If you’ve been on YouTube, and had one of those dreaded emails warning you to take your video down, or face the music. Despite the panic reaction, it did grate a little that every little piece of music, even if it played from a TV from all the way across your living room, could be the little pin that punctured your 10 minutes of internet notoriety.
THE INS AND OUTS OF YouTube’s ContentID SYSTEM
For starters ContentID is the reason why YouTube can detect and remove any video which doesn’t meet its standards, promptly after you upload it. It’s also the reason why so many entrepreneurs are coming up with so many channels. The more popular the video, the more you are sampled in other user videos. When they get popular, you can decide to share, the ad revenue generated by their video or take it down.
The software works by breaking down the video into its audio-visual components, and creating a fingerprint of its video and audio spectrum, finding the nearest, not the closest matches (which is why you won’t get away by simply transcoding your video into a new format).
The software however isn’t 100% original. The video fingerprinting technique is Google’s own, but after a spate of threatening lawsuits from the big guns of the evil, money hungry and righteous music companies, YouTube officials were hastily forced to license Audible Magic’s technology for audio fingerprinting.
Turns out, thanks to Scott Smitelli the programming, behind the whole corporate greed and copyright infringement scenario, can actually be duped. The tech behind audio fingerprinting is surprisingly tenacious in parts, and simply retarded in others. For instance it can detect almost any variation in the volume, and isn’t fooled by plain white noise anywhere in the track (unless it covers an overwhelming 50% of the track), at the same time, it detects only the first 30 seconds of a track, for its sampling needs. Huh?
The best part however, lies in the genius of pitch variation. Scott was pretty thorough in his tests, and found a lucky gem which shows once the track’s pitch is increased to 5% and beyond, the software fails to detect it. Now a 5% increase in pitch may sound excessive, but surprisingly, it’s undetectable (unless you know the song back to front, subliminal messages included). Here’s proof (UPDATE: video no longer available).
Scott has had more than 35 ContentID warnings for his tests (the entire range of 82+ test videos can be found on his channel “retnirpregnif”, that’s reverse for fingerprinter), but his account still seems to be up, and running. Go figure. (UPDATE: no longer working)
Technology exists to be bested. It’s the only way it can evolve. But some of the stuff WMG and other media moguls tend to do with their blanket bans, do deserve rotten tomatoes. Even now, YouTube doesn’t give an option when it detects even a snip of a WMG copyrighted audio clip in your vid, simply choosing to mute it and run that awful red banner underneath, but hopefully that will change.
Copyright infringement is a serious issue, considering the harm it brings to the artist, still most of them are more than happy, to share the revenue generated by a video specifically honoring and highlighting their creations. The methods outlined by Scott are just inconsistencies in programming software, not tips and tricks to upload copyrighted data on YouTube. It’s up to you what you decide what you put up on YouTube.
That still means no lolicon AMV’s.