Forgot your password?
Close
typodupeerror
Youtube

Ask Slashdot: Are YouTube's Subtitles 'Appallingly Bad'? 100

Posted by EditorDavid from the spelling-flame dept.
Long-time Slashdot reader Anne Thwacks frequently uses YouTube's subtitles "not to disturb others in the room, or because my hearing is not very good." But they say there's a new problem.

"The subtitling is terrible!" Almost every sentence has a huge error. Proper names are more often wrong than right. Non-English place names are almost always mangled to barely recognizable. And no effort whatsoever is made to use context to figure out whether a place name is Russian or Arabic, and often complete garbage is used in place of a common French, Spanish or Italian name!

If AI actually works (I have my doubts about this), surely it would be possible to figure out language contexts. If it is about an event in Italy, then expect a lot of Italian names! If it is about the Russia-Ukraine war, then expect places in Russia or Ukraine to be more plausible than mindless gobbledygook! Does YouTube not know that there are places in the world that are not in America? (However, plenty of names of people and places famous in America are also regularly screwed up.)
They argue the subtitles are "appallingly bad" — and that "the situation seems to be getting worse," wondering why the problem isn't addressed with some basic spell-checking. ("I'm sure that the vast majority of foul-ups could be fixed by the use of a dictionary.") Have any Slashdot readers seen similar problems? A friend of mine noticed that YouTube's subtitles even bungled this innocuous song from the 1966.

ANNETTE FUNICELLO: "If your love is true love, you can tell by his touch."
YOUTUBE SUBTITLE: "If your love is too lava, you can tell by his touch..."

Share your own experiences and thoughts in the comments. And do you think YouTube's subtitles are "appallingly bad"?

Ask Slashdot: Are YouTube's Subtitles 'Appallingly Bad'? More | Reply

Ask Slashdot: Are YouTube's Subtitles 'Appallingly Bad'?

Comments Filter:

  • Nah (Score:3, Insightful)

    by devslash0 ( 4203435 ) on Saturday May 02, 2026 @06:43PM (#66124650)

    It's just that the entire YouTube is appallingly bad.

    • Youtube has plenty of good content. The caveat being what is pushed on their main page is utter garbage so you have to actively search for it. Take Adam Savage for example. He toured a facility that scans imax film and showed the original one off film scanning machine built 25 years ago. It's an engineering work of art. https://www.youtube.com/watch?... [youtube.com]

      • Youtube has plenty of good content. The caveat being what is pushed on their main page is utter garbage so you have to actively search for it. Take Adam Savage for example. He toured a facility that scans imax film and showed the original one off film scanning machine built 25 years ago. It's an engineering work of art. https://www.youtube.com/watch?... [youtube.com]

        I like Savage's content as well. His workshop vids are like siding down with an old friend, and I find most of them relaxing.

    • It's just that the entire YouTube is appallingly bad.

      A lot of the audio production in individual videos is really bad. This isn't anything to do with YouTube per se, not their compression algorithms or other features. A lot of YouTubers have absolutely no concept of microphone placement, of using audio compression, of reducing background noise. All of which are things which will drastically affect audio quality and the ability of a speech-to-text model to create subtitles.

      It would be nice if YouTube would normalize all the uploaded videos to one set standard.

      • Re: (Score:2)

        by pjt33 ( 739471 )

        YouTube does already do something to the audio. I notice it when streaming guitar and vocals: the sound which comes out of YouTube emphasises the guitar, so if I balance listening to the mixer feed I have to balance the guitar low.

    • The number of medical misinformation ads is crazy.

      • Like Dr Gundry's " water makes you fat" claims. What dumbass believes that shit?

        The one that needs outlawed is the AI lady hocking free grocery cards to Medicare recipients.

      • Re: (Score:2)

        by shanen ( 462549 )

        More the branch I was looking for. Business model makes EVIL.

        But the story is actually related to a topic I wanted to AskSlahdot about... It would be interesting if there was a video website where you paid by helping to train the AIs. As it applies here, you would earn credit for correcting errors in the subtitles. Easily leveraged against the users by checking each participant's corrections and suggestions against other participants' work.

        But for the greater glory and never-sufficient profits of the ever-m

    • Google's subtitles are terrible, and what's worse is that a lot of AI crap trains off it.

      Like all my videos have perfect subtitles because it comes directly off the input. When I've let youtube "auto" caption, it can't tell what is speaking most of the time. So you get really worthless captions on anything that is not a one-person podcast with no music and an American accent. If they have an Aussie accent, not a chance. If they have a Quebecois accent, ZERO.

