A music recommendation engine that 'works by listening to the music' ! Tell me it isn't so !
Any real detail ? No... ?
Boffins in California say they have produced music-recommendation software which produces playlists "as good as" those from Apple's iTunes Genius - and which has the advantages of collecting no user data and having in its repertoire a lot of music that "Genius knows nothing about". iTunes' thoughts on what a Shadows fan might …
A music recommendation engine that 'works by listening to the music' ! Tell me it isn't so !
Any real detail ? No... ?
All these systems are, quite frankly, doomed to failure for one simple reason. Enjoyment of music is not for most of us a rational choice. Two songs can be very similar and you can like one and dislike the other. Hell, there could even be two similar versions of the same song where you like only one of them. The reasons for this are manifold. It could be that you don't like the timbre of a particular singer's voice or a particular instrument. Maybe you don't like a particular musician's phrasing. Or less subtly it could be bias against a certain artist.
Look at the Susan Boyle cover of Wild Horses. I don't particularly like it because this version has absolutely no character, it's as bland as if it were produced by a computer capable of reading sheet music. Each of the members of the Stones, put their own particular interpretation and inflection onto the music, and then their interaction as a whole coloured that to end up with a very simple song with an awful lot of character. There are lots of cover versions out there. Some of them are facsimiles of the original and fall a little flat, others have been filled with the cover artists' own character and mostly work very well. This latest version has no character or interpretation at all it's just one note after another.
Can this new system deal with all of these subtleties? I doubt it. Can it deal with personal bias for and against particular acts? Not without storing your personal preferences it can't. That last is very, very important. Say you happen to be a fan of a particular artist who sings soul and they release a song which is country and western. You are much more likely to give that a quick listen than you are to listen to Billy Ray Cyrus. Nothing in analysing the accoustics of your whole record collection would be likely to suggest you might like country and western. I know people who claim only to listen to one genre who actually listen to many others, simply because their preferred artists are more open minded than their fans.
And by the way I quite like some of the Shadows output, but I can't think of a single Herman's Hermits song I've ever enjoyed.
Errr, isn't that what Pandora does ?
Paris - come on, who doesn't want to open Pandora's box!
Signals guys do some very neat stuff, but the math was a bit over my head. Anyway, this sounds very impressive-- it'd be cool to see a slightly more technical description of the techniques they're using for the analysis-- and I hope this comes out soon so I can play with it :-).
I can finally toss that FM radio and escape any lingering chances of hearing anything that might challenge my narrow tastes? Hoorah!
Not much cop at lyric-driven songs then.
Horses for courses etc.
At least it's better than Apple's first attempt, "Just For You (beta)" which said:
“You bought The Shadows, Live at the Paris Olympia. We recommend Basement Jaxx, Kish Kash.”
What you obviously don't realise is that this was another example of Apple pushing the envelope in an innovative and unconventional way. Where their more pedestrian competitors might offer albums similar to the ones you already own, Apple’s groundbreaking system leads customers in a much more creative and original direction, encouraging them to widen their aural horizons by offering albums from genres entirely unlike the music they’ve bought so far.
Pandora's been doing this for quite a while... wish they'd had the funds to open it with such a fanfare though!
I don't see that the recommendation is so eccentric. The Genius approach (as far as I understand it) is essentially a "people who bought X also bought Y" system. If lots of people who bought the Shadows also bought Herman's Hermits then it seems a reasonable recommendation.
The Genius approach builds in a lot of non-musical stuff into its recommendations (e.g. tracks that were around at the same time but aren't musically related, etc) just through this approach. To be honest, it seems to work quite well for me.
It would seem that the "listening to the music" approach would yield a different type of answer, but might end up recommending things which are musically similar but which people wouldn't buy both of. As an example, if someone listened to a lot of reggae it might recommend UB40 due to the similarity of the music, but this would not be a good recommendation for a lot of people.
Perhaps the ideal solution is a combination of the two.
Indeed, which is why the "real world" analysis that Genius apparently uses seems to be the better choice. Effectively it's using Wisdom Of Crowds (though not to predict Lottery results as that imbecile fobbed onto us) for musical tastes. i.e. majority of Band X fans also have an extensive Band Y collection, suggest Band X to Band Y fans and vice versa. Obviously this will favour mainstream bands though.
