We’ve been head down here at the Echo Nest putting the finishing touches on what I think is a game changer for music discovery. For years, music recommendation companies have been trying to get collaborative filtering technologies to work. These CF systems work pretty well, but sooner or later, you’ll get a bad recommendation. There are just too many ways for a CF recommender to fail. Here at the ‘nest we’ve decided to take a completely different approach. Instead of recommending music based on the wisdom of the crowds or based upon what your friends are listening to, we are going to recommend music just based on whether or not the music is good. This is such an obvious idea - recommend music that is good, and don’t recommend music that is bad – that it is a puzzle as to why this approach hasn’t been taken before. Of course deciding which music is good and which music is bad can be problematic. But the scientists here at The Echo Nest have spent years building machine learning technologies so that we can essentially reproduce the thought process of a Pitchfork music critic. Think of this technology as Pitchfork-in-a-box.
Our implementation is quite simple. We’ve added a single API method ‘get_goodness’ to our set of developer offerings. You give this method an Echo Nest artist ID (that you can obtain via an artist search call) and get_goodness returns a number between zero and one that indicates how good or bad the artist is. Here’s an example call for radiohead:
http://developer.echonest.com/api/get_goodness?api_key=EHY4JJEGIOFA1RCJP&id=music://id.echonest.com/~/AR/ARH6W4X1187B99274F&version=3
The results are:
<response version="3">
<status>
<code>0</code>
<message>Success</message>
</status>
<query>
<parameter name="api_key">EHY4JJEGIOFA1RCJP</parameter>
<parameter name="id">music://id.echonest.com/~/AR/ARH6W4X1187B99274F</parameter>
</query>
<artist>
<name>Radiohead</name>
<id>music://id.echonest.com/~/AR/ARH6W4X1187B99274F</id>
<goodness>0.47</goodness>
<instant_critic>More enjoyable than Kanye bitching.</instant_critic>
</artist>
</response>
We also include in the response, a text string that indicates how you should feel about this artist. This is just the tip of the iceberg for our forthcoming automatic music review technology that will generate blog entries, amazon reviews, wikipedia descriptions and press releases automatically, just based upon the audio.
We’ve made a web demo of this technology that will allow you try out the goodness API. Check it out at: demo.echonest.com.
We’ve had lots of late nights in the last few weeks, but now that this baby is launched, time to celebrate (briefly) and then on to the next killer music tech!
April 1, 2009 at 12:46 pm |
I see Coldplay got a 0.02 on the goodness scale! Was this mandated by Brian? But DMB got a 0.44?? Vanilla Ice is 0.73?
I don’t understand how an artist can have a “goodness” score? It depends a lot on the person, no?
April 1, 2009 at 12:59 pm |
@dale – there is a clear objective criteria for music goodness. This is science. Don’t argue.
April 1, 2009 at 1:03 pm |
Seriously. Thank God somebody thought of this.
Additionally, if you guys just want to send me an e-mail, I’ll tell you whether or not something is good.
Not as robust as an API, but I’m up pretty late and check my e-mail pretty often.
It would be quick too. If you say “Is Neko Case good?” I’ll say “Hell yeah” but if you ask “Is Deerhoof good?” I’d say “Not many sane people like them, but you could give it a shot.”
April 1, 2009 at 1:06 pm |
@zac = thanks for the offer. Is the new Decemberists album any good?
April 1, 2009 at 1:07 pm |
This is clearly broken. I tried “Loof Lirpa”, which everyone will agree is the best band anywhere, ever, and not only did it not award it a perfect 1.0 but it didn’t even know the band !
April 1, 2009 at 1:10 pm |
@bethor – if we don’t consider the output of an artist to be music, we don’t add the artist to our database. Consider a result of ‘don’t know about this artist’ to mean that we don’t consider what that artists produces to actually be music.
April 1, 2009 at 1:13 pm |
@Paul “Not many sane people like them, but you could give it a shot.”
