Comment by elliekelly | Hacker News Reader

Comment by elliekelly | Hacker News Reader

Parent Full thread

Since this is posted here I'll ask a question that has been bugging me for the last year: Who does the captions on these YouTube videos? Is it automated? I ask because every CS193p video starts with the same opening scene/music and the caption describing the music is different on every single one. Is this an auto-caption algorithm just getting it wrong? Or is there a person having fun behind the scenes waiting for someone like me to notice and spend an irrational amount of time wondering what happened? I feel like it has to be a person, right? YouTube's captioning wouldn't give 12 different results.

Edit- For anyone curious, here's how the music is described in the captions at the beginning of each lecture:

1. regal music

2. mellow ambient music

3. light music

4. gentle music

5. air whooshing

6. gentle music (turns out there is one repeat)

7. peaceful music

8. [no description]

9. ethereal music

10. celestial music

11. serene music

12. dramatic music

I have a feeling that there is a chance that this comment has made some anonymous subtitle transcriber with a sense of humour very happy.

Oh, these are (mostly?) manual (often done by students working with SCPD, if I recall correctly). Any student can sign up and record for a class after some basic training, which I think can also include CC'ing some parts of it.

It is definitely someone just having a bit of fun :)

I think you are right - people sometimes have too much fun with subtitles: videos from https://www.youtube.com/technologyconnections typically subtitle the jazz at the end of the videos with a relevant adjective (e.g. 'cool jazz' or something for a video on HVAC operations.)

I know for a fact that Alec either transcribes those videos himself (not hard to do from a script he already writes for himself), or gets it done by someone else before uploading the video.

I'm any case, they're always uploaded at the same time as the main video.

Also, YouTube recently removed the ability for individual users to contribute to captions of creators. There was a thread on HN about it.

people sometimes have too much fun with subtitles

If you watch WKRP In Cincinnati on MeTV, the subtitled lyrics for the closing theme are different every time.

YouTube automatically transcribes captions for its videos. So that is why they are so funny sometimes.

As a YT creator, it is possible to upload your own captions if you have any transcribed, which will then be used in place of the automatic ones. But 99.9% of YT videos are using automatic captions.

Edit: Ok, I started watching the first video in this series and these are manually uploaded captions. So Stanford got the videos transcribed and manually uploaded the captions to Youtube. I wonder if various students did the captioning. So each student described the music differently. I laughed because the first video describes the music as "Regal Music".

I wonder if various students did the captioning

It's possible. But there are plenty of companies that do subtitle transcription pretty cheaply.

Local TV newscasts used to pipe the scripts from their TelePrompTers into the closed captioning encoder, but that wasn't good enough when the law changed in the late 90's because then there wasn't any text for interviews or weather segments.

Now you can hire a company like Denver Caption that will have someone dial into a special phone line at the TV station and transcribe the content in real time, sending it directly back to the station's encoder.

This is why when more than one person is talking, you can see in the text how the person captioning gets overloaded. Or they might use the wrong form of a word or phrase that would be easy to pick if someone was watching video, but easy to get wrong if they're only listening to audio over a phone line.

Given the precision of things like ponctuation, and certain words like ViewModel being correclty spelled (instead of "view model" for example), those captions are probably manually edited.

The YT automatic captions are marked as such and are usually easily distinguishable because the subtitles appear word-by-word instead of whole lines at once.

My guess is that those are manual.

My partner worked for a bit as a transcriber on Rev, which I believe is the source for a bunch of these— they have a turnkey option for "caption Youtube video X": https://www.rev.com/

Stanford hires students to transcribe all classes that are also available through SCPD. My suspicion is that some bored student out there is having some fun.

The course's videos have options for both "English" and "English (auto-generated)" captions. The non-generated one is uploaded by whoever uploaded the video.

Reply on news.ycombinator.com