Slashdot Mirror
Audioscrobbler (Anyone Remember Firefly?)
asciirock writes "RJ, a University of Southampton grad student in the UK has just put his final year project online. Audioscrobbler is a free plug-in for Linux XMMS and Windows Winamp2. It tracks every tune you play, cross-references with others in the Audioscrobbler community and serves up recommendations. There's also msging, stats and user homepages. In other words... Firefly lives!"
Listen to *one* Britney Spears track out of curiosity, get distracted by something to do in another room, forget that it's playing repeatedly for 3 hours in the meanwhile, get labelled a teen music sheep by the system and get recommendations for more degrading music. Arg!
Recommendations are nice, but what I want is a tie in to Fast Track. I want a list of DATs that I can plug in to Kazaalite and download based on what I play.
Wow, I guess they can make really nice profiles of their users.. and then sent the customized advertisements. Nice ;-)
Can you imagine just how valuable the kind of information generated by a project like this might be to the RIAA? They somehow gain access to a server and now they know everything you're listening to
Argh, why's there no support for Winamp 3? Now i have to choose between having my Media Library, or goin old-skool, but bein able to use this cool plug-in
Everything sucks except musicandstuff
While it might seem cool at the first sight to produce such a tool which creates recommandations from other people's playlists, it's in fact counterproductive at pratical applications.
What we have here is a stabilizing feedback loop, songs often heard will be heard more often. This can be described by the following simple equation (h(t) - hear rate):
dh(t)/dt = h(t) * c + sin(h(t)) * phi(dt,H(t)) where the last term is a stochastic diffusion corrector which models connection drops etc. This means that after a 3c/pi annealing time new injected songs (c1,...,ck) have no chance to be heard at all, because the system reenforces to old songs. The only possibility to get something new into the playlists, is to get an external stimulation at e.g. t0: phi(c-h(t0). Such a high current can be only injected be a very strong source covering a large part of the system.
In simple words: after some iterations an equlibrium is reached and all new song turning up in the recommendations are the top 24 played at MTV.
In fact, you are just replaying the shitty MTV mainstream taste.
I can't think that this is very good, first you don't need a computer program to recommend the MTV top 30 when you have a TV and secondly you only get boring mainstream stuff and nothing like exciting french chansons or so.
Owner of a Mensa membership card.
What i would like is a window that lets me rate a song the first time i hear it, or just add a rating to the ID tag thing.
And also maybe keep track of the amount of times i play it..
That way it could find songs that i like, and i could have a category by which to order my songs when i can't decide what to listen to..
What I really don't get, and sounds a bit fishy to me, is the whole username/profiling thing.
I mean, this can be done without it:
Have an anonymous user handle on that site. No email, no nothing. (sure, they can have your IP. They can have mine, too, if they want, it's a dynamic one)
Whenever you hear a song, it sends the info: a user who heard (set of songs) decided to hear (new song), and of course heuristics of how much any song is heard, bla bla bla.
The server keeps this huge database. When you want recommendations (downloaded every 15 min? or something) your program asks what the database recommends for someone who listened to (the set of songs you listened to). You're not giving away an email, no personal info, just an anonymous username (created automatically, or something. There is alot of 'or something's here)
There's no real reason for the server to know who you are or what you like for this to work.
Perfect profiling is also not nessecary, in my view, but that's a different issue altogether
My other
I think that associations, in computing, is a great idea for user interface.
A program like this (lets disregard the Big Brother for one second, and look at computer+user alone) tells you what songs it thinks you'll like, based on what you've heard before.
It could also tell you what songs you'd like to hear NEXT, based on order of songs you had before, and make these easier to access on the playlist (like, on the recommendation list. I'm getting out of hand aren't I?)
The whole idea of associating user actions can be great. Suppose you work on a project. Slowly, the computer (the brand-new GPLed Associator program) associates a certain directory, where all the files are, with the files themselves, your favorite editor, the compiler for that language, and certain sites you visited researching for it.
via some UI, it'll make all these accessible when 'triggered' - when it is pretty sure you're working on the project right now, or going to.
In some sense (in a small amount of cases), the computer will be 'one step ahead of you' - holding the line when you're just about to ask it to call...
My other
Don't worry, they won't. The RIAA has no power in the UK and none over the government.
Part of the fun of being British these days is the RIAA can't bribe - sorry, fund - polititions in Westminister nearly as easily as in Washington.
This is my sig.
Correct. They are currently valid nowhere in Europe, although there are unfortunately plans to change that.
However, this doesn't prevent an unethical company to sneak just such a patent past a sleepy patent office clerk, and once it's on the books, they can bully whoever they want with it. True enough, eventually the judge will decide in favor of the defendant, but until that date the defendant has to cope with a number of hassles (lawyer's costs, and more importantly: injunction to force him to change his software, so as not to use the disputed features, etc.). Just let's hope nobody patents the light switch!
Say no to software patents.
Forget all the bogus attempts to move your antique business model online. What you want to do is license this technology from its creators, and build a mechanism to sell digital copies of the recommended tunes.
Imagine a dialog box comes up and says: Hey, people who like Weezer and Radiohead are also listening to Wilco. Want to download their latest single for 50 cents?
Combine that with some fair-use-friendly DRM software, and you've got THE application that gives the recording industry legs for the digital age.
Why not today?
Don't be embarrassed about it being crappy code, all code is crappy in the early stages. :) Put the GPL LICENSE file in the root directory, and follow its directions for adding notification to your source files. Then tar it up and call it 0.01.
Put it up and keep putting it up as you update it. If you think you might have security issues, best that you open the code now before your user base gets any bigger -- let people review it and send you suggestions. If you don't think you have security issues, you have no reason not to release it.
For a project that demands community participation, a promise of GPL code in the future is worthless. What's valuable is the code itself.
Licenses, releases, security feedback, other feedback... this is all part of doing a project like this. It's something that isn't normally taught in a university, but if you really want to run a project that depends on its community, this is not extra-credit, this is a prerequisite.
Just my opinion :)
is this by ID3 tags or by filenames or what? i have a great deal of digitized (ogg) LPs that are named "~/albums/Artist/AlbumTitle/xx - trackname". there's not even playlists, because i'm lazy and added xmms to open-with for directories in nautilus. how will audioscrobbler respond to me playing those tracks?