Slashdot Mirror
Pandora Radio from Music Genome Project
kramthegram writes "The Music Genome Project, an attempt to define music by it's traits in a way similar to DNA defines traits in humans has led to the development of Pandora. Pandora uses the song choices you make to see what traits appeal to you and present you with custom radio station. While limiting you to thumbs up or thumbs down, the "gene" heuristics allows for a very quick adaptation to your musical tastes." Not sure how deep it goes, and I'm not sure I like that it led me from The Who to Styx and Def Leppard. But this is a neat little tool for discovering new music.
http://slashdot.org/article.pl?sid=05/10/07/173021 5&tid=141&tid=187
At least this one took over a month.
...Rob
The American Dream isn't an SUV and a house in the suburbs; it's Don't Tread On Me.
How useful will it really be? Sure, I like Punk, Alternative, and Metal. But the different bands have diffrents styles within the genra and I may love one band, but hate another similar one.
Someone save me from this sanity.
Is it anything like Last.FM, or does it run independant of other users? If it runs independant of other users, I'd say Last.FM would win in that category, because it's showing you what other people that listen to the same music that you do like.
I think Last.FM and this have the same aim, recommending music you might like, but I think Last.FM pulled it off better.
...Six Degrees of Kevin Bacon.
Actually, it seems like an interesting idea. We all have libraries of CDs based on our likes and I suspect if the libraries were analyzed we'd find slighlty deeper relations between the disparate music we collect. I've got a very eclectic collection of music and I'd be hard pressed to see the link between Reba McIntyre, Pink Floyd, and David Sanborn, but maybe there is one.
Of course some conspiracy theorist is going to use this to determine that the music industry is actually selling the same 5 songs over and over again, just in different keys and rhythms. Because we all know it's true.
GetOuttaMySpace - The Anti-Social Network
The best resutls come from submitting a song that you like. Using the Artist will most likely get you going down the wrong path. This is due to not all songs sounding similar from one artist. I have had the best results by putting in 2 or 3 songs that are similar to my ear that like. If you can't think of any songs by an Artist, Google it first. Also, there is a "rating" system. If you like the song, then give it a Thumbs Up, if you don't, give it a Thumbs Down. This will help your station learn what kind of music you are actually looking for. Rate the songs for better results.
I've been loving Pandora for about a week now. Just this morning I thought, "Hmmm... maybe I should try to get this posted to Slashdot. I'll bet a lot of the Slashdot crowd would dig this," but then I thought again to myself. I said, "Self, why would you want to slashdot their server and rob yourself of this little jewel?"
If you dirty buggers bring down this server... so help me steve...
I do like the site, unfortunately though after around 3 hours of using it, it stopped giving me new songs that I liked; it just played song's I already said I'd liked, or songs I didn't like. One interesting thing is that is uses basic mp3 files for the music, so it's actually not too hard to download the mp3's directly from the server if you log the right packets.
Pity they'll be putting ads on it (soon).
At least this one took over a month.
That's because it was on Fark.com (yesterday? the day before?), so the people that like to copy links from one site to the other thought it was new.
I want to drag this out as long as possible. Bring me my protractor.
... I've given up. The Pandora player insists on using Flash local storage, which I had disabled. Now, no matter what I do with the local storage settings, Pandora just keeps telling me I need to enable Flash local storage. Following their instructions doesn't help.
Too bad.
Sean
A very similar concept that actually works is www.music-map.com. This engine takes the input of all users into account and really let's you discover new artists from the genres you like.
Comment removed based on user account deletion
I once bought a highly recommended cd from Amazon, and it was the worst piece of noise I have ever heard (I won't mention the artist in question). But almost all the reviews were five starts and glowing. Finding new interesting artists which match your taste in music is a hard task. Could a classification system help to make suggestions?
I find Last.FM a better method to find new music. Granted, the new Player REALLY SUCKS, I used it before they started the new player thing. There is a Proxy being developed which restores the old functionallity more or less.
Oh, that and the Pandora music project is not free:
Q: How much does it cost?
Pandora is available in two forms. Both versions have exactly the same features.
The first form is an advertising-supported version which is entirely free. Over time we'll be incorporating ads into this version of Pandora.
