A musician who made a 10-hour long video of continuous white noise - indistinct electronic hissing - has said five copyright infringement claims have been made against him.
Sebastian Tomczak, who is based in Australia, said he made the video in 2015 and uploaded it to YouTube.
The claimants accusing him of infringement include publishers of white noise intended for sleep therapy.
"I will be disputing these claims," he told the BBC.
In this case, those accusing Mr Tomczak are not demanding the video's removal, but instead the reward of any revenue made from advertising associated with it.
Without the claims, Mr Tomczak would receive such revenue himself.
"I am intrigued and perplexed that YouTube's automated content ID system will pattern-match white noise with multiple claims," he said.
His video was originally made along with other 10-hour recordings - including one of a continuous electronic tone.
The claims relate to specific portions of similarly lengthy videos of white noise also posted on the site.
http://www.bbc.co.uk/news/technology-42580523
Very interesting. Looks like the content ID algorithm itself can be affected by placebo effect like human. It it driven by deep learning?
Obviously an uncredited cover version of something ;-)
Reading the article, I found this: https://juliareda.eu/2017/09/when-filters-fail/ The web page of a Pirate Party MEP - politics alert, but the examples are ridiculous:
- a purring cat faced a takedown. The cat in question, Phantom, has filed a dispute and hopes to reclaim his rights.
- a Harvard Law School lecture on copyright
As well as public domain material and the artist's own work.
It should be noted here that copyright claims are not as severe as takedowns. AFAIK, with claims, you still get the ad revenue back after the claim is removed, for the duration the content was claimed.
Random guess: both white noise samples were encoded using similar or the same encoders, and since there is nothing in the source audio except random noise, the pattern matching algorithm noticed the pattern introduced by the encoder's requantization and confused that for the signal.
Random guess: both white noise samples were encoded using similar or the same encoders, and since there is nothing in the source audio except random noise, the pattern matching algorithm noticed the pattern introduced by the encoder's requantization and confused that for the signal.
I would not bet too much against "white noise" appearing similar to "white noise", especially not if the ID system is tuned to catching background music in cellphone recordings.
It should be able to distinguish "white noise" and noise music, though ... IMHO.
That said, I know nothing about how the generators draw the pseudo-random sequence in practice. The signal does not have to pass "randomness" tests in order to sound like white noise, I guess ... uh ... I bet some self-proclaimed golden ears will dispute that.
By the way, someone wrote the story into this Wikipedia entry: https://en.wikipedia.org/wiki/White_noise
Some people inverted one audio channel in their videos to bypass the system. It sounds like garbage when playing on the speaker of my phone due to mono mixdown.
https://youtu.be/MrpmqUldAas
Some people also deliberately desync their audio and video, covering logos on video, mirroring etc. These algorithms remind me of antivirus software, and those so-called "heuristic" analysis...