I guess so.
I have checked. The captions are mostly in sync with the video now. I think they can be improved once Jeremy wiki-fied your post. Thank you for the super hack.
I guess so.
I have checked. The captions are mostly in sync with the video now. I think they can be improved once Jeremy wiki-fied your post. Thank you for the super hack.
I guess Youtube only saves the last 2 hours of Live Video, which might have messed with your timings
Thanks a lot for you effort. Hack is indeed mind-blowing
Brilliant hack. Love it
Thank you! This made my day
Pretty fast ; This is really impressive. Thank you.
I’ll be releasing an edited video soon BTW - so I’m afraid these timings are all going to change! Generally I release an edited video 2 days after the lesson.
No worries. If people find this useful now i am glad. I think it is possible to generate new timestamps when the edited video arrive by comparing the transcripts from before and after the edit since they contain timestamps.
While thinking of that, i realized one can take the transcript and easily convert the xml to markdown with links using regex replacments. I find Sublime text to be really useful for that as you see your regex matches while you type them.
Here is an raw example:
(I changed so that each sentence is a link instead of each word)
Perhaps someone can find it useful. To make this again with the new video is very straightforward.
Pre break part 1:
so let's welcome both of them to the
stage and I'm really excited about this
kick-off so please hello everybody
thank you David and thank you PJ this is
a great space so for those of you that
are sitting in the audience you're
actually a small fraction of the people
who are watching right now there's
…
[edit] removed rest as this pointed to the old video and made scrolling a pain. I added one for the new video in a secret gist.
pre break part 2:
- the other thing to mention is that the
faster a library's in a real designed in
a very interesting modular way and
you'll find over time that when you do
use import star there's far less
clobbering of things than you might
…
[edit] removed rest as this pointed to the old video and made scrolling a pain. I added one for the new video in a secret gist.
You are awesome man!! thank you so much.
Post break part 1:
Post break:
so we we've not so hopefully oh great
so where we got to was we just we just
trained a model we don't exactly know
what that involved or how it happened
but we do know that we're three or four
lines of code we've built something
which smashed the accuracy of the
…
[edit] removed rest as this pointed to the old video and made scrolling a pain. I added one for the new video in a secret gist.
That’s good! Are you using google voice recognition or something to create that transcript? Got a script we can see?
post break part 2:
but to really get it good you have to
call unfreeze and unfreeze is the thing
that says please train the whole model
and then I can call fit one cycle again
and uh-oh
the error got much worse okay why in
order to understand why we're actually
going to have to learn more about
exactly what's going on behind the
scenes so let's start out by trying to
get an intuitive understanding of what's
going on behind the scenes and again
…
anyway, these will be redundant soon but are easily re-generated
[edit] removed rest as this pointed to the old video and made scrolling a pain. I added one for the new video in a secret gist.
Just posted this but may get lost since you added more transcript.
If you click the CC button on the video player, it loads youtubes auto-generated transcript.
if you inspect the ajax calls using the inspector you see that it loads an xml file.
that file can be manipulated into markdown or html etc.
<s.*?>
with empty string (we don’t need that each word is a link)</s>
with emptry string.*t="(.*?)".*?>(.*)</p>
with [${2}](https://youtu.be/7hX8yKCX6xM?t=${1})
[0-9][0-9][0-9]\)
with )
to get rid of last 3 digits in timestamp.then you should have a markdown list of linked sentences
Have you seen this hack?
He created a caption and timestamp search tool for fast.ai lecture videos using a few Python scripts and libraries like youtube_dl
and webvtt
.
That’s awesome.
Re-edited video released! Thanks Jeremy!
I updated the post so that it links to the edited video and adjusted the times. The post could perhaps be wikified so that people can adjust it if they want. I have found the links useful myself when trying to recap some of the things mentioned.