Split youtube videos into chapters task

Youtube automatically analyzes uploaded videos and split them into chapters with meaningful titles.

Does anyone knows how this task is named in the literature? Is it just video classification?
Is there any service that does that?


How does one turn this functionality on? because we’ve been manually setting timestamps for part1 videos of this course?

Video summarization would fall into this category. I’m not aware of a video-based service. I use snipd for podcasts, though.

See option " automatic chapters"

Not sure of the quality, haven’t used myself, not a YTber. Still, I didn’t understand why I couldn’t find something like this in Google Cloud among the AI products.


I haven’t tested yet, but I found this service here: Introducing Auto Chapters - Summarize Audio and Video Files

1 Like

Wow that is very interesting

You might be interested in checking out our demo for an application that attempts to tackle this problem …

Our team is actively working on this application so if you have any thoughts on it, good or bad, it would be great to hear them


That is so interesting! Does it rely on the audio of the video only (video → audio → transcriptions → chapters + summarization) or does it take the video into account to divide into chapters? I guess that for lessons, audio is good enough.

How can I know more about this project?

Our current topic segmentation model only considers the transcribed audio (I think we want to be able to apply this to audio without visuals … e.g., podcasts, interviews, recorded lectures, etc…).

We did a poster session for the FSDL course and may be doing another. If so, I’ll try to remember to post something here and/or on twitter. Once we recover from the week we worked to get the demo out and formalize things a bit, I’ll be able to share more information about the project as a whole.

1 Like

As Wayde mentioned, we(me, @wgpubs & @kurianbenoy) will probably post more formal information once we recover from the last week a bit, but in the meantime you can check the tweets below.

And YES, you can definitely try out the MVP. It’s running on CPU servers, so it’ll take a while(depends on the video length) but it’ll be done. Hopefully, we can keep this alive for a while.

(P.S. - PLEASE DON’T SUBMIT the Part 2 2022 Fast.ai course urls, as these are not meant to be public yet, and everything on our MVP is currently public.)

Would love to hear feedback. :beers:


An updated loom and link to our presentation describing our approach can now be found here: Project Showcase - Full Stack Deep Learning