Dataset Viewer
Auto-converted to Parquet Duplicate
video_id
stringlengths
11
11
title
stringlengths
2
101
text
large_stringlengths
1
778k
duration_seconds
float64
5.31
54.5k
source
stringlengths
0
75
url
stringlengths
0
43
priority
int64
5
9
speed_ratio
float64
6.1
611
vidCX_dMCu0
Lec 02. How to Train a Neural Net
Okay, so welcome to lecture two of deep learning. Today we're going to talk about how you actually train a neural network. And I imagine that some of this might be a review for some of you but hopefully kind of the perspective or the way we look talk about the process of training the neural network might be a slightly ...
4,773.952
9
67.6
8zzfcYIELdo
Lec 15. Generative Models: Representation Learning Meets Generative Modeling
We're going to continue on today and this week with generative models. So today we'll be on variational auto-encoders. And we're going to give a perspective that generative modeling and representation learning are very tightly coupled. If you can solve the generative modeling problem, it should help you solve the repre...
4,839.317333
9
63.7
7hbf4klU3ks
Lec 20. Scaling Laws
Okay, so today we're going to talk about scaling loss. And this lecture is, you know, we're at the final few weeks of the semester. So this is, we're moving into territory, which is much more kind of ongoing science. Like I'm gonna talk about recent papers and results that are empirical in nature for which we don't kno...
2,302.037333
9
69.9
IiHknRHA-Gk
Lec 10. Architectures: Memory
"So today we're going to be doing our last of this mini series on machine learning architectures or (...TRUNCATED)
4,407.616
9
70.7
bxVkZ4M-hIE
Lec 04. Architectures: Grids
"So today we're going to start off the first of a series of lectures on machine learning architectur(...TRUNCATED)
5,036.309333
9
65.9
zaMcHuJwe1w
Lec 16. Generative Models: Conditional Models
"OK, so welcome. Happy Halloween. Nice to see a few little costumes. OK, so today we're going to tal(...TRUNCATED)
4,891.285333
9
63.1
-eC0-5mXHQg
Lec 13. Representation Learning: Theory
"Okay, thank you everyone for coming to the lecture. So this was the example on the, do people remem(...TRUNCATED)
4,520.106667
9
73.7
QxOzQRtd440
Lec 11. Representation Learning: Reconstruction-Based
"Good. So welcome to today's lecture. And today we're going to talk about representation learning. S(...TRUNCATED)
4,863.616
9
65.7
RUdQMHV-7KM
Lec 19. Transfer Learning: Data
"We're going to finish up our short series on transfer learning today. And the other sort of logisti(...TRUNCATED)
4,543.104
9
68.2
hJlrAHqGOS8
Lec 14. Generative Models: Basics
"Okay, yeah, welcome everyone. So now we're going to get coming to the, um, another section of the c(...TRUNCATED)
4,877.909333
9
67.3
End of preview. Expand in Data Studio

Massive YouTube Educational Transcriptions

Large-scale educational content transcribed from YouTube using distil-whisper/distil-large-v3.5.

Stats

  • Videos: 59,355
  • Characters: 1,539,022,925 (~384M tokens)
  • Audio hours: 35,890
  • Model: faster-whisper (CTranslate2) with distil-large-v3.5
  • Hardware: 2x RTX 5090 + 2x RTX 4090 at 165-185x realtime

Fields

Field Description
video_id YouTube video ID
title Video title
text Full transcript
duration_seconds Original audio duration (seconds)
source Discovery source (channel/course/playlist)
url YouTube URL
priority Educational priority (9=university, 8=lecture, 7=documentary, 5=general)
speed_ratio Transcription speed (realtime multiplier)
content_category Content type (university_lecture, conference, individual_educator, etc.)
license_risk License risk level (green/yellow/orange/red)

Content Categories

  • green: CC-licensed or public domain (NPTEL, Khan Academy, MIT OCW, Yale OYC)
  • yellow: Standard YouTube license, fair use for research
  • orange: Commercial content, needs review
  • red: Non-educational, excluded from transcription

Methodology

  1. Discovery: Channel crawling, related video walking, CC-focused search
  2. Quality filter: 40+ reject categories, duration >= 15min, 3-tier priority scoring
  3. Transcription: faster-whisper CTranslate2, 1.2x audio speedup, beam=1, no VAD
  4. Classification: Channel name -> title regex -> priority fallback

Code

github.com/thepowerfuldeez/massive_yt_edu_scraper

License

MIT

Downloads last month
122