LWN: Comments on "Echoprint: Open acoustic fingerprinting"

MPEG-7?

rzm — Fri, 08 Jul 2011 16:37:16 +0000

I wonder how into all this fits MPEG-7. Is it impractical? To complicated?

Echoprint: Open acoustic fingerprinting

lukaslalinsky — Thu, 07 Jul 2011 14:42:00 +0000

There were these two options:

1) Proprietary solution
2) No fingerprinting at all

I think it's a fairly obvious choice. If there was an open solution back then, MusicBrainz would have used it, but there wasn't and MB was never interested in developing its own fingerprinting technology because it's not the primary goal of the project.

Echoprint: Open acoustic fingerprinting

jamesh — Thu, 07 Jul 2011 07:06:14 +0000

Of course, this isn't the first time MusicBrainz has had its proprietary fingerprinting solution disappear from under them.

The PUID system they currently use was a replacement for a different proprietary system called TRM. The company behind the TRM fingerprinting had moved on to other things, and the server software was unreliable for the load it was being put under.

It was a bit surprising to see MB switch to another proprietary solution after their prior experience, but it is good that they've finally got an open system they can rely on.

Echoprint: Open acoustic fingerprinting

giraffedata — Fri, 01 Jul 2011 00:22:39 +0000

That link says that they've performed one billion identifications

Thanks for that. I couldn't derive that meaning either from the referenced web page or LWN's corrected text.

It's one of the language foibles that always irritates me: people say "I did X to N things" when they mean "I did X to something N times." Like when a transit agency says 50,000 people ride the the bus on a typical day, when they really mean there are 50,000 boardings (by about 22,000 people) on a typical day.

Echoprint: Open acoustic fingerprinting

xtifr — Thu, 30 Jun 2011 19:53:01 +0000

No way there are [one billion] songs in the world.

Actually, I suspect there are far more than that! The number that have been recorded and distributed on the Internet may be much smaller though. But even there--there's a lot of music coming out of Japan and India, you know. Not to mention the rest of the world. As for not-recorded--my niece made up at least two songs in the last week alone! :)

It's also going to depend on how you define "song". Is the legendary Jimi Hendrix cover of Bob Dylan's All Along the Watchtower a different song from Bob's original? I suspect most people would say yes, but then what about the version Bob recorded in collaboration with the Grateful Dead? Is that the same as Bob's original? Or is it the same song as the version the Grateful Dead released on their own a couple of years later? Both? Neither? What about the over 1.5 million liberally-licensed live tracks hosted on the Internet Archive's Live Music Archive? (The overwhelming majority of which are from the USA.) What about the guys I regularly see on the streets around here selling privately made CDs of their own group's work (some of which probably does end up on the Internet)?

I suspect your estimates of the current database sizes are accurate, but I'm not so sure about your proposed theoretical limits.

Echoprint: Open acoustic fingerprinting

iabervon — Thu, 30 Jun 2011 17:35:32 +0000

It looks to me like this is evidence of the opposite: it's great to rely on a proprietary component, because a better open-source component will arise, either before you need it or soon after. In general, it's often good to start with some dependency as a temporary measure, until there is something really good available. The fact that having your dependency be proprietary is doomed in the long run is actually a benefit here, because it means that you will actually get around to making the transition.

Echoprint: Open acoustic fingerprinting

jengelh — Thu, 30 Jun 2011 15:23:43 +0000

Nothing beats Perl's soundex! :-)

Echoprint Database License?

giggls — Thu, 30 Jun 2011 12:48:13 +0000

Looks like ODBL (http://www.opendatacommons.org/licenses/odbl/) of Openstreetmap fame would have been a perfect match here.

Echoprint: Open acoustic fingerprinting

felipec — Thu, 30 Jun 2011 09:51:35 +0000

Years ago I told MusicBrainz guys the dangers of relying on a closed source component so much.

Now, if there only was a way to compile for Linux.

Echoprint: Open acoustic fingerprinting

pabs — Thu, 30 Jun 2011 08:08:47 +0000

There is also a more general perceptual hash library:

http://phash.org/

Echoprint: Open acoustic fingerprinting

jimparis — Wed, 29 Jun 2011 20:06:33 +0000

Shazam claims more than one billion songs in its database.

No way there are that many songs in the world. That link says that they've performed one billion identifications. This article from Dec 2008 says that Shazam's database holds 8 million songs. So a "resolvable catalog" of 13 million songs for Echo Nest sounds pretty good!