Soundcloud - now scans for copyrighted material

So i've been doing a talk radio show with my sister for the past year, recently when we upload our show to soundcloud it actually will block our upload because it finds within seconds (This is not a real person listening to it and detecting this, its an algorithm) a copyrighted snippet of a song we're using for the bumper music. What other websites do this? I find this a very dangerous precedent for anyone who's made a DJ mix or who's sampled copyrighted material in their own music. I don't want to give the mainstream record labels any ideas but what if they just buy up a huge chunk of all the digital music out there and as a death cough retroactively sue everybody who is flagged on this 'copyright' detector algorithm.

so i guess mixcloud doesnt do this, which would make sense why so many people are moving over there. I'm just kind of shocked that i paid for a premium soundcloud membership and a year later it starts doing this.

edit: if anyone here is thinking 'this has never happened to me what the hell is awepittance talking about' try right now to upload a mix or something with a section from a song from a major record label, it wont let you

As someone who never samples and who writes all original material, I'm hardly concerned by this.

But I do see where you are coming from.

PS-

I'm really impressed that their algo can pick up snippets in the middle of a talkshow.

  On 4/5/2011 at 9:13 PM, acidphakist said:

As someone who never samples and who writes all original material, I'm hardly concerned by this.

 

But I do see where you are coming from.

 

PS-

 

I'm really impressed that their algo can pick up snippets in the middle of a talkshow.

i imagine the algo was gicen to them?

but this awful, why do they want to stifle peoples creativity, this makes me so angry! choking on my own words here!

vhs head is in trouble

so what sites dont do this ? So far it seems to be only youtube and soundcloud, im sure its only a matter of time before the algorithm is sold off to whoever else. Right now it seems to only have in it's database things signed to big major record labels.

  On 4/5/2011 at 9:19 PM, Awepittance said:

so what sites dont do this ? So far it seems to be only youtube and soundcloud, im sure its only a matter of time before the algorithm is sold off to whoever else. Right now it seems to only have in it's database things signed to big major record labels.

on the plus side, there is now a new genre centering around hoodwinking The Algorithm, and sneaking copyrighted bits by.

Have there been any news stories written about this? Seems like it just kind of snuck by

Very interesting. I wonder how sensitive the algorithm is at detecting the copyrighted material. If you applied a filter to the song or altered it only slightly, would it still recognize it as that song?

that's a great question, i've been wondering the same thing. I guess i wrongly assumed an mp3 encoded song was unmatchable in a spectral analyzer to its original

  On 4/5/2011 at 9:13 PM, acidphakist said:

As someone who never samples and who writes all original material, I'm hardly concerned by this.

 

That's a little bit solipsistic. surely you have friends who put out DJ mixes and things like that?

i mean i could understand if you're not personally concerned about your own stuff, but i think a great deal of your fellow musicians will be effected by this.

  On 4/5/2011 at 9:49 PM, patternoverlap said:

Very interesting. I wonder how sensitive the algorithm is at detecting the copyrighted material. If you applied a filter to the song or altered it only slightly, would it still recognize it as that song?

it is likely "heuristic" in the sense that it applies a whole litany of techniques. does the spectrogram match? 10 points. does the rhythm match? 10 points. etc. etc. and if it's over 50 points it triggers the copyright slammer.

a human (or a human using a pool of statistical data) will have gone in and set up the weights -- how many points the spectrogram gets, and how many points before it gets flagged etc. you can either calibrate it to get most copyrighted stuff, but also block legal things by accident (false positives) or let some copyrighted work slip through, so legal things are far less likely to be accidentally blocked (false negatives). the weights are probably re-tuned regularly, new matching techniques added, etc.

i am betting they err on the side of caution, as blocking legal crap would piss a lot of people off, so there's benefit of the doubt you can use to your advantage.

i'm not terribly much on the maths of doing FFT matching etc., but you could make a list of the most sane techniques to detect copyrighted material, write a song intended to bust up the effectiveness of these techniques, and then try it out. the most challenging and fun part would be trying to suss out exactly what sets it off!

