The Role of Content Identification in the SNOCAP System
SNOCAP uses acoustic fingerprinting technology to identify audio content submitted for registration. In this article, I explain a little about acoustic fingerprinting, how it differs from other methods of content identification, and why this level of identification is important for SNOCAP and the integrity of our open registry.
First, it’s important to know exactly what is represented in the SNOCAP Registry. The Registry contains three types of information: 1) sound recordings, as embodied by audio files, 2) rights holder information, describing the artists and labels who own sound recording rights, and 3) distribution terms, describing how the rights holders want to make their audio content available. The SNOCAP Registry is described as an “open registry”, meaning any rights holder is free to create a SNOCAP account (providing the rights holder information) and register tracks (providing the audio files representing sound recordings for which they own or control rights). To prevent misrepresentation of rights ownership, it is important to compare newly submitted content against content that has already been registered. This is where the content identification technology comes into play. We have to make the comparison based on acoustic properties of the file submitted, since it is the sound recording that is being registered, not just a specific file containing the sound recording.
If audio files were all created the same way and could not be altered without significantly degrading their quality, perhaps acoustic fingerprinting would not be necessary. Files could be identified by using a standard hashing algorithm based on the bits in the file. But there are many ways to create a digital audio file and the tools and parameters available for doing so generate files that contain different bits, but sound more or less the same to the human ear. So, standard hashing techniques are inadequate to serve SNOCAP’s content identification needs.
Audio file formats contain the ability to add metadata describing the file contents. If this metadata were always accurate and could not be changed, then perhaps this information would be sufficient to identify a particular sound recording. But metadata fields are easy to change, and the tools and services that supply the metadata (based on track information ripped off a CD and looked up in a database, for example) vary. Metadata is certainly not sufficient when making definitive claims about copyright ownership.
Acoustic fingerprinting is like the automated equivalent of having a person listen to all the audio files submitted to SNOCAP and remembering which ones have been heard before. Just like a human, the fingerprinting system must accommodate noise or variations in sound quality, or even versions that have been shortened by eliminating portions of the song. To accomplish this, the fingerprinting system slices the audio file into short frames, and for each frame calculates information about the acoustic features contained within. When this data is reduced into numerical values, it can be compared to other data frames in the system. If enough of the frames from a new track match a previously registered track, we determine that it is sufficiently similar to constitute the same sound recording.
Some acoustic fingerprinting systems only analyze a small subset of a track against a database of fingerprints. This approach is fine if you only want reasonable assurance that the track in question is contained in your database (to look up track metadata or get recommendations for other, related music, for example). But when validating representations of ownership, it is necessary to analyze the entire file and compare all analyzed segments against the database. This is a computationally intensive process, but it is necessary for our purposes. Additionally, the acoustic fingerprinting technology we use is built to scale linearly, so that lookup time is constant no matter how many fingerprints are in the system.
There is another use of acoustic fingerprinting technology in the SNOCAP system. SNOCAP provides services that identify user-submitted content or content made available by individual users on a file sharing network. In these cases, it is necessary to obtain an acoustic fingerprint for the files in question and use this information as a key to look up ownership and distribution rights specified in the SNOCAP Registry. When dealing with user submitted content, the effectiveness of the system in matching altered versions of audio files against the registered masters is paramount. And to scale to the volumes associated with such systems, it is critical that fingerprints can be generated efficiently on the client systems and used for lookup purposes.
The ability to reliably identify content is central to the integrity of the SNOCAP Registry. All retailer and consumer-facing offerings enabled by SNOCAP rely on this capability. This solution must work in the case where masters submitted by rights holders are compared to previously registered content, and also the case where user-submitted and shared content is identified and the resulting fingerprint used to look up corresponding distribution rights. The acoustic fingerprinting technology employed by SNOCAP was selected based on its ability to perform in these cases and do so in an efficient and scalable manner.
See also: http://ismir2002.ismir.net/proceedings/02-FP04-2.pdf, a description of the acoustic fingerprinting algorithms that form the basis of the SNOCAP content identification system.
March 2, 2007 at 3:57 pm
[...] fingerprinting is a technology used for “content identification.” Content identification helps to protect original works. It can also identify false claims [...]
March 6, 2007 at 4:45 am
Thank you for taking the time to review and comment on our agreement. I’m writing to confirm and stress that what you call SNOCAP’s “pitch” is indeed true: SNOCAP is non-exclusive. It’s your music. You own it.
Terms of Service agreements are legal documents and yes, legal-ese
language inevitably creeps in. We’re always reviewing ours with the
goal of clarifying the language (in fact, we’re in the process of
changing the wording in the agreement this week). I believe you have
misinterpreted our terms in a few places and I want to offer the
following in hopes that I can shed some light on the areas you have
identified.
1) Terms of service for any service do change from time to time. We’re
a new business using a fast-evolving technology in a rapidly growing
Internet marketplace and as we add new products and features, updates to
the terms of service might be required. We inform our artists of
substantive changes we are making, and the artists always have the
opportunity to opt-out of the agreement if they don’t agree to the
changes.
2) We communicate changes and updates to our artists both through our
web interface and when possible, through email. We can’t rely on
reaching everyone by email because sometimes emails don’t go through
(those wicked spam traps!) and sometimes email addresses change without
our being notified. Artists using the service will be visiting the
interface regularly to upload new tracks, check on sales, track payment,
etc., so the interface is an obvious place to post any changes to our
terms of service.
3) An acoustic fingerprint is just that, an internal identification
mechanism. Fingerprints alone can’t be used to distribute or exploit in
any way artist content - they are completely separate from the sound
recording rights that are registered in our system or the audio files
that are sold through SNOCAP’s MyStore. We keep fingerprints because
the integrity of our content identification process relies on this
database; those fingerprints don’t confer any rights to SNOCAP for the
use of the content.
4) SNOCAP cannot and will not play your stuff however it sees fit. We
ask for the right to modify content solely for the purpose of generating
30 second clips so that consumers can sample music before they buy, just
like iTunes, Amazon and others. We ask for the right to do this without
paying mechanical royalties for each play - that would make it
economically impossible to provide the preview feature.
5) SNOCAP bears all the costs of running our service and there are
administrative costs for maintaining the database and individual artist
accounts that go beyond what is covered in the $.39 per track. Thirty
dollars a year is very competitive when compared to other music
distribution services. We waived the fee in the first year of service
to lower the barrier to entry and encourage artists to try us out, and
we reserve the right to waive this fee in the future depending on other
business offerings or considerations. And, if at the end of the first
year of service an artist chooses to not continue the service, then
there is no obligation by the artist at all going forward.
I know it takes a lot of words to say it, but it’s your music, you own
it. And we hope you’ll set up a SNOCAP MyStore soon.