MySpace to Use Audio Fingerprinting

dptalia writes "MacWorld reports that MySpace is going to start implementing audio fingerprinting to prevent copyrighted material from appearing on their site. The new technology will be used to review all uploads and prevent 'inappropriate' material from ever seeing the light of day."

Re:Just like real finger printing today...

[arun_s]

The 'article' is woefully low on information, apart from a mention of Gracenote MusicID being used. From Gracenote's own page [gracenote.com] (Its on mobile music recognition, but I assume the principle is the same):
How it Works
1. When music fans hear a song they want to identify, they tap a command on the phone keypad to start the audio recognition process, and then hold the phone up to the music source.
2. The phone captures a few seconds of the audio and extracts a waveform fingerprint of the snippet. The snippet can be from any section of the song, even the last few seconds.
3. The fingerprint is sent to the Mobile MusicID recognition service from the service provider that may be located anywhere in the world.
4. The Mobile MusicID recognition server compares the fingerprint to its database of reference fingerprints and responds with the exact match.
5. The artist, song title and related information, as well as content like album art and download links are relayed to the fan.

Gracenote's own article on this

[in2mind]

Gracenote's own article on this:

http://www.gracenote.com/music/corporate/press/art icle.html/date=2006103000 [gracenote.com]

Re:Just like real finger printing today...

[IKnwThePiecesFt]

But... we're talking about MySpace, not YouTube?

I think this is where the confusion comes in...

Re:Silver lining...

[joshetc]

Its ok because they analize it when it is uploaded and block it from being uploaded.. not every time the file is downloaded.

Re:I've been waiting for this moment

[IDontAgreeWithYou]

No one else gets hurt? Let's just imagine that you are in a moderate accident. Say another car's tire blows out. Their car loses control and crashes into yours, legally making the accident their fault. Your car rolls over an embankment and you are thrown out of the car and killed. Now the person has to live with causing an accident that killed you, when you would have been fine had you worn a seatbelt. Plus, do you really think that first responders enjoy scraping your dead ass off the highway or that other motorists want to see your internal organs spread out all over the road, all because you weren't quite comfortable enough with a seat belt on. No, it definitely does hurt other people.

Re:I've been waiting for this moment

[stunt_penguin]

Excuse the fuck out of me, but when the vehicle you are travelling in hits a solid object (like another vehicle) then most of your momentum remains until *you* manage to hit something solid. If you're a passenger in the back of a car, then that solid object is likely to be the person in the seat in front of you. If you're in the front and it's a side impact, then there's a 50% chance that you're going to slam into the driver at most of the speed your vehicle was travelling at. The results are never pretty.

It's bloody minded ignorance like this that makes the roads a more dangerous place. I don't mean this unkindly, but I just hope you get a harmless scare that is bad enough, or see an accident victim cut up bad enough to make you wear your belt. You sure as hell aren't going to listen to reason.

Re:The evaders will win

[upeters]

I developed a system using AudioID technology (http://www.m2any.com) from IIS Fraunhofer (the people who created MP3) during the past year. I can assure you, a few distortions won't harm the audio recognition at all. The company I developed the project for works with commercials which should be aired several times each day by radio stations. To check if the commercials were actually broadcasted, the whole radio programming was recorded during 12 hours, and then analyzed with AudioID. Note that some radio stations (AM and FM) were pretty distant, so distortions were a given. And to keep the recording in managable sizes, the recording was sampled with "telephone" quality, keeping the 12 hours recording about 40 MB in size. The files were scanned with AudioID in about 5 minutes (or a bit more if the audio had lots of background noise), and was able to find commercial spots, most are only 30 seconds, with 100% confidence. The software uses overlapping time frames, and if the sound pattern of any part of the recording was similar to a fingerprint stored in the database, where we kept our commercials which should air during that day, a log was produced pointing where a spot was found. Naturally, if a spot uses a certain background music, and the radio station happens to play that music, an entry is created as well. However, the confidence will point out that this isn't the spot and it won't reach 100%. To evade the software, there can't be no 4-seconds interval where the song still sounds similar to the original, or else it will be matched against the fingerprint. I am not sure if the Gracenote software (licensed from Philips) is as powerful as the Frauhofer's product, though. It was too expensive to even consider it for testing.

Re:The evaders will win

[(H)elix1]

I wonder how well this will actually work.

The technology works surprisingly well on a cell phone. One of the guys I work with in the UK showed me Shazam. I picked a random track from his vast MP3 collection, he dialed a number and held out his phone for a half a minute, and shortly thereafter they SMS'ed him the artist. Not a quite background either...

http://www.shazam.com/music/portal/sp/s/media-type /html/user/anon/page/default/template/what_is_tagg ing/music.html [shazam.com]

Take that same technology and do it on what should be even cleaner audio than what you send over a cell phone speaker - I suspect they could get most of it.

It may be hard to defeat, I worked on audioprints

[jdoeii]

Depending on specifics of the algorithm, it may be very hard to defeat it if you still want the music to be recognizable by the listeners. I am familiar with the audio fingerprinting [sloud.com] algorithm from another company. The false positives are not a problem. The hash space is huge thus collisions are very rare. The false negatives can be a problem, but if they can weed out even 95% of attempts to upload copyrighted music, their life is going to be much, much easier. And if you distort the music enough to defeat the fingerprinting, then maybe you just have created a new masterpiece (c) you :-).