Mufin, a powerful music recommendation engine that actually works, has launched to the public. We last covered the site in early October, when it opened in a restricted private beta.
The site, which uses technology created by the same organization that developed the now-ubiquitous MP3 file format, uses an advanced algorithm to ‘listen to’ songs and identify similar sounding tracks based on over 40 characteristics. Such automated systems are very hard to pull off (which is why Pandora, another music recommendation engine, uses human experts), but in my testing Mufin had fared surprisingly well.
Since we last wrote about it, Mufin has introduced a Facebook widget that allows users to get song recommendations from within their friends’ Facebook profiles. The site has also relased a plugin for iTunes, which generates playlists based on tracks in a user’s library (iTunes Genius does exactly the same thing, but it relies on metadata rather than an analysis of the music file itself).
Really tough to tell how this works with only clips to play.
In this day and age with mp3 searches everywhere, there is no reason not to play full clips.
I came across this site the other day,
http://theperceptron.com/
and Mary (the sole founder/creator) is doing a great job on the technology, and she plays full tracks!
MusicIP.com is another one to look into in this space, though they are more like genius in that they build playlists based on your personal collection. They just launched a beta app for blackberry.
I came across the same site too, and i was impressed! Was gonna comment on it but i guess u beat me to it.
Oh btw
Check out http://www.jobstaxi.com
New Jobs. CauseForce. Strands. Carbine Studios. NCsoft.
@Yasser and @pedalpete I’m Mary, the author of http://theperceptron.com Thanks so much for the kind words about the site. Let me know if you have any suggestions for improvements.
Recommend me some international music please.
Well that really narrows it down doesn’t it. You’ve only asked for music from the entire world!
I agree, automated recommendation is probably awful hard to pull off. But someone’s done it before, namely Deezer, which I actually came across because it was mentioned on TechCrunch. When it can’t stream the songs you want, it offers something called SmartRadio, which plays similar songs. I was amazed at how well the feature works and have discovered quite a lot of good stuff I didn’t know about. (Actually, I don’t for sure if their system is automated, but given the sheer volume of their inventory, I’d say it’s got to be.)
I wasn’t impressed. Some of the results are laughable actually.
They need to be more transparent about why specific similar tracks are displayed. Maybe a visualization that illustrates the similarities, etc?
I tried three different artists/songs that I wanted more music like and not a one of them turned up any results. This isn’t exactly out-there stuff either (Ben Folds, Postal Service). Perhaps it’s just luck of the draw, but I won’t be coming back any time soon.
Oh I really really hope Microsoft buys these guys. With all you can eat music subscription on Zune and a killer recommendation engine powered with mixview, it would be an unbeatable combination.
I like it – it is not perfect but it is useful. I tried to trip it up with stuff like Rage Against the Machine, Black Sabbath, Lou Reed and yeah they had some weird matches but you can weed through the wrong stuff and find stuff you have not seen in a while.
The UI is super intuitive and you get enough of a snippet of the song to see if you like it. I really like the simplicity of it all – you don’t have to be really thinking about it and you can come across new music that fits into what you like. It is not hard for me to see myself wasting some time on it and then buying the “similar track” using their iTunes link.
Cheers
Eric
I wish they’d play full versions of songs rather than just previews, though that would probably be….illegal…for now at least
Found some good Daft Punk sound-a-likes
I’ve spent a bit of time with this, and I think it fails miserably. I don’t discover any new music through its suggestions, and most are off the mark anyway.
@Glenn Agree 100%, this is a giant FAIL, site is not good. What am I going to do with previews and second the few songs I searched it brought up the same song several times , or songs that were not very good.
Remember theFilter? What happened to those guys?
anyone remember the mongomusic guys? that stuff was by far the best algorithm based recommendation system ever. did not require any history or metadata. i wonder what happened to those people?
MongoMusic was acquired by Microsoft a long time ago.
I just tried their service on a few artists/songs and it seems to me, as others already pointed out, that, actually, it doesn’t work very well. Most of the queries turned out the same apologies: “Sorry, no tracks similar to * available”.
I am a last.fm user and I believe that, until we will see the first AI (which might be sooner than we think: http://ventureb...tive-computing/ ), this is as good as it gets in the matter of similar music (and other arts) discovery , we just have to rely on each other
Cheers!
Why do we “JUST” have to rely on each other? Undoubtedly some Computer Audition techniques (the science of audio understanding by machines) can be used to improve the meta-data and human recommendation system, it’s just a matter of how much and in what way.
With some sort of thumbs-up/thumbs-down system, they could add some machine learning techniques to improve the system over time. Anyways, don’t give up on the potential of a whole field of research just because this presentation of it doesn’t meet your expectations.
I saearch Enya as artist and it lists all Enya’s songs but I wanted to discover other artists like Enya.
sorry, just found out the option for discovering similar songs.
Might work well with main stream music but it’s results with electronic music were horrible