To all those working on Naruto fan dubs on the early 2000s I say:
I am sorry. I was too hard on you. It took me 20 years to realize what 'hot garbage " really is in the context of an anime dub.
To all those working on Naruto fan dubs on the early 2000s I say:
I am sorry. I was too hard on you. It took me 20 years to realize what 'hot garbage " really is in the context of an anime dub.
So do they not even have internal review?
OK, so it gets generated. Just like if someone sent a real recording, someone should listen to it, right? And make mention if there's parts that are noisy, or theres a random change in eq that feels jarring, or... I don't know, because it sounds like an adult voice on a 6yr old reading the lines?
Wow. I watched some in the embeded tweet and that was uhhh... That was hot garbage...
Dont watch the tweet if you don't want spoilers, there are important story beats included cause they're scenes with a lot of emotion that the ai dub completely mangles
Highly recommend the show. Content warning for sexual abuse and exploitation though, its a central part of the story and the show is a very difficult watch
Posted this in another community but I’ll leave this here too:
So back when I worked at Amazon I was playing around with AWS Skillbuilder. They don’t pay for any other training materials for SDEs (well they used to have ACloudGuru but ended that).
I was like “they charge money for this so it can’t be that bad right?” Well
1/3 of the courses were actually what I’d call “watchable”
1/3 were just SEO Blogspam masquerading as information
And the remaining 1/3 clearly used a text to speech software that was dreadful. It was incomprehensible.
I say all this to say that if there was ever a list of companies I would trust to do AI Dubs Amazon would be the bottom of that list.
It's insulting because there are a lot of LGBTQ+ voice actors who want to do this but the powers that be won't greenlight it.
Look at Amazon's relationship with Trump, Trump's positions on LGBTQ+ people, and ask why Amazon is doing this. They aren't contractually obligated to do so. They are doing various animation projects and paying real voice actors. Though Hazbin Hotel is pretty gay, and so is Helluva Boss (same people/same universe). The only other animation project I know at Amazon is Mighty Nein, and I wanna say those guys are LGBTQ+ friendly but I'm not sure they're on the spectrum. So I don't wanna say Amazon is acting prejudicial here, but it smells.
I mean, yeah, Amazon's relationship with Trump is far too cozy... however, in this case, I think it's more likely that Amazon was just trying to cheap out by using AI (and possibly test/showcase their use of the tech) rather than them actively deciding to cut out LGBTQ+ voice actors.
FTA:
As giant tech corporations try to jam AI into every possible orifice in the world, we are consistently getting new examples day in and day out of that going poorly. This week, that’s Amazon Prime Video introducing dubs to the beloved anime Banana Fish, which has needed them for a long while. The problem? They’re AI. Not just AI, horrible AI.
Had this decision been based in bigotry, I would imagine that they wouldn't have bothered with the dub at all.
Company like Amazon is huge. Like, the department to cozy up with administration is a completely different subsidiary than Prime Video anime importing. They don’t know to each other. They will cut each other’s throats if they are fighting for the same promotion.
This happens because 1) cost savings and 2) human garbage exist everywhere. They hide behind the decisions and use the current political climate to push their bigotry.
isn't interpreting language the ONE thing LLMs are supposed to be good at? jesus
They’re pretty bad outside of English-Chinese actually.
Voice-to-voice is all relatively new, and it sucks if it’s not all integrated (eg feeding a voice model plain text so it loses the original tone, emotion, cadence and such).
And… honestly, the only models I can think of that'd be good at this are Chinese. Or Japanese finetunes of Chinese models. Amazon certainly has some stupid policy where they aren’t allowed to use them (even with zero security risk since they’re open weights).
On one hand: It's impossible to dub every single anime at a reasonable time.
On the other: I totally see the big companies refusing to hire real actors for dubbing because 'muh profits'
Point 2 is true, Point 1 is simply not unless by reasonable time you mean immediately into every language.
There are ~25K anime in MyAnimeList. If we assume that there are 250K episodes with a total average length of 25 minutes, each, we are talking about dubbing 6.25 million minutes overall.
Let's say 80% of them is not yet dubbed, that'll be 5M minutes. Netflix delayed KaoruHana for at least two months to dub 8 episodes * 25 minutes, so they likely took three months to complete 200 minutes of footage i.e. a rate of 0.3 anime minutes/VA minute. Let's say there are 100 organizations with three teams of dubbers each, so the rate can be expanded to 90 anime minutes/VA minute overall.
Clearing the backlog will require 55K minutes. That's 925 hours/115 business days, or almost a year. I guess it is possible if we wait this long. And this calculation ignores time to select the right VA for the job, paying the amount the VAs request to accept the job, plus a whole lot of other factors I failed to mention or I don't even know exist.
