Slashdot Mirror

Sounds kinda like Relatable's TRM fingerprints, which are used by MusicBrainz and in the Neuros audio player.

IIRC, the fingerprints don't have any actual content in them, but instead describe the characteristics of the audio. So it's plausible, at least, that they can't listen in on your conversations, but could still uniquely identify what you're listening to.

In my experience, it works fairly well and only gets confused when the same recording has been released on multiple albums ("best of", remasters, etc).

Therefore every time you submit your MP3 TRM's to MusicBrainz, who in turn pass them onto relatable, his company can use that data to identify the songs on the P2P networks.

Far more accurate (although slower) than looking at the title of the files. Additionally, changing the metadata within the MP3 won't make a difference.

"One major concern with these interference technologies is that they will block files (in this case music) which are not illegal and thus hose "legitimate" uses of p2p technology, of which there really are a ton."

Of course it's a concern. That much is obvious. Thus I think we can safely assume that Audible Magic has thought of this. If I were heading up the project, it would be the #1 priority on my list. After all, why give the P2P companies any free ammunition?

"Which is to say, even though Napster traded in primarily (c) violating content that doesn't also mean that the filtering tech put in place wan't also massively over-inclusive in what it filtered, and if it was, well, that just really sucks."

You are referring to the "acoustic fingerprinting" technology. It was provided by Relatable. They are still around. You are correct in the abstract that if the Relatable technology were producing a lot of false positives, that would be a shame. But my web searches haven't found this to be a talked-about issue. It may be that Relatable, like Audible Magic, has a firm grasp of the obvious (perhaps they hired their coders off of Slashdot) and made accuracy their #1 priority.

No, it didn't. There are "hashing techniques" specially made for audio - "audio fingerprinting" so to speak, like Relatable's TRM and Gracenote's MusicID which do a great job of it. They identify the file correctly no matter what the source is - lossless audio CD, or even 128kbps MP3, you get the same fingerprint.

I've tried TRM personally through MusicBrainz, and ran it on around 1000 of my MP3s, some of them really horrible quality, and it managed to identify 99% of them (TRM fingerprint correlated with actual metadata is stored at MusicBrainz). I was surprised, but yes, it did work. And this technology is rather old too, I'm surprised not too many people know about this.

And the article specifically mentions this fact: ...The experts' claims center on technology for detecting copyrighted works through "fingerprinting" (sometimes also called "hashing") technology that identifies songs by analyzing the content itself. Such technology, which is provided by several firms including Audible Magic, GraceNote, and MediaGuide...

No, it didn't. There are "hashing techniques" specially made for audio - "audio fingerprinting" so to speak, like Relatable's TRM and Gracenote's MusicID which do a great job of it. They identify the file correctly no matter what the source is - lossless audio CD, or even 128kbps MP3, you get the same fingerprint.

I've tried TRM personally through MusicBrainz, and ran it on around 1000 of my MP3s, some of them really horrible quality, and it managed to identify 99% of them (TRM fingerprint correlated with actual metadata is stored at MusicBrainz). I was surprised, but yes, it did work. And this technology is rather old too, I'm surprised not too many people know about this.

And the article specifically mentions this fact: ...The experts' claims center on technology for detecting copyrighted works through "fingerprinting" (sometimes also called "hashing") technology that identifies songs by analyzing the content itself. Such technology, which is provided by several firms including Audible Magic, GraceNote, and MediaGuide...

I have decided to move to the not so popular .ogg format. Will this throw off the RIAA? I'm curious just how tech savvy these chumps are.

First, if the files are different in any way at all, wouldn't it give a completely different hash? Let alone a different format. What about encoding at a different bit rate?

Second, does the RIAA know about the ogg format? If everyone moves over, how long will it take before they notice?

Third, wouldn't it be better to use an audio fingerprinting scheme like musicbrainz uses to tag your files? It's similar to hashing but uses the actual audio qualities in the file.

Seems the RIAA is not only about 20 years behind in their business model, but they are about 20 years behind in technology.