      • And because the meta for YouTube is to pump out as many videos as possible, so that a channel owner has more 'content' for YouTube to be jamming ads into, from which the channel owner gets a sliver of payback from, there is a constant drive to push new videos out as fast as possible, so anything that slows down publishing new videos gets kicked to the wayside. The channels that use AI narration just push the script through an AI-driven text-to-speech system, and don't care whether it's using the right pronu

    • It's just that the entire YouTube is appallingly bad.

      Youtube has a lot of good material. Perhaps your interests are appallingly bad? I use Youtube to watch woodworking videos, to watch repair videos - like if I'm trying to figure out a problem on my son's car. Radio technology videos - computer teardown videos. I replaced a battery and hard drive on a laptop recently, so I previewed what it was going to be like before I tore into it. Even some programming videos. I recently became interested in Node-Red, so looked up some of the basic operations to get starte

  • But 20% of all the internet's traffic is still more than enough to make more money than God.

    Different set of incentives at play here.

  • For many years, my friends and I have noticed that Google Translate (and YouTube) have a better chance of getting a European language pronoun wrong than correct.

    We've noticed this: French, Italian, German, Spanish ... and most of the Scandinavian languages all suffer from this.

    It must somehow be related to how those LLMs are trained.

  • but they are better than nothing. I sometimes watch foreign content and i appreciate the attempt at a translation even if it is bad. Helps me to understand people like Angine de Poitrine.

    • Re: (Score:1)

      by Anonymous Coward

      uploaders can put in their own captions, if the video is considered incomplete that's where you point the finger

      as a stand-in auto CC is a courtesy feature and has been around longer than the septembers complaining

      i imagine this particular tourist will be complaining about AI not "realizing" context for a long time, how is deleting things MORE productive???!?!?

    • As a musician I can promise you trying to figure out the "foreign" context wont help you understand Angine de Poitrine. The whacky costumes and the like are part of a very english speaking tradition, punk absurdism. Likewise musically, they are coming from a group of musical genres thats about as english speaking as you get , prog rock and jam bands. That they are french speaking canadians is almost irrelevant, other than the fact that they've been known amonst qubecans longer than the rest of the anglosphe

      • What makes them different for me is the microtonal guitar. See also King Gizzard and Lizard Wizard if you are into that sort of thing. Intonation and accurate tonality of western style guitars has always been a sore spot. It is good to see musicians addressing some of those problems with custom fretting. I understand that before emigrating to Quebec, Angine de Poitrine inhabited the planet Claire. It just blows me away that Google is able to translate from their native Clarion to mostly understandable Engli

    • Re: (Score:2)

      by tlhIngan ( 30335 )

      Yes, they're terrible, but until someone forces all the creators to upload their scripts or create the captions themselves, that's the best we got - speech to text is never 100% accurate, and fails the more strongly an accent is.

      That said, some YouTubers do actually properly caption their video, and you can tell because it no longer says "auto generated".

      And the only reason YouTube has captions is because of ADA compliance demanding it.

      It's just a natural consequence of a site where production can range fro

      • Re: (Score:2)

        by tepples ( 727027 )

        Then perhaps what Google has failed to add on YouTube is a checkbox to exclude videos with only auto-generated captions from search results and recommendations.

  • It's quite hilarious for me watching foreign videos where I can't tell what they might have actually said because I lack any context, all I can do is boggle.

    I have seen improvement, but it's mostly in the software's ability to pick speech out of noise. The comprehension part is still nonexistent.

    • If you've never seen it; Watch some clips of Funky Forest in the original Japanese.
      I watched them and wouldn't find translations for years, I had a strong suspicion that there was no story at all and it was just nonsense video art but it turned out to mostly not be the case.

      I guess I should also mention this is not anime

  • Daisychained clickbait asking users to "Like" and subscribe, I watch some YouTube garbage when I am bored with nothing to do but I never interact with Liking or subscribing because it's just not worth it, I would guess for every million videos of clickbait there is one video that is actually has value from being informative

  • There was no need to phrase this as a question: if you have ever used YouTube subtitles, you'll know they are bad.

    The real question is why. While I think AI is overhyped in general, I don't believe that this is the best Google could do if they put in some effort. It seems the model is just picking the most likely word on an individual word level, not considering any context like the video title or description (which can contain the very names it gets wrong), the type of content the channel typically publish

  • Not on YouTube, but on the Firefly DVDs(!): "You're a rabbit, Mal!"
    • My most memorable subtitle WTF was also pre-AI, a documentary series about the American West via Prime. Don't even remember what it was (because that how my memory works), but it was a whole phrase phonetically substituted instead of select words.