It's not perfect, but nobody really understands why people like different kinds of music. It's an extension of Amazon et al with they're "other customers also bought". It could be skewed by a multitude of things. Buy a Hannah Montana album for a neice from Amazon and be bewildered by the new recommendations they make you.
"Apple’s groundbreaking system leads customers in a much more creative and original direction, encouraging them to widen their aural horizons by offering albums from genres entirely unlike the music they’ve bought so far."
MAYBE... but... the Apple system is based upon prodominently commercial/mainstream music with a handful of non-commercial/mainstream music thrown in. the recommendations will not be broad enough because it wont recognise the non-commercial music within a users collection. maybe im wrong, i understand that it "analyses" music but it cant recommend music it doesnt know (ie most GOOD music out there)
i dont and wont EVER use iTunes because firstly, it didnt work on Vista (windows problem no doubt im sure) but mainly because im bored of the apple popularity contest whereby people now only buy their equipment because its "in" or "trendy". their computers are brilliant, the ipod is what it is (ie a media player) and the apple business now seems to built on the idea that you should own their technology because its "new/innovative/quirky". in fact most of what they produce has and is being done by many other companies for lower prices without the inherent limits on software/OS compatability. add in that the only way they can seem to pry a bit of market share from microsoft is by pointing out flaws in windows (as if mac OS is flaw-free...not!) and they lose me as a potential customer.
anyway, i digress. iTunes does not offer me the music i want either through my looking (on friends computers) or via recommendations. it simply doesnt have enough non-commercial variety and thats sad.
Hopefully these guys will produce something similar that will analyse your music collection outside of a music player or within WMP (yes i use it and it serves me well!)
oh and for all of you out there who want something that will recommend music based upon what you're listening to (although you cant download through it) use LAST.FM. it links to media players and gives you info on the artist and similar artist recommendations.
"However he and his colleagues consider that Apple's Genius isn't smart enough, and in any case can't recommend music which isn't on iTunes."
So this new one is going to tell me to listen to music I don't actually have?
Consider the following three pieces of music:
1, Mars, God of War from the Planet Suite
2, Are you Losing your Mind? by Hawkwind
3, Am I Evil? by Diamond Head.
They all have the same chord progression - they're all quite different though. Will the software be smart enough to work out that Diamond Head fans might not want to listen to classical music or Hawkwind songs? I doubt it.
I suspect that this is another piece of software with all the appeal of Wolfram Alpha!
It has to be said...
Who are "Herman's Hermits"?
Come to think of it, who are "The Shadows"?
What's that? I've arrived in the wrong era? Damn space-time continuum thingy-whatsit, I was aiming for 2009.
Isn't this what Pandora have been doing for quite some time?
I like The Shadows. I like Herman's Hermits, too. Suck on that.
... and all your arguments are own goals.
You suggest that liking one version of a song will make the software suggest interminable covers by others. That is simply ludicrous.
Exactly what you say --- timbre, rhythm, and all that --- are things to capture in numbers, and hence compare across songs. For example, my dislike of pompous pseudo-operatic outbursts, so popular in pop by tv-show-selected nobodies and divas on the way down, cannot be hard to quantify into a strong penalty. Bring it on!
What you describe cannot be done.
I believe those students did a good job given what they wanted to do.
It's all about statistics and probabilities.
If you liked "Song X" by Artist X, there are good chances you'll like the cover by Artist Y, because it uses the same instruments, has the same rythm, speed, whatever they use to categorise songs. You may not like it because Artist Y sucks in general, but the software would be right to propose it because given your taste, you "technically" should like it.
Mind you, I for one use neither.
The one thing I'd love to hear is good stuff by relative unknowns. My local college and public radio stations have helped a few times, but 90% of the "cool new stupfh" from them is really some (often quite bizarre or banal) tripe that the DJ likes.
This sounds promising, and would make the Web a great new distro.
Which is why it would probably be squashed flat by the Big Boys. :(
Liking The Shadows means you might like Herman's Hermits, does it?
Sounds reasonable, as both are so archaic they might as well be playing lutes. Oh, verily, sirrah.
