Comment by bscphil | Hacker News Reader

Comment by bscphil | Hacker News Reader

Parent Full thread

A related issue: in some cases the industry has allowed anyone to upload a copy of a song, and so you run into the issue that you're at the mercy of whoever the uploader was when it comes to the quality of your upload.

I ran into a case where there were reports that a specific release of a CD has a much better mastering than other versions, it was a Super Audio CD (SACD) as I recall. I looked up the release on YouTube, since differences in mastering are clearly audible even in a lossy encode. The one upload was by a random person who had put it up in an HD video, but the volume of the release had inexplicably been boosted to the point that the sound was heavily distorted!

Another case: the most popular version of the Eminem "Lose Yourself" music video (over a billion views) is uploaded by some random person, with one of those ludicrous "no copyright infringement intended" messages in the description. It's seemingly the one that the industry approved as the legitimate version of the music video on YouTube, however. I actually tried tracking down the official version on the artist's page, and (at the time) it had actually been taken down by the studio with one of the standard DMCA takedown messages YouTube shows!

Regarding the Eminem song, I found this on Reddit, don't know if it's true:

> YouTube allows copyright owners to choose to leave videos/channels up when they report them and simply divert all monetization from the videos to the copyright owner. That is likely what is happening here.

Could be some sort of authenticity play from Eminem, him wanting to look less "corporate".

I've been uploading a copyrighted show on a small YouTube channel[1] for fun for a long time now. This is exactly what happens[2] and why I've never had a strike or anything.

[1] https://www.youtube.com/c/uktotheworld/videos [2] https://i.imgur.com/r2Gl9cB.jpg

That's interesting phrasing. "This video was found to contain copyrighted material." So what? Tons of stuff contains copyrighted material, under fair use. But there's no acknowledgement of that and it's not obvious that they would allow you to dispute that your use of the material was fair use. (Clearly, in your case, that's not at issue here. I just think it's strange.)

Is there some deep-learning tool to reconstruct songs that got distorted in various ways?

"Content aware fill" works on the basis that you just want "some grass" or "some rocks" and don't know or care whether it matches the exact grass that was in the original picture.

If people have already heard the original song, then the computer isn't going to be able to fill in the lost information in a way that matches their memory.. even with deep learning

I'm not convinced. Counterexample: if the original song is a sine wave, and the distortion turns it into a square wave, then a simple algorithm can turn it back into a sine wave with the right frequency, very easily.

Counter-counterexample: your algorithm was too naive. The original song was much more interesting:

A sine wave at a single frequency, but with rhythmic volume swells from nearly inaudible to 0dBFS. The slow hopeful crescendos in the opening give way to a more playful bounce in the middle, before building to a frantic ending with nearly nauseating swings.

Pitchfork gave it a 10.0, and called it “Better than the remaster of In The Aeroplane Over The Sea. If our scale allowed, it would have gotten a 12.” What a ride.

The distortion that transformed it was fairly pedestrian: turn it up until it exceeds the dynamic range of the recording medium.

The result was that boring square wave at constant frequency and amplitude. Can you expect an AI to reconstruct this original, given both of our proposed original songs map to the same output?

I’d also argue that the original proposed algorithm is not simple, since it would have to remove the aliased frequency components when converting from the square wave back to a sine.

Reply on news.ycombinator.com