[HN Gopher] Show HN: CTRL-F for YouTube Videos
       ___________________________________________________________________
        
       Show HN: CTRL-F for YouTube Videos
        
       This is a small project i made years ago and updated to whisper
       last year, i still use it from time to time and thought it might be
       useful to others, or just put the idea out there for someone better
       than me to make a better implementation!
        
       Author : ewild
       Score  : 55 points
       Date   : 2024-04-13 15:17 UTC (7 hours ago)
        
 (HTM) web link (github.com)
 (TXT) w3m dump (github.com)
        
       | popf1 wrote:
       | That's cool, but there is also Firefox extensions that does
       | something similar. There's one for searching comments, and one
       | for searching caption.
       | 
       | https://addons.mozilla.org/en-US/firefox/addon/youtube-capti...
       | 
       | https://addons.mozilla.org/en-US/firefox/addon/ycs/
        
         | ewild wrote:
         | ahh never really looked cause i built my original one in 2019
         | off of Deepspeech haha just updated it for fun mostly. I know
         | youtube captions themselves are good, but one thing on his code
         | would be that not all videos have captions. Since mine actually
         | downloads the audio and runs it, it would still have values on
         | those older videos that never got captions
        
         | sammyatman wrote:
         | Ctrl-F across all of youtube: https://www.askyoutube.ai
        
           | modeless wrote:
           | https://www.youglish.com is also a sort of search engine for
           | YouTube captions, though mostly aimed at short phrases.
        
           | popf1 wrote:
           | I searched for my YouTube username and then for the exact
           | title of one video I posted and it didn't find either one....
           | instead it said the title of my video was not true because it
           | didn't interpret it correctly (but it didn't link to the
           | video).
        
             | sammyatman wrote:
             | I'll look into it, could you send the request to
             | askutubeai@gmail.com?
        
       | modeless wrote:
       | Ctrl-F can already search the transcript on YouTube. I use it all
       | the time. I guess this could be useful for videos YouTube doesn't
       | have captions for.
        
         | stainablesteel wrote:
         | im not able to do this, can you explain?
        
           | atahanacar wrote:
           | You can find a button for the transcript in the description
           | (or the three dot menu near the dislike button if it's still
           | serving you the older interface). You have to open the
           | transcript first, then Ctrl+f
        
           | modeless wrote:
           | 1. In a desktop web browser, visit a YouTube video with
           | captions, which is almost all of them
           | 
           | 2. Click the video description to expand it
           | 
           | 3. Scroll down and click the tiny "Show Transcript" button
           | near the bottom (whoever decided to bury it down here was
           | very misguided)
           | 
           | 4. Ctrl-F and search any word. Occurrences in the transcript
           | will be highlighted and you can press enter to scroll the
           | transcript to the next one. Click the transcript to seek the
           | video.
           | 
           | I see that this extension shows occurrences on the seek bar
           | which is cool. There is also a slight problem with regular
           | ctrl-F: if you search for a multiple word phrase you might
           | not find it if the phrase happened to be split between two
           | chunks of the transcript. So that could be better in this
           | extension. And of course not every YouTube video has
           | captions, but most do these days.
        
             | ewild wrote:
             | you are correct, originally youtube didnt have this when i
             | made it in 2019 with deepspeech, now they do but i just
             | always preferred the idea of it being on the timebar to
             | just click and go right to it. tbh i should just make a
             | simple addon to take the youtube timestamps and slap it
             | onto the timebar. also for the split chunks this would have
             | no problem there as the transcript is actually stored in a
             | json file, so any concurrent words will always be matchable
             | for phrases. ofc downside being you need to run the model
             | lol
        
               | modeless wrote:
               | I'd use an extension that made the transcript show by
               | default on every video and added a transcript search bar
               | in the page. That would be great.
        
               | ewild wrote:
               | i guess i might aswell do it so i dont need to run a
               | model everytime myself too lol ill have it done in a day
               | or two
        
               | ewild wrote:
               | would you prefer if the timestamp was hidden since it
               | takes up a bigass portion of the screen or that being an
               | option to hide it in the extension settings?
        