Whenever I see mine i go back in my cave for three months.
'The one thing I'd love to hear is good stuff by relative unknowns.'
Get yourself onto Last.fm then - it's great for that.
to recommend me the music I like. Reason it's obvious, the preference for a piece of music has nothing to do with statistical analysis of frequencies or timings. Piece of cake to prove it : take a song and have a representative sample of people listening it. Some of them might like it, some of them will hate it, however the results of the statistical analysis is the same. Now if the majority of those participants in this exercise will like it, this will force a correlation between the results of the sound analysis pushing the conclusion that the rest of the people must like it too.
Do not forget that the beauty is in the eyes (or shall we say ears) of the beholder. It's the same with every other preferences.
Also one more question, how can this software tell if I really liked the music I've just listened ? Maybe after listening to it I realized it's not worth pirat.. sorry, buying
What have the FUCKING HIVES got to do with The Ramones? Every damn place I've bought Ramones music from online, their smart-arsed recommendation system offers me Swedish bloody fake punk kiddie rock hogshit fucking bollocks. I wouldn't mind if I occasionally they offered me the first Vibrators album, but JESUS H. CHRIST ALMIGHTY, not the FUCKING HIVES. OK? I hate these idiot systems. But I do like the Ramones.
iTunes determined that because your sad enough to like massively non-relevant outdated naff guitar riff driven cowboys 'n injuns western sounding music by an ancient band called "The Shadows", your very likely to be sad enough to enjoy that blisteringly gay old-timers band, "hermans fuckwits"
... I'm henry the 8th I am, henry the 8th I am I am...
... second verse, as shit as the first ...
I haven't tried it, but I have heard of it. The only issue is that my music listening time is restricted. I'm best off doing "podcast" type stuff, as opposed to streaming. I used to listen to RantRadio while I worked, but the music made me too happy ;).
I like the idea of independents putting their music on the Web, having a "smart aggregator" find it for me, and cut out the coke-addled jerks at Universal/Sony.
I have had a good time with the genius thing. I was playing a track and it recommended an album by Seal. I laughed and said I would never buy a Seal album. I visited a friend and he happened to be playing the recommended album. Bugger me, it was excellent. So I purchased it.
@ AC 14:23 GMT
I have the Planets and a lot of Hawkwind in my collection, the friend mentioned earlier has Hawkwind and Diamond Head in his collection.
where there are multiple recording of the same work - the Westminster recording of Tallis' Spem in Allium knocks my socks off, but the version by Cantillation is 'meh', but Cantillation's Ave Verum still gives me goosebumps when I hear it.
My iTunes collection is almost all old school punk. I pick almost any song in my library, and the Genius can never find any recommendation. Oddly enough, on the rare occasions it can find a match, it always recommends the same Ramones song, "Somebody Put Something In My Drink." I think the Genius had something slipped into its drink.
Anybody ever thought the reason iTunes on MS blows is that there is probably a thick gooey nasty off-centre shim of cocoa/carbon/whatnot to Win32 API. I am thinking if that is the case then a lot of Mac stuff could possibly run on Windows (badly?).
I think its good that there are various methods for this, the final arbiter of course being your taste.
On what planet is it a bad, or even unexpected thing, that a music recommendation service does not recommend stuff that is is unable to supply?
How stupid would it be if it recommended tracks and then told you that it didn't have them?
Also really disappointed that the article did not even mention lyrics, or the fact that I for one, prefer cover versions that are different from the original - and usually can't be bothered with ones that sound really similar to the original - this system seems to be designed to recommend the former, and ignore those that have some originality.
I'm a broad minded 27 year old
I like Trance, classical and The Shadows, Dire Straits, Johnny Cash, Lenny Kravitz and Jimi Hendrix
Leave these good artists alone and label me I dare you.
"What you describe cannot be done."
Which was exactly my point. It can't be done so don't bother trying.
It's also pointless. One of the important things about music choice for people who actually make a choice* is that they have found the music themselves, or rather that they think they have found the music themselves. Whether they came across it on the radio, heard them as a support band or were given a personal recommendation it is important that people think it was entirely their own choice. As such an impersonal recommendation by a computer system simply won't cut it.
* So that would be don't watch X Factor then.