For those who want to steer clear of advertising, subscriptions are available in two different flavors:
ANNUAL: 12 months of unlimited use for $36
QUARTERLY: 3 months of unlimited use for $12
while the last.fm is free unless you want a "personal" radio.
Ubuntu is an African word meaning 'I can't configure Debian'
The site is not responding (big surprise). But from the description it sounds like it is similar to Music-Map.
and now that I know C a little, maybe I'll try out making a plugin or something..
/right now/.
/would/ go well. (various probability weighting schemes, decreased weight as we move on, requiring much use before it really knows you, blah blah blah...)
I have lots of MP3s. I like most of them. However, I'm not always in the mood for all of them. There is very little music I've dismissed completely as bad, so "Thumbs up" || "Thumbs down" is pretty lame (,stupid, closed-minded, moronic, a horrible basis for anything, encouraging of the already prevailent general-dumbness of people whose music I tend not to be in the mood for, etc)
What I've wanted is a system by which music can be automatically catagorized based not on whether or not I like it, but rather based on whether or not I'm likely to enjoy it
How this would work: Start with the standard "Shuffle", picking at random any song. Then, if I hit "next" right after a song starts, decide "This song doesnt go well with this other song right now", and instead try selecting one which my lack of hitting "next" in the past has indicated
The closest I've seen has been plugins which weight the shuffle based on a rating you choose, which doesnt ever fluxuate.
Point: Playlists should be quaint by now. Why should I need to choose in advance what I'm in the mood to listen to an hour from now?
-- 'The' Lord and Master Bitman On High, Master Of All
Sue them!!
-- RIAA
(the funny/scary part is that it's not far fetched to me that they actually will, for being too accurate in handing out music a user wants to listen to)
Beware: In C++, your friends can see your privates!
I have been a happy (and donating) user of Last FM formerly Audioscrobbler. They do a really good job of matching up music tastes and their radio program is good. They also have a plugin that allows you to submit what you play from most major audio players so you can track what you listen to and compare with others. They have full tagging capabilities and extensive forums as well as music 'groups' of like minded appreciators. I have been very impressed and I admit I haven't played with Pandora much but it doesn't seem too much better/different.
As someone who has read enough books on music to know what they are talking about when they say 'We chose this song because it features x and y', I can honestly say this doesn't mean much. I think genre and statistical comparisons between users has a far greater impact on what songs it chooses than what they suggest in their FAQ/info pages.
The whole idea of analyzing a song for different qualities is great, but it really doesn't get you very far with something like this. I can think of a million songs with 'Mild rhythmatic syncopation' and 'Major key tonality' (just an example of the reasons it told me it was playing a song), and I would probably only like a small portion of them. I suspect that the genre of my song (eg 'Hard rock roots' or 'punk roots' etc) is the biggest deciding factor in what it plays -- not the actual style of the song.
I am skeptical of any algorithm that purports to gauge or classify taste. People listen to music for complicated reasons and they often listen to very different genres.
A better solution is to point people to "taste-makers".
I found by illegally downloading music using limewire, that I could find very interesting new music by simply broswing the collections of the people that were downloading from me. That really opened up my horizons as far as taste is concerned. I don't think an algorithm could come close to that.
You know, most of the media world is pretty excited about these concepts of "personalized media"... where the media that is presented to you is based on the types of things that you already like (it's just starting to take off in music, but watch for it in the future in television, movies, and internet sites). But I see this as somewhat of a problem, where people are never exposed to new things. If everything in our world is personalized and created specifically for our tastes, how do we define our tastes? When do we ever get a chance to listen to something we don't like, and say that we don't like it? Or listen to new things we've never heard of, and that may not be in any way related to our database of media we like, and say we like it?
The situation presents us with two possibilities: either we get pidgeonholed into a "genre" artificially created by the content distributors (as broad or narrow as that genre might be), or our tastes enter a feedback loop, where the only things we listen to are the things our personalized media players play for us, whose choices are based on things we listen to in our personalized media players.