My issue with the RIAA is, why can't they just say to the courts "This guy is downloading music confiscate his computer." There are no watchdog groups to make sure these guys are actually verifying that someone has the copyrighted information.

Can you imagine the bandwith costs the RIAA have just to download the number of files they do. Just so they can check the hashes on each file? Verifying the legality of files must be extraordinarily costly. Wouldn't it be great to start flooding their network with their own fake files just to WASTE (link pun intended) the RIAA's bandwith and time downloading the junk they disseminate?

Sorry, the sig field is temporarily out of order, you will have to read whatever I write here.

As a class project, a friend and I built a music recognition database. You can read our paper.

The general approach is fairly straightforward. You extract a set of "features" (typically several Mel Frequency Cepstral Coefficients, or MFCCs) from each sample of the song, say 10ms. You then pick several (say, 16) arbitrary points and iteratively generate that many "average" feature vectors, along with their weights so that they all sum to a one vector. This data is turned into a Hidden Markov Model (HMM). To see what audio you have, you run it through each of the possible HMMs and see which produces the greatest likelihood.

This method is typically applied to speaker recognition, where a linear search through HMMs is reasonable. This obviously isn't the case when you know about hundreds of thousands of songs, so a large part of the challenge is narrowing the field of HMMs to check (which is one of the focuses in our paper). Relatable, who were working with Napster a long time ago, have clusters that can classify 1,000 songs per second; I'm pretty sure they use this technique.

This technique has several important features. First, it doesn't depend on any properties of files themselves. Checksums would be trivial to beat, looking at a file's length could be circumvented by inserting silence, etc. Since this creates an average of sample data, a song would need to be changed quite a bit to fail to match. (The system is robust to, for instance, changes in bitrate, slowing the music down, and rearranging bits of the song or putting it in reverse.) We didn't have enough "derivative" music to test how it handles sampled music vs. the original -- it depends how much is changed.

Finally, this sort of system is useful for much more than song identification. You can build a model for an artist or genre and determine how to classify the song. One of my focuses in the paper is unsupervised genre classification -- my tests indicated some fairly reasonable groupings. This technique could be used for music recommendation -- "You like Dropkick Murphys? Well, they sound like Flogging Molly, so you might want to check them out."

Like the article mentions the concept of finger printing music using a beats and tones method is nothing new.

The concept of using it to enforce p2p transactions quickly falls down though. It is obviously impractical to design any sort of p2p system that would require the content to be uploaded to a central server for authentication and beat matching. Any system that relied on client trust for the content matching could be easily circumvented, and would be by a community that has been built on the desire to pirate this music in the first place.

Why someone would want to use an application like kazaa to buy thier music in the first place is beyond me anyway. Who wants to pay a record company for a song that's been encoded by an unknown party, often at lower bitrates and without quality encoding software. And to be sure, people do not use current p2p systems because they find them to be the best designed, or best UI, or easiest. They do it to pirate the music.

The future of pay for download systems rests firmly in the professionally produced content, delivered from server to client such as iTunes and Rhapsody. You can always include peer bandwidth augmentation of official downloads, though i would expect thats not really nescesary.

Trying to monetize p2p is a red herring.

I recall that in its dying days Napster was talking about adding this to appease the recording industry. The variation then was from a company called Relatable. Sounds like Shawn is stuck in a recursive loop.

That's right. TRM is a product of relatable. Last I checked, they give away an open-source program that generates a fingerprint, and have a private database that maps the fingerprint to actual songs. Napster licensed their technology in 2001 to identify illegally shared songs, as mentioned here and here.

Do we really want to help them build a bigger database?

That's right. TRM is a product of relatable. Last I checked, they give away an open-source program that generates a fingerprint, and have a private database that maps the fingerprint to actual songs. Napster licensed their technology in 2001 to identify illegally shared songs, as mentioned here and here.

Do we really want to help them build a bigger database?