  • Somewhere between the 25th and 50th percentile of a stenographer that hasn't pre programmed names.

  • Not all (Score:3)

    by PPH ( 736903 ) on Saturday May 02, 2026 @07:33PM (#66124740)

    Some [youtube.com] are OK.

  • Appallingly bad (Score:3)

    by ConceptJunkie ( 24823 ) on Saturday May 02, 2026 @07:44PM (#66124758) Homepage Journal

    I rarely see _any_ subtitles that are not appallingly bad, including YouTube.

    • Re: (Score:2)

      by 0123456 ( 636235 )

      Beat me to it. Yes, Youtube subtitles are generally appallingly bad, but when I watch movies with subtitles on DVD or Bluray they're often also garbage. In some cases it appears the person creating the subtitles is working from a different script to the movie as shot because the words bear no relation at all to what they're saying on the screen.

      • In commercial content, I think this is more of an issue with dubbed audio tracks. On the other hand, subtitles tend to be fairly faithful translations of the original audio, which -for practical reasons- is often not an exact match for the dubbed version. In fact, a good dub is usually adapted to the target languageâ(TM)s customs rather than being a literal, word-for-word translation. Additionally, dubbed dialogue has to account for lip-sync so it doesnâ(TM)t break the viewerâ(TM)s immersion

        • Re: (Score:3, Interesting)

          by 0123456 ( 636235 )

          It's possible. But in some cases we've been watching with English subtitles on an English TV show with people speaking in English because my girlfriend doesn't understand English English. The words in the subtitles still don't necessarily match the words they're saying in the same language.

          • Playing devilâ(TM)s advocate. Thereâ(TM)s a âoelimitâ on how much text the average viewer can âoereadâ per second -so to speak-. This means professional subs sometimes âoesummarizeâ or change some wording to something simpler to not overload the screen⦠OR⦠maybe this show had really awful subs *lol*

      • Not sure what movies you're talking about, but closed caption subtitles on a typical Bluray are word for word transcribed in literally every movie in my vast collection. There is of course the occasional hiccup, a typo or word missing, but if that happens once in any feature film it's already worse than normal.

        Are you talking about foreign language films or something?

        • Re: (Score:3)

          by 0123456 ( 636235 )

          Here's one current example. My girlfriend is watching a detective show on Amazon Prime.

          This is an American show where the characters are speaking English and the subtitles are English. Not even sure what show it is, but it clearly has a decent budge that could afford to spend a few bucks creating the subtitles.

          Main character's daughter is fixing main character's computer and says "I exported all your data."

          Subtitle says: "I extorted all your data."

          Now, the girl's accent is a bit heavy, but it's clear from t

    • I watched a Japanese pronunciation video which had been run through English speech-to-text processor to create subtitles. Even the English words were incorrectly detected.

      I mentioned last week, that such processors can't handle homonyms: Always writing "brake" instead of "break".

  • Nah (Score:1, Funny)

    by Anonymous Coward

    Does YouTube not know that there are places in the world that are not in America?

    Why should YouTube be any better than the average American?

  • BUT if you Subscribe! We will fix the captions. ;-)

  • Heat (Score:2)

    by troff ( 529250 )

    Every single stupid time there's just music playing and the stupid subtitler prints "Heat. Heat" and I'd like to see the whole of the YouTube/Google management complex to experience flashpoint/ignition heat.

  • Over the years there have been many lists published of lyrics people thought they heard but were completely nonsensical. To some apparently Abba sings, "feel the beat of the tangerine."

    So I'm not surprised youtube can't understand song lyrics. Regular speaking, it's fairly good. And not too bad at translating the subtitles.

    • Over the years there have been many lists published of lyrics people thought they heard but were completely nonsensical. To some apparently Abba sings, "feel the beat of the tangerine."

      If you were ever beat by a tangerine, you'd think differently, you insensitive clod! ;^)

  • YouTube's subtitles turn themselves on after every ad for some reason.
  • Than Google's voice to text.
    I use it pretty much every time I'm writing something (like right now)
    And it is pretty darn High error rate.
    For instance why did it capitalize High and why did it make has into is?
    Some of it, I'm sure, has to do with working with really poor pronunciation. Most people tend to blur and slur when they're speaking.
    And then depending on the listener to straighten out what it was they said.

    • Re: (Score:1)

      by Anonymous Coward

      I don't think they're really any worse.
      Than Google's voice to text.

