Music Created by Learning Computer Is Impressive | Hacker News Reader

Music Created by Learning Computer Is Impressive

singularityhub.com

37 pointskkleiner16 comments

NPR covered David Cope's software, named "Emily Howell", this weekend:

http://www.npr.org/templates/story/story.php?storyId=1137194...

It's funny that All Things Considered, not ArsTechnica or singularityhub, included a little detail that will amuse some HN readers:

RAZ: What exactly is Emily Howell, and how does she, it work?

Prof. COPE: Well, that could be a very long answer, but simply put, it's a computer program I've written in the computer programming language LISP.

Sample MP3s are here: http://arts.ucsc.edu/faculty/cope/mp3page.htm

I think the Chopin one is eerily good.. Perhaps there's an uncanny valley for music too?

I studied Cope's book and synthetic music in college.

His Chopin pieces on the surface kind of sound like a Chopin piece, but when you look at the underlying forms the complexity just isn't there. Musicologists can consistently identify the synthetic pieces.

I feel that the lack of the underlying complexity is what makes this almost beyond the ability of a human to create in that removing ourselves from patterns/meaning when creating art is incredibly difficult. And to some people the lack of meaning is the beauty. So I am not disagreeing with you - I was just inspired by your comment to think about what I like about music like this.

Yeah, that's what I found eerie about it! It's close but not quite Chopin.

Maybe the Bach and Joplin pieces didn't bother me so much because baroque counterpoints and piano rags are already kind of formulaic.

I'd love to hear (or maybe be creeped out by) their program trying some jazz greats. That's a style I'm much more familiar with.

I obviously can't promise this isn't my own bias, but the Joplin-eque piece just doesn't flow the same as any of the actual Joplin rags I'm familiar with. Same with the Bach. There's something compelling in the originals that seems to be missing from the ersatz version.

I'm a little skeptical about the role of snobbery in this. It's impossible to do a properly blinded experiment, because the folks with the necessary knowledge already are able to identify any Chopin (or Bach or Beethoven) piece, and thus know a priori which is the synthetic piece. This seems rather too similar to the story about winetasters (e.g., here, sorry for the cache link: http://74.125.93.132/search?q=cache:SB6S068de0gJ:www.latimes... )

Snobbery is not the right word, but you make a valid point.

The download links are asking for a password.

The last part of this Radiolab episode features EMI. Overall it's a very interesting talk.

http://www.wnyc.org/shows/radiolab/episodes/2006/04/21

Did anyone find the Beethoven subtly similar to Beethoven's Moonlight Sonata?

Having played a lot of Bach when studied piano, I recognized a lot of parts sounding exactly like his pieces.

More than subtly.

It was kind of clear that these songs were modeled off of specific previously written songs. The Bach sounded a lot like Invention 8, Beethoven's and his Sonata, Joplin and Maple leaf rag.

See also Wolfram Research Lab's Mathematica-driven ring tone generator:

http://tones.wolfram.com/

Or maybe popular music is simplistic and this is impressive compared to it?

Reply on news.ycombinator.com