Vulnerability Development mailing list archives
Audio fingerprinting (was Re: hacksdmi?)
From: Geoff Schmidt <geoff () TUNEPRINT COM>
Date: Sat, 14 Oct 2000 02:02:51 -0400
I'm very curious what use the watermark has. It seems that image recognition techniques could be applied to music such that a particular performance could be reliably identifed without adding anything to the format.
Of course destroying the recognizable patterns bypasses this; then you have what I would consider a different (possably derivate) work.
Funny you should mention it :) Check out the audio fingerprinting algorithm we're working on at www.tuneprint.com. The idea is that the fingerprint is based on the large-scale psychoacoustic features of the music, so changing the music enough to damage the fingerprint is like running the audio through a shredder. The plan isn't so much to use the tech to enforce copyrights (although if anyone wants to investigate along those lines we're not going to stop them) as it is to build kickass search engines/peer-to-peer networks/popularity charts/collaborative filtering/etc. Once you know what track that mp3 is a recording of, you can do lots of cool stuff. But more on topic, lossy encoding and inaudible watermarks are kind of like oil and water -- from an information theoretic point of view, you've got a certain amount of redundancy in a series of PCM samples (ie, a raw WAV file) and you can either squeeze that redundancy out and make the file smaller (lossy compression) or inject a signal and use that space to store information (watermarking.) In other words, an mp3 encoder's function in life is to strip out the chunk of spectrum (loosely speaking) that a watermark algorithm uses. So every advance in lossy audio compression is also an advance in watermark stripping. In particular, the next generation of 'generative' audio codecs are going to be a big challenge to the current breed of watermarking tech. (I heard that Monty, the mastermind behind the new Vorbis format, has already had some thoughts about generative models, but don't quote me.) On the other hand, it only takes a couple of bits to store a watermark, and there's huge amounts of redundancy in audio, so the hide and seek game could go on for quite a while. geoff
Current thread:
- Audio fingerprinting (was Re: hacksdmi?) Geoff Schmidt (Oct 13)
- Re: Audio fingerprinting (was Re: hacksdmi?) Thierry (Oct 14)
- Re: Audio fingerprinting (was Re: hacksdmi?) Geoff Schmidt (Oct 14)
- Re: Audio fingerprinting (was Re: hacksdmi?) Bluefish (P.Magnusson) (Oct 16)
- Re: Audio fingerprinting (was Re: hacksdmi?) Geoff Schmidt (Oct 14)
- Re: Audio fingerprinting (was Re: hacksdmi?) Lincoln Yeoh (Oct 15)
- Re: Audio fingerprinting (was Re: hacksdmi?) Geoff Schmidt (Oct 16)
- Re: Audio fingerprinting (was Re: hacksdmi?) Thierry (Oct 14)