New search engine "listens" to music to help you find new tunes
By Nilay Patel 
posted August 18th 2007 2:27AM
posted August 18th 2007 2:27AM
Peer-recommendation services like Last.fm and Pandora are pretty good at leveraging the power of the community to help you discover new music, but a recent grant from the National Science Foundation to the College of Charleston aims to take the concept to the next level, by creating a search engine that "listens" to music and creates critical comparisons between works. The system, as described by Ars Technica, involves a neural network that is trained to recognized the composer and style of music, an evaluation engine that's supposed to simulate human taste, and a set of objective metrics like pitch, tempo, and duration. The results are then combined and the system can then recommend matches to find similar music. The researchers have already demoed a similar system with good results, so here's hoping the grant money helps them refine things further -- we've been looking way too long for the next Wham![Image from O'Reilly's Digital Media Blog]
Name that tune just got a whole lot easier, just hum the few lines you know in to a microphone and the computer will try to match it.
I read about and saw this technology 8 or 9 years ago. Some (Asian) girl at MIT was demonstrating this tech on television back when MP3.com was the biggest site on the web. Good to see it finally come to fruition.
Lots of techno music.
I have an application on OS X that I use called ieatbrainz or something like that.
Anyways, it scans the song and checks the audio against a database to fix ID3 tags.
It needs some work (It pretty much gave all my albums it could detect the BEST OF [insert artist] tag, and screwed some Roll Tide for the 3's a company theme song.
I wonder hwo much better this is though.
Also don't mind my bad typing, I'm on a Anykey keyboard right now, and I'm way too used to my Model M.
so it is similar to this...http://www.musiclens.de/contest/
Pandora doesn't use peer recommendations like Last.fm, it actually suggests songs that are musically similar according to the Music Genome Project.
That aside, this sounds pretty friggen cool.
pun intended?
No, if you read the article, I think the point of this is to listen to the music and match it with songs with similar melodies. Personally, I'd be skeptical of this because I wouldn't want too many different songs that sound the same, so it would have to change it up a little.
If it were just song recognition based on listening, then that certainly wouldn't be news. I have a program on my phone that does that.
And what if I listen to underoath or opeth or lamb of god, which does not have much(if any)tune? :)
you've obviously never listened to much Opeth then son, check out Damnation
Will it be able to differentiate between a Linkin Park song and an mp3 of two howler monkeys mating? I can't.
Can you ultimately search a tune by singing it?
Isn't there already a site that does that?
I thought Midomi has been doing that for awhile. I haven't used it, but it sounds similar to what's being described up there: http://www.midomi.com/index.php?action=main.profile&recording_id=f056e9acd6ccace94975264a3e4afaee
Verizon's Vcast song ID works pretty well, but this seems to be much better,
also there's the musicmarker music match thing thats built into Venzero DAPs
Here's a Windows/Mac app that can listen to music from the computer's mic and recognize the artist and album name
http://thinkabdul.com/2007/08/07/tunatic-free-windowsmac-osx-application-to-recognize-identify-artist-name-title-of-song-or-music-album-through-microphone/
We've discussed this may times and one experienced developer of musicology software points out that we love Bob Dylan, but nobody wants to hear anyone who sounds like him.