So where do we get outside input? My suggestion at this point would be to do away with artificial genres and create relationships between media based purely on a database of what people like and don't like. (Last.FM does this now.) Then I would like to see the media player throw in a randomly chosen selection once in a while, just to test its own theory, so to speak. However, for that to work, the selection would have to be truly random; no fair throwing in something that you are marketing heavily (I'm talking to you, [RI|MP]AA...) just to get people to hear it. So instead of choosing music based only on your tastes, your media player will choose music based *mostly* on your tastes, and then throw you a curve ball once in a while to see how you react. Who knows? Maybe that diehard punk fan would enjoy a Beethoven piece or a 70's pop song. But the media player would never know that unless it tried.
For security, the MD5 hash of this message and sig is 09f911029d74e35bd84156c5635688c0.
Basically what you are saying is that they are lying about how they choose songs. But why would they do that? If they were using social networking or grouping by genre, and it worked, why wouldn't they say so?
Furthermore I haven't seen anything that would lead me to think that they are grouping things like you claim. Try typing in a band like Ween. You won't get anything that is remotely related by genre or popular tastes at all.
One of the biggest things that I like about this site is that it does play different artists than I find with other systems that determine thier suggestions by genre or social linking (people who liked A liked B), and I have liked many of them.
I can think of a million songs with 'Mild rhythmatic syncopation' and 'Major key tonality' (just an example of the reasons it told me it was playing a song), and I would probably only like a small portion of them.
They are rating on hundreds of different factors. The fact that two songs are related by just one of those factors would not cause it to be played. It is the fact that it is related on a large number of those factors. And it isn't surprising that music within a genre share many traits with each other.
I suspect that the genre of my song (eg 'Hard rock roots' or 'punk roots' etc) is the biggest deciding factor in what it plays -- not the actual style of the song.
Again I don't see any reason to think this. I have found it rare for it to only play music within a specific genre, and to the extent that it did, the songs were all musically simular.
The Pandora app was written using OpenLaszlo ( http://www.openlaszlo.org/ ), a free software rich internet application development platform. Why do you suppose they did that instead of using Macromedia's tools and runtime?
Also, the personalization algorithms don't even really represent your own tastes all that accurately, at least, not until they've built up a considerable database about you.
For example, at one point in time or another, I bought some Star Wars-related product from Amazon.com. This was years ago. But to this day, every time I go to Amazon.com, they are recommending me the latest Star Wars novel or toy or DVD bonus package or what-have-you. Just what is it about my buying habits that makes them think I like Star Wars that much?
Not to mention the fact that my Amazon.com purchasing habits don't necessarily represent my purchasing habits as a whole. It's funny; I probably buy a ton more books than I do Star Wars DVDs. The thing is, I don't really like to order my books through the Web. I prefer to wander down to my local independent bookstore on my lunch break, thumb through the pages a bit, smell the paper, and then proceed to buy the books at the checkout counter using my 10 percent discount card.
As a result, Amazon, which by all rights should know exactly what my purchasing habits are, doesn't actually have the slightest idea.
Likewise, my iTunes (or Windows Media Player or whatever) probably doesn't have the absolute best idea of what music I listen to. Yes, the music I play on iTunes is probably music that I legitimately like. But it's also music that I downloaded. I do legitimately rip my own CDs to MP3 format, but I do that for my portable player, not to listen to at home when I have the original disc sitting right there. So if you were using iTunes to judge my preferences, you'd only really know about the music that I think is kinda catchy, but either I don't feel like paying for or else is too obscure for me to be able to track down and buy. You wouldn't know about any of the stuff that I liked enough to drop $18 on.
Breakfast served all day!
you should try IMMS, I think it does exactly what you want.
it has no interface other than the player's next/prev and playlist, and is fairly easy to port in case your player isn't supported
(there's only a small plugin that needs porting, currently supports XMMS and BMP)
For those using unix, the files are cached to /tmp/plugtmp/access*
If you wanted to copy them to say... ~/pandora you could then make that your working directory and:
#!/bin/sh
for i in `ls`; do mv $i $i".mp3"; done
which will set your extensions... quick and dirty, but hey we're not keeping these songs, r-i-g-h-t-?
Be who you are and say what you feel, because those who mind don't matter and those who matter don't mind. - Dr. Seuss