That's right. TRM is a product of relatable. Last I checked, they give away an open-source program that generates a fingerprint, and have a private database that maps the fingerprint to actual songs. Napster licensed their technology in 2001 to identify illegally shared songs, as mentioned here and here.

Do we really want to help them build a bigger database?

Well, there's always Relatable and their audio fingerprinting technology. Given an MP3, you can get a key to determine artist, album, etc. Kind of like what Napster did in 2001. MusicBrainz uses Relatable for their community music database.

While i believe there is/was at least one startup that was working to match music using a beats & tone analysis method that could match to songs that had been shifted or obscured in some way

That was Relatable.

i'm not sure this technology has ever been in real use.

Napster 10.x used it. MusicBrainz uses it.

11,000 albums heavily compressed to 160kbps still takes approximately 600gb

Relatable claims that its tech can identify songs down to 16 kbps.

I do not think there is any software that could scan an Ogg Vorbis file and determine at all what song it is.

You think wrong. Relatable offers audio fingerprinting technology that creates a hash of an audio clip, which is useful for determining what recording it belongs to. Apparently, Napster was thinking of using it until the service was shut down and converted to a completely opt-in system.

And yes, it may get confused sometimes on cover songs, but that's the whole point: under copyright law, a songwriter is entitled to a royalty for every sale of a CD containing his or her copyrighted work.

I think that collaborative filtering is a much better solution than a recommendation system, although only time will tell. The advantage of a collaborative filtering system is that it can be passive, not requiring any explicit input from users. The software can just examine your playlist, and (anonymously) upload the information to a server. (perhaps a username and password so you can identify the same user repeatedly, but no way to tie the username back to anything else)

If you want to get fancy, you could even hook a Gnutella client up to it, and have a virtual custom radio that downloads and plays music that it thinks you'll like. (except, of course, that would be illegal ;-))

I think the FreeAmp project is a great one; it's a cross-platform, GPLed, music player that even supports Ogg Vorbis.

It just needs a little love.

To make a smooth, free collaborative filtering system, we really need a free software implementation of music fingerprinting software, along with an open, non-profit database of songs. MusicBrainz is headed in that direction, but, they to, are tethered to Relatable's technology.

Does anyone want to step forward to work on music fingerprinting software, who is interested in using it for the good of consumers, without catering to the recording industry?

How about a collaborative filtering database for music? If you were willing to settle for per-CD resolution, it'd be pretty straightforward to add this technology to FreeCDDB.

-- Agthorr

What does all of this have to do with napster and you? Well, freeamp allows you to download/stream music from emusic fairly easily (for a fee--something like $10 a month). So, if the napster distribution channel dries up, they become a quite attractive alternative. No more crappy searches, no more little red dots beside the songs, linux integration, artist-tipping support. Now, I'm not saying that emusic's actions here are good or bad, but do have a legal approach to digital music, while napster/gnutella/etc are questionable at the very least. They do support an open source project as well.

or use yet more vapourware - a tool to analyse a song and guess which song it actually is

That's what it does. It uses Relatable's TRM open source "audio signaturing" technology. See my post below for a link to e-music's teleconference about the system for much more info.

Also, see Relatable's page on TRM - also, check out the TRM-enabled freeamp.

Also, relatable's technology, which seems similar, though I hesitate to jump to any conclusions, is format-ambivalent. It supports any audio format; thus using ogg or wav or something other than mp3 won't defeat it.

Relatable does something similar to this already.

If you're using the stunning, portable, MP3 player FreeAmp there is the option of "signaturing" all the MP3's in your playlist.

The intention is that then you can do automatic lookup of them on the Relatable site - and I suppose perform automatic MP3 info tagging.

Its designed to completely supercede CDDB by providing a free and open database for information without the cumbersome album model and cumbersome license agreements. Its being sponsored by Emusic.com and Relatable.com.

The project leader is Robert Kaye. He's currently at Burning Man so he's not here to defend himself.

Jacob Everist
jeverist@fairtunes.com