it's like a stupid video game i play. you design cars to race, and you can set a license plate. it won't let you use "dirty language." my immediate reaction to this was to try and find a way to sneak as much filth by it as possible ! down with control.

interesting theories. i wonder how much this algorithm costs if one were to buy it

  On 4/5/2011 at 10:15 PM, Awepittance said:

interesting theories. i wonder how much this algorithm costs if one were to buy it

probably a subscription service. what isn't, these days?

now here's music by the founder of Soundcloud, Eric Wahlforss:

http://www.youtube.com/watch?v=qpFJEOnoirk

cut, paste and build something. quite a paradigm shift indeed eh?

Whether they scan everything uploaded or they use reporting methods (there's a button "report copyright infringement"), I'm not that shocked (and yes, I have a premium account too) since, to my knowledge at least, it's a common thing on such social networks (youtube for instance). I make DJ sets but haven't had a problem so far with any of my mixes, and yes, some of them contain one or two major label tracks, although they are mostly made of independent labels music. I wouldn't say I'm shocked by this since soundcloud never claimed to be an underground or rebel website. It's a behaviour to be expected from such fast-growing websites that deal with music sharing.

so you have not encountered this yourself, i guess the only time it 'caught' me is when i was trying to put a 20 second clip from a Soft Cell and then a PJ Harvey song in the radio show

any other places you've heard about doing this besides youtube/soundcloud?

would it recognise a song in a mix if it wasn't at the exact tempo of the original?

either way there are plenty of other good site to go too and it does stop skanks ripping off other peoples work I suppose

I haven't encountered this myself. I have only uploaded full mixes so I don't know whether (if) they are fully scanned or not.

I'm gonna try this right now.

What about Bandcamp, don't they have the same thing over there ?

This is stupid though, reagrding dj sets or radio shows, a track from your label being played can't hurt your sales.

What, specifically, did it say when it rejected you? Just curious to see if it is, in fact, based on copyright...

Trying to upload Micheal Jackson's Smooth Criminal.

You have a point Awepittance.

  Quote

Your upload "1-05 Smooth Criminal" from April 5, 2011 contains "Smooth Criminal (Radio Edit)" (Michael Jackson) which is owned or licensed by rightsholder Epic/Legacy and has been made inaccessible by request of the rightsholder.

 

If you are certain that you have all necessary permissions from the rightsholder to upload and share this content, you can submit a claim below.

Again, even with the downloads disabled.

  Quote

Your upload "I just cant stop looooovvvving youuuuu" from April 5, 2011 contains "I Just Can't Stop Loving You" (Michael Jackson) which is owned or licensed by rightsholder EPIC and has been made inaccessible by request of the rightsholder.

 

If you are certain that you have all necessary permissions from the rightsholder to upload and share this content, you can submit a claim below.

i think XXX is onto something. The Shazam 'app' i believe is free on all platforms. How to make money off something like that? Sell it to the wealthiest rights holders and let them wreck havoc on all media sharing sites

  On 4/5/2011 at 11:11 PM, soundwave said:

would it recognise a song in a mix if it wasn't at the exact tempo of the original?

this i'm not sure of, i haven't tried to trick the algorithm yet. It would be useful i think for people to know the existing work-arounds for uploading stuff to get by this algorithm. I'll write an article about it for her website if i can find surefire ways.

unfortunately it's only a matter of time before Bandcamp and other sites have it. Another side of this whole thing that i'm curious about how do you as an independent musician or label owner who has copyright claims join in on this database? Why should i not have the same sledhehammer power to stop people from uploading music from my label as these megalomaniacal companies do?

edit: here are some things i'm going to try to get past it

-bandpass, hipass, lowpass filtering

-pitch shifting/time stretching/overall speed adjustments

-reversal, will it detect a song in reverse?

-varying lengths, what is the shortest snippet it can detect?

if anybody else wants to join in on the fun try what anonymous forumite did, upload smooth criminal after these or other adjustments