And this is for just one language (English), and I'm still waiting for the Italian dub of JoJo Part 5.
If we assume
bruv every anime listed in MAL has runtimes, lengths, and additional metadata. Your overestimation is simply wrong, with the fact there are few N1 localizers willing to work at the 硬貨 on the دينار for whatever AMZN is valued at. If you don't pay what we are worth, continue waiting on Italians to localize Jojo for you at N5 levels of knowledge.
How is “every anime has one season of 10 episodes running 25 minutes each” an overestimation? I’d call that an underestimation. Yeah, there are anime with much shorter runtimes, but there are also many anime with well over 100 episodes.
Not to mention, are you really expecting this random person to write an interface for their API to get exact numbers for an off the cuff online discussion?
Because dub time doesn't equate to run time. Most anime has PLENTY of no voice work, and openings and endings can be copy and pasted.
are you really expecting this random person to write an interface for their API to get exact numbers for an off the cuff online discussion?
I fear more you are not aware where in the internet you are. This is basic 101 scripting work that happens every second your application notifies you your new anime got updated. Basic scripting is what our /c/ and threadiverse do for most of us who actually moved on from fashit.
This is basic 101 scripting work that happens
Cool, so you can share the total runtime?
At 10k€, sure. Monero or?
But it’s simple basic 101 scripting, by your own words. If you expect the other user to do the work for a one off calculation instead of assumptions, I’m sure it should be trivial for you to throw something together?
It is a simple for loop, or a WHILE if I ask Jicetus nicely. I just don't do threadiverse homework for free anymore.
So, are trivially going to pay up, or do you need me to send a simple PM?
My dude, you’re missing the point. You expect them to give accurate numbers, but aren’t willing to do the same. Double standards are pretty lame.
Nah: I said the overestimation was completely wrong. I also said runtimes ≠ voice times.
But you're missing mines: How many N1 localizers are you gonna hire inorder to localize all the anime in MAL in order for you volunteers to voice act your dubs.
Honestly N1 is probably unnecessary. I think most people who've passed N2 could do a great job with most shounen stuff. It's not like they need to do it in real time.
Depends the anime. Idol anime like the Pretty series or Pretty Cure require a lot of creativity to localize from the era they are portraying. But general shows like in shounen, sure.
not like they need to do it in real time.
🤝 Exactly. And the fact we are scattered everywhere, doesn't mean we have a window to localize from 0900-1700. Tis one of the reasons I want to emigrate to Europe ASAP, so I don't have to karoshi.
I prefer proper VA dubbing versus AI dubbing. This is more of a question if the viewership prefers AI dub vs. no dub if proper dubs are not yet available.
bruv every anime listed in MAL has runtimes, lengths, and additional metadata.
Fair, I don't have the time right now to be 100% accurate at my figures, so I went with a rough estimate. I tried to be as clear as I can on that point.
there are few N1 localizers willing to work at the 硬貨 on the دينار for whatever AMZN is valued at
If it's only easy for fandubs to be readily available. Like there are few N1 localizers willing to work at Amazon's assessed rates, few IP holders are willing to say yes to dubbing their shows at the money fandubbers can afford. There's also hiring the proper voice actors for the characters, and the ones who do the anime justice/will not have the fans crucifying the anime for doing a craptastic job at dubbing deservedly ask a premium.
viewership prefers AI dub vs. no dub
According to the beta testers, and the Internet, listeners abhorred the LLM localization & actual tone-deaf Speech audio dubbing. Keeping the original dubbings is simply what folks want, esp. if it's labeled abridged.
[components of dubbing]
At the least you are aware why this /c/ prefers subs, because it is that much cheaper and errorless to output.
According to the beta testers, and the Internet, listeners abhorred the LLM localization & actual tone-deaf Speech audio dubbing. Keeping the original dubbings is simply what folks want, esp. if it’s labeled abridged.
Yes, at its current state. Will it stay that way? The tech companies are burning cash in attempts to make it not so. My hunch says even Vocaloid-tier AI dubbing will be enough for a large sector of the audience. Then the human vs. AI dubbing debate could be analogous to debates between lossy (more accessible) vs. lossless (higher quality) audio.
Now, LLM localization is the greater challenge. I highly doubt those, including the classic machine-learning models, can reach N1-level localization quality.
The only thing funny about mentioning Vocaloid is the fact that Vocaloid synthesis has to be manually pitched, tempod, and toned🤣. Glad you honestly believe capitalists want to invest more on disqualifying tone deafening pitchless speech waveforms.