      Fun fact, youtube auto-subtitles are produced with none other than googles first gen voice to text!
      So you're exactly right, they are not worse and not better, it is the same thing.

      It's also been in place a lot longer than most people think.
      Over 10 years ago, about when they made their v3 api available, all uploads had their audio transcribed to text and stored with the other metadata. It was accessible and searchable via API.

      It was many years later before they started using it for subtitles.

      Other fun facts

  • tiny model? (Score:3)

    by algaeman ( 600564 ) on Saturday May 02, 2026 @09:16PM (#66124874)
    I use Whisper to generates subtitles on occasion. If you use a tiny model, it is pretty terrible. If you use a large model, it does a pretty good job, even doing phonetic spellings of unusual names. The large model takes some real cpu time to do the analysis, and YouTube isn't gonna burn money for free features.

    • That makes sense. What doesn't make sense to me is that I feel that it has gotten worse. I remember thinking that a few years ago that live transliteration seemed to be getting pretty good. I remember testing out early Dragon speech-to-text software and not being impressed, so maybe I had set my expectations so low that that even muddling translations were a real improvement, and I misremember how well it was really doing.

  • AI has always been and will continue to be terrible with accents. Each new accent is like a whole new language it has to learn. Niche accents will never have enough training data.

  • Same thing happens with some newer animes on Crunchyroll. The CC is sometimes complete nonsense. If the content is live, I can totally understand the CC being a little rough. But the content I'm watching is pre-recorded and has a script. The script could easily be provided, stripped of everything not needed for CC, and then synced to the video. Hell, use the AI to sync it, but there should never be any reason an AI is just making up random, wrong shit for pre-recorded content that has a script available. It

    • Re: (Score:1)

      by drn8 ( 883816 )
      imgur doesn't work. I can't figure out which of the dozens of domains to allow to get an image to show up. Use postimg or something more sane.

  • Closed Captions are even worse. It's in many news outlet Youtube channel.
    It's like it was translated to some other language and back to English with a 90's translator, with lot of missing words, in all capital letters and with a 15 seconds delay.
    Mind you, are the authors that provides the content of these hateful CC's.
    They're supposed to be at least decent, because they're (supposedly) human-made and directed to hearing impaired people.
    Oh and Youtube don't let you fallback to regular automatic subtitles. It

  • I happen to know a lot about this, and am familiar with the state-of-the-art text-to-speech models, and for some reason, Google definately isn't using one. My guess is that they have so many youtube videos to transcribe each day that they just dont have the compute to use a state-of-the-art model to do it, and have to use a fast cheap one with a higher error rate. If that is the reason, then Google should provide the option to people posting videos to pay for having their video transcribed by a SOA model.

  • Not just YouTube. Instagram is almost unusable. YouTube shilling their shorts and auto-dubbing is pushing me to the edge.

  • The modern trend of mumble acting doesn't help. When actors transitioned from stage to screen they came armed to annunciate. They could even speak clearly and act at the same time.

    Admittedly my hearing isn't great, but it pisses me off when actors mumble and slur their way through dialog because they and the director both think that makes the performance brooding or edgy.

    The AI can be forgiven for getting it wrong when regular people also struggle to hear it.

    Watch great actors speak. Listen to Hopkins as Le

  • They COULD be good (Score:1)

    by Anonymous Coward
    They are bad because they are using a lightweight model with a supertight budget. If the translation model could spend some (or quite a lot) more compute on the translations, the results would be much much better.
  • They never match the spoken words and are slightly worse than the auto-generated YT crap that can at least be turned off.

  • It depends on how much money you are willing to burn per token. For offering free subtitles on YT? The quality is going to be "budget" tier AI. We have a massive scalability, capacity, and energy problem to overcome before AI actually does a decent job at average people's mundane tasks.

    Of course capitalism is not generally going you give away a higher quality service than it needs to. Especially if the end user is not a customer but the product.

    • Re: (Score:2)

      by AmiMoJo ( 196126 )

      My Pixel phone has on-device voice recognition, and it's extremely accurate. Like better than a human's first pass accurate. There are some things it struggles on, like odd place names, but even then if it's something like Google Maps it tends to figure out what I meant anyway.

      So I'm not sure why YouTube is so bad, because clearly Google has better technology than this and with YouTube has the luxury of being able to take its time, rather than having to produce real-time output.

      • For starters, there are a lot of tricks used in devices. You don't need inference to do decent voice recognition for dialing and commands.

