Brewster Kahle

AI Audio Challenge: Audio Restoration based on Expert Examples (we have 1,600 examples that could train a model to do it on the other 400k recordings)

this would help the internet archive and could be fun for AI hackers.

http://blog.archive.org/2023/04/24/ai-audio-challenge-audio-restoration-based-on-expert-examples/

AI Audio Challenge: Audio Restoration based on Expert Examples - Internet Archive Blogs

Hopefully we have a dataset primed for AI researchers to do something really useful, and fun– how to take noise out of digitized 78rpm records. The Internet Archive has 1,600 examples of quality human restorations of 78rpm records where the best tools were used to ‘lightly restore’ the audio files. This takes away scratchy surface […]

Internet Archive Blogs
Apr 24, 2023 at 6:23pmWeb
Show threadKent PitmanApr 24, 2023

@brewsterkahle

I'm curious: Train the model for how to act on data, or how to manage controls for tools that algorithmically adjust data?

Show threadBrewster KahleApr 25, 2023
@kentpitman I would say either, but I am really hoping that some magic happens where the AI's learn how to clean up directly.
Show threadmirabilosApr 25, 2023
@brewsterkahle just. stop. that. so-called “AI”. nōnsense.
Show threadJon UdellApr 25, 2023
@brewsterkahle Great idea! It falls into my category of things you can ask a model to do that yield results that are easy for humans to validate.
Show threadChristian SteinmetzApr 27, 2023
@brewsterkahle Nice. I am wondering if you also have the original unrestored recordings from the 1600 examples?
Show threadBrewster KahleApr 27, 2023
@csteinmetz1 yes, the unrestored are in the item. either in the zip file or top level