Tv Archive AI pipeline // Internet Archive, Dan Schultz

Notes

  • SLIDES: https://docs.google.com/presentation/d/17s8tvqYY8zb4cGVk0MPi2HeA85af-x0macLn-WijAdc/edit?usp=sharing

  • Internet Archive! https://archive.org/

  • Petabytes of data! The TV Archive! The TV News Archive! Recording TV, cable news, Philadelphia local news, CSPAN (people LOVE it).
  • There's an API written in Python Clip-based metadata storage. "This clip was fact-checked, this clip is a politlcal ad" Check it out: https://archive.org/details/tv What is in the data? VIDEO / AUDIO / CAPTIONS / but also information about… PROGRAM / CONTEXT / AUDIENCE Not all of this data is easy to grab but each have signals where data can be pulled from.
  • [ see excellent slide with detailed examples for each ]
  • How can we use these signals to provide more information?
  • Speaker diarization / who is talking?
  • Experimentation at the Internet Archive:
    • They want to enable experimentation as a library, not necessarily do it (but they do it too)
  • Example: Political TV Ad Archive (https://politicaladarchive.org)
  • Frequency analysis, turning sound into a hash and easily searchable. FIngerprinted each political ad and found all other copies with that ad.
  • Example: Face o matic http://tv-research5.us.archive.org:8000/auth
  • Recording in real time, 4 hours to convert into mp4 and other things, run through an algorithm that "doesn't take much time at all"
  • This is a prototype: What do YOU want to use it for? - What format would you like this in?
  • Example output in #face-o-matic on hyperaudio slack
  • Example: Chyron Extraction Extracting the "lower thirds" and how do they compare? That will be discussed tomorrow*

  • The Vision: What's happening on TV? What's all the metadata associated with what's happening? They don't have a great API for these things but want to use Sockets to push things out.

  • Be able to look historically, last month, etc. INTERNET ARCHIVE IS A LIBRARY! Dan wants to make sure you know this. ← it’s true.

  • Question:

    • Can you add more data to the archive? How can submitters use all this cool stuff?
    • "Good question."
    • "Reach out to me, basically." → DAN @slifty SCHULTZ
    • "We should talk" [email protected] everyone talk to him!

results matching ""

    No results matching ""