(Sort of just kidding. I liked their other stuff but this one is *really* hard to get through)
April 1, 2009 at 1:16 pm |
Ok, I’ll admit it (shamefully) : due to the limitations inherent in textual communication, I have no idea if you noticed I was pulling your leg in response to this (very funny !) announcement and pulled right back, or if my attempt at humor was a complete failure :D
(if its the latter : the artist Loof Lirpa would be a perfect candidate for ‘back-masking’, so to speak.)
Either way, of all the amazing announcements I’ve read today on various blogs, this was definitely the best ;)
April 1, 2009 at 2:24 pm |
[...] looks like the guys over at the Echo Nest have added a new metric, called Goodness, to their developer [...]
April 1, 2009 at 2:37 pm |
recommend music just based on whether or not the music is good. This is such an obvious idea – recommend music that is good, and don’t recommend music that is bad – that it is a puzzle as to why this approach hasn’t been taken before.
C’mon, Paul.. I’ve been saying this to you for years :-)
April 1, 2009 at 2:48 pm |
@jeremy – I know you have. And I was thinking of you when I wrote that. I was going to cite you, but then I thought that if I didn’t, I’d be sure to get atleast one comment from a reader ;)
April 1, 2009 at 2:50 pm |
[...] via Killer music technology « Music Machinery. [...]
April 1, 2009 at 2:56 pm |
Oh, wait, no.. I’ve just been April Fooled, haven’t I?
D’oh, and double D’oh!
My fault for not reading the comments first, eh? [wipes egg from face]
I still stand by my desire, foolish or not, to have music recommended to me not based on whether it is similar to music I already like, or similar to music my friends like, but whether it really is.. good.
Sandra Uitgenbogert had a very early Music IR paper.. I think it was published at ACM Multimedia in 1998.. in which she used a number of heuristics to estimate the “interestingness” of music. Things like repetition, entropy, etc.
April 1, 2009 at 3:00 pm |
Paul, I was hoping for a complete description on the theory of goodness! I thought this was my chance to finally understand the inner workings of Brian’s musical tastes.
Now that goodness is defined, the music mixes running at the Nest must be completely conflict-free!
April 1, 2009 at 5:06 pm |
[...] Killer music technology We’ve been head down here at the Echo Nest putting the finishing touches on what I think is a game changer for music discovery. For years, music recommendation companies have been trying to get collaborative filtering technologies to work. These CF systems work pretty well, but sooner or later, you’ll get a bad recommendation. There are just too many ways for a CF recommender to fail. Here at the ‘nest we’ve decided to take a completely different approach. Instead of recommending music bas [...]
April 1, 2009 at 10:29 pm |
[...] more from the original source: Killer music technology « Music Machinery :artist, concert, goodness, hacking, music, music-machinery, research, startup, Technology, [...]
April 2, 2009 at 5:54 am |
At least music criticism is more exciting, such as having more options to gauge whether a particualr song is good or bad.
April 2, 2009 at 6:41 am |
So good it was bad.
Suitably Suckered (until Nitzer Ebb gave me “This one makes me cry every time”)
April 2, 2009 at 7:31 am |
What did you use for artist names? Because my band came up and some of my friends’ bands came up, but others’ didn’t.
April 2, 2009 at 9:08 am |
We crawl the web for artist and band names. A good way to ensure that your band gets coverage is to make sure that it has an entry at MusicBrainz.org
April 3, 2009 at 3:01 am |
[...] Killer music technology We’ve been head down here at the Echo Nest putting the finishing touches on what I think is a game changer for music discovery. For years, music recommendation companies have been trying to get collaborative filtering technologies to work. These CF systems work pretty well, but sooner or later, you’ll get a bad recommendation. There are just too many ways for a CF recommender to fail. Here at the ‘nest we’ve decided to take a completely different approach. Instead of recommending music bas [...]
April 3, 2009 at 3:34 pm |
That was hilariously deadpan. Nicely done.
And Zac, you win the comments.
April 8, 2009 at 9:20 am |
Argggh- I hate going over items a week later in my reader and then realizing they were posted on april 1st. The april fools hangover affect always gets me.
April 9, 2009 at 5:59 pm |
“good” vs. “bad” = subjective… i think this is science the same way leggos are construction materials.
April 9, 2009 at 6:08 pm |
It is science but only on April 1. ;)