               | modeless wrote:
               | I think the timestamp is OK, my biggest complaint is the
               | huge amount of whitespace between the rows and the small
               | size of the box. If I designed YouTube I would put the
               | transcript on the left side above the video description,
               | with a button that expands it to full height so there's
               | no separate scrollbar for the transcript anymore, it's
               | just all directly in the page.
               | 
               | BTW when I went to look at a video just now, YouTube
               | actually served me a "Search in Video" box at the top of
               | the transcript. So I guess the feature exists, they just
               | haven't rolled it out to everyone yet.
        
               | ewild wrote:
               | damn i see this after im 90% done and just have to make a
               | fancy button lol
        
             | philsnow wrote:
             | > visit a YouTube video with captions, which is almost all
             | of them
             | 
             | Depending on what you're watching, you might never come
             | across a video with good subtitles but rather Youtube's
             | auto-generated subtitles.
             | 
             | Whisper can do a better job in a lot of cases, but not
             | all... I wonder if they've had multiple generations of
             | auto-captioning and not gone back and redone the ones that
             | were done earlier.
             | 
             | This extension is really interesting to me because in the
             | past I've tried (and failed) to make a similar one that
             | adds a new .vtt to the list of available subtitles for the
             | video. I sometimes struggle with auditory processing,
             | especially in a noisy environment, and following along with
             | subtitles helps me out immensely, so it's frustrating when
             | the auto-generated subtitles are poor quality. I've
             | bookmarked the extension to see if I can fork it for that
             | purpose in the future.
        
               | ewild wrote:
               | if you have any questions feel free to ask!
        
       | lopkeny12ko wrote:
       | https://filmot.com/
        
       | qntmfred wrote:
       | I've been using https://www.appblit.com/scribe to get transcripts
       | into a more readable/ctrl+f-able format
        
         | ewild wrote:
         | yeah I remember the whole transcript youtube coming out a
         | yearish after i made the first version of this in 2019, but i
         | still perfer the timebar highlighting, but thats just a
         | preference thing
        
       | BetterWhisper wrote:
       | Developed https://www.videototextai.com/ exactly for this reason
       | as it was quite impossible to search videos otherwise. Also you
       | can copy the transcript into a LLM and ask questions from video
       | content like that.
        
       | a_wild_dandan wrote:
       | If we had an extension to skip all the filler garbage in YT
       | videos, I would be ecstatic. Maybe that's doable now? YT captions
       | -> identify fluff timestamps via a browser LLM -> insert segments
       | onto the video timeline, which automatically skip, a la
       | SponsorBlock.
       | 
       | We could slash through Youtubers repeating themselves, making
       | hack jokes, narrating their video title & outline, vapid
       | explanations of common knowledge, etc. Any of which can be
       | customized to your taste via a system prompt!
       | 
       | This kinda semantic filter would actually be an immensely
       | powerful UI tool for _all_ webpages and media, now that I think
       | about it...
        
         | sergiotapia wrote:
         | just use sponsorblock today, works fine on all my devices.
         | https://github.com/ajayyy/SponsorBlock
         | 
         | from mobile phone to tv to pc.
        
           | jsheard wrote:
           | Do check the settings too, SponsorBlock is best known for
           | skipping sponsored segments but it also has markers for
           | things like intros, previews, self-promotion, and filler
           | jokes/skits which aren't skipped by default but can be if you
           | want them to.
        
         | vidvizgpt wrote:
         | Hey! We're building an extension that is working on this
         | problem. It's not perfect, but we're kind of in an open beta.
         | Currently working on video and soon other forms of content.
         | 
         | https://vidvizgpt.com/
         | 
         | https://chromewebstore.google.com/detail/vidviz-gpt-ai-for-y...
        
       ___________________________________________________________________
       (page generated 2024-04-13 23:00 UTC)