But please, never stop supporting espeak!
It's amusing to me how long people have been saying "yes, AI is crap, but it might not be crap some day, so just you wait!" Despite all the money tech companies have thrown at AI, it's still as crap as it ever was, and I don't see any reason to think it'll get better.
Meanwhile, Crunchyroll doesn't care if it's crap, so long as they can get around the cost of paying humans (which is another can of worms). If they're willing to buy this level of quality, what incentive is there for quality to improve?
yes, AI is crap, but it might not be crap some day, so just you wait!”
I mean, there's a gap between the capabilities of Cleverbot and ChatGPT, as referenced in this very comments section. As much as one wishes it not be so, it would be foolish to ignore past technological leaps—and how people back then laugh them off as impossible.
I don't see any significant differences between ChatGPT and Cleverbot, if I'm honest. It might have a wider array of responses to pick between, but it's still making the same mistakes.
It would be foolish to ignore past tech bubbles, and how people back then claimed they'd fix all their problems in the near future and you need to jump on now or you won't survive (and how none of them survived).
Unlike Cleverbot, you can add your project-specific context in ChatGPT. That was extremely helpful in my creative writing process as I use it as a virtual assistant.
It would be foolish to ignore past tech bubbles, and how people back then claimed they’d fix all their problems in the near future and you need to jump on now or you won’t survive (and how none of them survived).
While largely true, that none of them survived is false. Amazon is a survivor of the dotcom bubble. Pets.com died, but Chewy perfected the concept later on. Circling back to the topic, if/when the bubble bursts, we could be talking about 90% of the AI-centric companies going under, give a decade or so, a "stabilized" form of AI dubbing could resurface and establish a long-lasting presence.
Now, LLM localization is the greater challenge. I highly doubt those, including the classic machine-learning models, can reach N1-level localization quality.
There's no chance it's happening any time soon. Many manga and anime lean heavily on visual context as well as the context of the story in general to clear up situations where the language would otherwise be ambiguous, so until the translation software can also use all of that context it's basically impossible.
I mean if you think about it, Japan solved that problem given that every series is already dubbed in one language.
It's impossible to dub every single anime at a reasonable time.
I mean, not really. For all already existing ones, maybe, but for new ones? No reason for it to be impossible. The Japanese dub doesn't appear out of thin air. The Western streaming services companies just aren't willing to pay up.
The argument of "just watch subs" falls flat when they're using AI to do soulless subtitling too.
Jesus fuck I hate AI so much. The other day Google AI literally gaslit me. I remembered a joke I had heard in a movie but couldn't remember the movie title. Google used to be really good at this kinda thing so I Googled it. The results were wrong and the AI was like "It's Mike and Dave Need Wedding Dates".
Now, I've never seen Mike and Dave Need Wedding Dates so I knew that couldn't be the movie so I told it "no that's not it". And what did the AI do? Did it say "oh, well then maybe it's this?" Did it say it didn't know? No, it doubled down insisting it MUST be Mike and Dave Need Wedding Dates.
I COMMANDED it to stop suggesting it was Mike and Dave because it was definitely not it and it STILL kept saying that was the movie.
So I asked ChatGPT. Yeah, I know it's another AI but I was at my wits end here and Google was undeniably wrong. Anyway, ChatGPT figured it out instantly.
So I went back to Google and was like, "ChatGPT figured it out. It was 'The Art of Racing in the Rain' you stupid AI."
Rather than admit it was wrong, Google then gaslit me saying I was misremembering the movie and that it was Mike and Dave Nerd Wedding Dates.
I was furious so I told Google to prove it to me. Then it got all vague like, "well, I searched my database and web records." I asked for the specific records. It gives me a link to the movie script for Mike and Dave Need Wedding Dates on some website. I do a simple word search and do not find the joke anywhere in the script. Nothing even remotely close.
When I point this out to Google it was like" well, maybe it was adlibbed or not in that early version if the script, but it's in discussions of the movie so I know I'm right."
"Show me these discussions"
"Sorry, there are no discussions"
"So you hallucinated the whole thing and completely made it up?"
"Yes"
Fuck you, Google.
So I went back to Google and was like, “ChatGPT figured it out. It was ‘The Art of Racing in the Rain’ you stupid AI.”
lmao
You know 15 years ago it was common pastime to see how quickly we could trick Cleverbot into admitting it was a chat bot and not actually intelligent. Now we have chat bots competing with us to prove that we must not be intelligent.
And they're winning.
While not quite as long a thing I have had the ai do this. Maybe you are misremembering and its just that it could not find it.
