect either. The translated CC are already very useful when I want to see content in other languages -...

          In many cases, they won’t. And the translation won’t be perfect, but they don’t need to be perfect either. The translated CC are already very useful when I want to see content in other languages - that is effectively the source of the dub, so I think it’ll be fine. With video you have the video as context.

          They also show that content creators have optionality on the contents of the dub. Although it still requires some language knowledge, it’s fundamentally much less work to merely read, verify, and edit the automated transcript that will be used for the dub.

          It’s up to the viewer using the tool correctly. I don’t see it as any more liable for misuse than the automated translated subs we have now. It is just merely more accessible.

          7 votes
    2. pete_the_paper_boat
      Link Parent
      I remember when people started advertising their own channel on PewDiePie's closed captions. That was a real shame cause it was rather useful, but too easily abused. However this is still good for...

      I remember when people started advertising their own channel on PewDiePie's closed captions. That was a real shame cause it was rather useful, but too easily abused.

      However this is still good for people with vision problems.

      1 vote
    3. [2]
      Jedi
      Link Parent
      Are you referring to TikTok’s “voice”? Then I agree. Google has some pretty advanced voice models though.

      AI generated voices are almost all just shy of nails on a chalkboard

      Are you referring to TikTok’s “voice”? Then I agree.

      Google has some pretty advanced voice models though.

      1 vote
      1. Moah
        Link Parent
        They're auto generated subtitles are garbage, I don't see why their auto generated voices would be better

        They're auto generated subtitles are garbage, I don't see why their auto generated voices would be better

        1 vote
    4. unkz
      Link Parent
      As someone else said, advanced models are astonishingly good compared to what you’re talking about. Thinking a step ahead though, wouldn’t it be interesting if voice cloning were able to be...

      As someone else said, advanced models are astonishingly good compared to what you’re talking about.

      Thinking a step ahead though, wouldn’t it be interesting if voice cloning were able to be applied to make the dubs in the voice of the actual people in the scenes?

  2. [3]
    CptBluebear
    Link
    What do you mean "forget subs"? Those are an accessibility requirement, not an optional tool to reach people that speak other languages.

    What do you mean "forget subs"? Those are an accessibility requirement, not an optional tool to reach people that speak other languages.

    7 votes
    1. [2]
      stu2b50
      Link Parent
      "Forget X, Y" is a phrase/template in English used as an intensifier for Y. It's not literal. That being said, the ADA has basically given up and agreed with Youtube's stance that having CCs on...

      "Forget X, Y" is a phrase/template in English used as an intensifier for Y. It's not literal.

      That being said, the ADA has basically given up and agreed with Youtube's stance that having CCs on youtube content is overly burdensome. Of course, if you have a separate source of ADA requirements, like being a government entity, you still need to provide subtitles. Manual subtitles by youtube creators are, in practice, an optional tool for multinational outreach.

      7 votes
      1. Eji1700
        Link Parent
        It always struck me as odd that this wasn't pushed more on the creators and for videos getting over X views. If you've got a literal production company because everything you put out hits millions...

        It always struck me as odd that this wasn't pushed more on the creators and for videos getting over X views.

        If you've got a literal production company because everything you put out hits millions of views, you can have a few people who's job it is to transcribe your videos at LEAST into your native language within a week or two of it coming out.

        I get there's edge cases here, but it's always struck me as youtube just bullying it's way past the ADA by claiming that since they can't do all content they shouldn't have to do any.

  3. petrichor
    Link
    Forget subtitles, indeed. Google's killing of community captions three years back ranks very high on my Google-crimes list. The system of (inaccurate) auto-generated audio recognition and...

    Forget subtitles, indeed. Google's killing of community captions three years back ranks very high on my Google-crimes list. The system of (inaccurate) auto-generated audio recognition and streamline-contracting-to-third-parties that replaced it is utter garbage.

    For those unaware: community captions allowed the YouTube community to write and sync subtitles themselves and submit them to videos - where after being approved by the video uploader, they would be publicly available. It was a brilliant system and worked incredibly well - particularly for YouTubers and those with audiences, as fans would naturally put time and care into subtitles for people they were fans of: separating different people talking, transcribing visual gags, adding context and occasionally commentary (commonly separated from descriptive subtitles as English (Canada), in my recollection).

    Comparing any video with subtitles on from a major channel five years back and now showcases the astounding drop in quality. I do not have any hopes for this to be useful, as per any change to YouTube.

    6 votes
  4. sloslosloth
    Link
    No, I will not forget subs. I can't even tolerate dubs when it's done by people who are trying hard to replicate the emotions. So I can only imagine what AI translation with AI voice will sound...

    No, I will not forget subs. I can't even tolerate dubs when it's done by people who are trying hard to replicate the emotions. So I can only imagine what AI translation with AI voice will sound like.

    I can't wait for this AI frenzy to die down so that we can go back to just bitter arguments with other humans over nothing.

    2 votes
  5. [2]
    drannex
    Link
    This could be interesting for language learners, and will be entirely cool once they inevitably integrate vocal sampling from the video and have the generated dub be in the same vocal style (and...

    This could be interesting for language learners, and will be entirely cool once they inevitably integrate vocal sampling from the video and have the generated dub be in the same vocal style (and cadence? potentially) as the original video / creator. Uncanny, but certainly doable and exciting.

    1 vote
    1. sandaltree
      Link Parent
      I don't really think so. There's no reason to use this instead native real content. Just like ChatGPT doesn't really offer much over real content, that there is an abundance of already.

      I don't really think so. There's no reason to use this instead native real content. Just like ChatGPT doesn't really offer much over real content, that there is an abundance of already.

      1 vote
  6. [2]
    Moah
    Link
    Yeah those ai generated voices still be very useful to us hard of hearing people.

    Yeah those ai generated voices still be very useful to us hard of hearing people.

    1 vote