        Offline inference is pretty spotty for me on my Pixel 9A. Like unable to understand me well enough to take down an address. But on the other hand when it has good signal it is good enough for sniff conversations in the room and start recording products my wife and I are discussing. Probably because the engines are far better trained on product names than on place names. J

  • The UK ball-drawn lotteries are shown live on YouTube and have auto-gen'ed subtitles. With the official UK lottery site scrapping a list of the balls in drawn order in Jan 2026 (they only have them in ascending order now), it's remarkable that the only official text form of the drawn order of the balls is that subtitle text stream.

    Yep, that official subtitle stream has regular parsing errors such as "[music]" instead of a ball number (yes, there's video background music that can interfere with the narrator'

  • Not everything needs to be autogenerated. Upload something with subtitles.

  • So if there are pro bikers speaking: it does not now the names of teams, brands, bikers, routes or climbs. It just tries to hear, but still plain stupid. Aliens visiting this planet are laughing their asses off.

    Why is there not a checkbox called context in the video upload form, check cycling and he should learn that for this video. Is that so hard to come up with?

  • Auto-generated subtitles on all platforms are appallingly bad. The trouble with them is that they do generate some text that is good enough for most people not to bother to watch the video and correct them before posting.

  • The ones I love are the animated ones where there is text printed on the frames and the automatic comments textify the spoken portions. Its entertainment on its own, the mangling of proper names are even random so the same botched rendering hardly ever repeats. An entertainment all by itself. I routinely play youtube at 1.5x, tried 2.0 but the speach compression totally fails at that rate. Just watching stuff for entertainment so really dont care about the accuracy. I prefer words in a row to hearing a talk

  • They could use a better model and spend 50x the cost to generate 30% better subtitles, no doubt.

    It's good enough for their "summarize the video" feature which I now use to avoid listening to AI narrators. Especially the ones programmed to maximize runtime with nonsense, repetition, and restatements.

  • A ton of videos have big, colored, hard-coded subtitles, especially for voice-overs. Those are incredibly distracting and the absolute worst. At least you can turn these off.

  • do you think YouTube's subtitles are "appallingly bad"?

    No, they're much, much better than having no generated titles at all, and by a lot.

    mpv won't generate any subtitles for me; it can only show what's included in the data.

  • Whenever I watch a Star Trek-related video that happens to touch on the species that oppressed Bajor, Youtube will almost always render the name as "Kardashian".

  • I did some video where I had a script. So it was easy to extract subtitle and compare to script. Yeah, its pretty bad. Probably every other sentence had some obviously incorrect subtitling. And these videos were on clean audio inputs: no background noise, good mic.
    I ended up having to correct the subtitles using the transcript.

    I have seen a study recently on noisy audio in a group setting, so the speakers maybe not be perfectly mic-ed and all. The speech-to-text was done with a couple of Whisper models. The

  • Opening a French video out of the blue and hearing some weird English translation no one asked for instead of simply adding subtitles is such an awful example of American exceptionalism...
    I'm not necessarily going to set my browser to every language I want to hear with subtitles, but auto-voice translation is just wrong.
    If you managed to get to Youtube, you can read and choose to activate it if you want, but stop with this English-centric view.

  • Yeah, those subtitles can be pretty 'orrible, no question. I've often wondered if there was work there for humans, until I remember that most humans' spelling, grammar, punctuation are pretty weak too. Still, better than nothing, and I do miss it when the subtitles are not available. And it's always a surprise when I turn them on for a particular thing, and find them to be in Russian (!) or Spanish or German or whatever. I leave them on (well, except for the Russian ones because I can't read the script

  • While manually entering them on Youtube is hassle, it comes down to the laziness of the creators.
  • Final Cut Pro is really good at generating transcriptions, and you can easily export them for import into YouTube (or you can burn-in the subtitles if you want them on all the time for everyone).

    YouTube is really bad at auto-generation, and in particular gets punctuation wrong much of the time. There are better solutions out there than YouTube, and it should be something every creator does by default rather than leaving it to YouTube.
  • Lately, my love *has* been too lava.
  • As someone is pretty hard of hearing, even when I have my hearing aids in, I often struggle to make out conversations in movies/TV Shows. I watch mostly Youtube and very little regular TV and the subtitles really help.

    Yes, they are sometimes inaccurate, but I'm using them to supplement what my ears are hearing. For example, someone might say, "50 CAL" and the subtitles say, "50 COW" but even though I might not hear "CAL" properly, the subtitles help me figure it out, despite being wrong.

    It's like th

Slashdot Top Deals

"Experience has proved that some people indeed know everything." -- Russell Baker

Close