At any point in time did you ever consider doing a non-AI search? The sheer fact you say you hate AI… yet you continue to use it is a problem in and of itself.
The results were wrong
Yes... maybe I should have emphasized that the SEARCH results were wrong. And have you used Google lately? It forces the AI on you. You can't turn that shit off. At least not easily.
But yeah, apparently if I had asked Bing it seems that it would have gotten it. Who would've thought that Bing would now be better than Google? But even Bing forces Copilot on you.
So please, oh master of internet searching, what search engine doesn't force AI on you? I guess I could pretend it's 1999 and use Lycos...
Dude, ever hear of Duck Duck Go? It’s a more privacy focused search engine that you can turn off any AI summaries with one click. It uses Bings DB but its own interface.
Oh snaps, I forgot that existed, mostly because it used to suck. But if Bing's better than Google now...
I'm glad I asked and sorry for being such an ass. Well, mostly sorry ;)
If Bing's database isn't good enough for you, Startpage supposedly uses Google's database and doesn't shoehorn AI into the results. You'll want an adblocker, though. (My usual order of query is Duckduckgo, then Startpage, then Mojeek if the other two didn't find anything useful.)
So if I understand correctly, they fire professional voice actors and replace them with grotesque AI voice bots. I hope there will be more fandubbing in the future
Honestly, the AI-driven subtitles are intolerable enough.
The sooner this AI bubble crashes and burns the better. If this trillion-dollar corporation can't keep up with the expected reputation... then it needs to stop claiming that's it's worth that much.
Voice acting is so incredibly disrespected, its not a surprise at all they'd try to replace it with this garbage.
Oh my god I fucking love these dubs.
It's so fucking bad.
I need more garbage like this in my life
Well good news, it'll probably run locally by next year.
I'd love to be able to watch any show from any language on like a one-minute delay, with the robot faking the original voices. But I'd also like a fully in-character version of Revenge Of The Sith with the infamously bad Chinese subtitles.
I recently watched Last Samurai Standing from Netflix I think, which I actually thought was pretty good, but I noticed the English dub seemed to use some kind of AI to make the character mouth movements match the English dub words.
At first it wasnt too noticeable, and I thought they had maybe filmed in English. But there was a lot of things that seemed like maybe they had some native Japanese people on the team because English speakers would have either not known about it or would feel culturally compelled to change it. Then at some point I noticed it, the mouths moving unnaturally. I think it was a close up of someone talking, but once I noticed it, it was pretty distracting.
I honestly like the idea though. It might actually be a legitimate use case for "AI," as some people are distracted when character mouth movements don't match the words they say, but in this case I think it is still too early for deployment. It was done well enough to pass on people that can't see very well, but it was pretty distracting for me once I noticed it. It would be too much work to try and film all the actors saying languages they likely don't understand to try and composite. Is it mandatory? No, but it would be nice for people that get distracted by mismatched audio.
I can't seem to watch the example ATM, but I seem to be allergic to subs, so there are MANY shows that I would love to have even trash-quality dubs.
You may be underestimating how bad these AI dubs are.
Honestly.... given the reaction seen here, I was expecting FAR worse. Assuming there is approximately zero human input/effort, I actually find it remarkable for a push-button auto-dubbing solution. In fact, if I had such a auto-dubbing "button", I would most surely make heavy use of it, and maybe even pay extra for it as a service... but that might just be me.
Here is another one, do you really think this is good:
https://xcancel.com/OtakuSpirited/status/1994662919498600488
This also has a dub originally but Amazon decided to go with AI. Cheaper.
I prefer this over AI: https://youtube.com/watch?v=m5LjA0gjeX8
I understand that it's weird, and it sucks that they would lift this above a curated human dubs (I guess they have separate ongoing license fees?), but yeah... as best I can tell it would be watchable (which for me is better than subs-only). Though, I don't know what it might do to my enjoyment (or brain) after a whole episode/season of ingesting this. :)
I think where I land on this issue is that this ought to be a third pole. We have subs, dubs, and this... and IMO it is misleading to call it a dub, or mix these in with wittingly-dubbed content, but I certainly see value in it.
Secondary effects may be another matter... part of me is afraid of a "raise to the bottom" effect, where we find ourselves in a position where nobody pays for high-quality dubs at some point. But in theory, lowering the barrier to dubbing does not necessarily need to bring everything down to trash level; if practically overnight everyone and their dog can auto-dub anime, there is no reason to have sub-only anime, pro-dub becomes a standout feature, and watchful streaming services could even see those autodubs that get high watch counts as a demand signal to pay for professional dubs.
all 18 comments