Digital Creativity

[Ogg Vorbis]

I wonder when computer programs will be able to create music as convincing and "musical" (whatever that means) as human composition.

Will we ever see a side-by-side comparison of a human-composed piece of music and then a computer-composed piece where most listeners will not be able to tell which was which?

This discussion will likely fall into two camps. One camp (probably most people) will say that there is some emotional stuff which a computer won't be able to reproduce no matter how complicated the algorithm.

Another camp might be that computers will be able to emulate human creativity so convincingly that soon we'll hear music that will blow people away.

I guess I don't see either option very clearly. I don't believe emotionalism defines music, but I also don't think it's algorithmic either.

Do you think AI-based music is a dead-end street?

[synthgeek]

I'm tempted to say it is possible, for two somewhat-related reasons: many very "emotional" pieces of music are quite simple, and I've seen computers do other things I would have said were impossible at one time.

It may not be a "works every time" kind of thing, but then even great human songwriters have a dud sometimes, right?

[psenior]

I have little experience with this kind of software. The one put out by Microsoft, it's a bad joke. The most impressive type of software I saw like this was one where a jazz pianist played and the machine was able to figure out what they were doing and emulate it eventually. That was interesting.

Lots of musical styles are very formula-based and could be done by machine, no problem. I use Band in a Box all the time; for certain things it works well. However, I am telling it what chord changes to play and the parts it generates are samples of real people playing real instruments.

I hope AI, in all its forms, keeps getting better and better.

[Gamma-UT]

Will we ever see a side-by-side comparison of a human-composed piece of music and then a computer-composed piece where most listeners will not be able to tell which was which?

Yes. Arguably there already is with David Cope's Emily Howell: http://arstechnica.com/science/news/200 ... oversy.ars

At least in the early stages, the best chance for a computer in a blind test will be music that has a lot of structural rules, such as a fugue.

People tend to argue that the computer can't produce emotive music. But there is a growing body of research that suggests that years of musical conditioning mean that people are highly susceptible to certain techniques for conveying emotion over and again (things like chromatic mediants, deceptive and delayed cadences etc). David Huron's book Sweet Anticipation isn't about AI in music but provides some clues as to how this kind of research is likely to proceed.

[jopy]

I'd say there's a fair sized proportion of the population who think that involving a computer in any way isn't as musically creative as performance with physical instruments. They'd reject even the stuff where a person is doing most of the work and just using a computer as a tool. No one who comes here, of course, but it's not an uncommon opinion in my experience.

[Ogg Vorbis]

Will we ever see a side-by-side comparison of a human-composed piece of music and then a computer-composed piece where most listeners will not be able to tell which was which?

Yes. Arguably there already is with David Cope's Emily Howell: http://arstechnica.com/science/news/200 ... oversy.ars

At least in the early stages, the best chance for a computer in a blind test will be music that has a lot of structural rules, such as a fugue.

People tend to argue that the computer can't produce emotive music. But there is a growing body of research that suggests that years of musical conditioning mean that people are highly susceptible to certain techniques for conveying emotion over and again (things like chromatic mediants, deceptive and delayed cadences etc). David Huron's book Sweet Anticipation isn't about AI in music but provides some clues as to how this kind of research is likely to proceed.

Interesting article. I remember hearing David Cope lecture at my university when I was a student and he was very engaging. In this article he makes a wonderful point about the human element in music, which is to ask "who cares where the music came from, we are hearing it."

I suppose the real threat to humanity would be replacing the audience with computers, not replacing the writers.

[Gamma-UT]

I suppose the real threat to humanity would be replacing the audience with computers, not replacing the writers.

Well, that's where David Temperley comes in: http://theory.esm.rochester.edu/temperley/

[jancivil]

I think it's quite possible given an extremely high level of conditioning on the intended audience's part.

In classical music terms, that isn't as much a stretch as some may think.

I mean there are people who are susceptible to certain phenomena more than others.

I'm pretty much immune to being manipulated by, eg., a certain kind of chord change with the right notes, into 'experiencing' the 'intended emotion' and need more than that, but I'm a special kind of junkie compared to some.

I'm less immune to let's say a singer (or a real performer on an instrument, 'singing' with the instrument), but/and/so I am extremely skeptical you can AI the reality of a person anytime soon

[Musicologo]

I believe computers already do that. Many programs (those of David Cope are fantastic!) already do that, but you have more stuff, look at Karlheinz Essl work also, and why not, Brian Eno. Have you already seen the application "Bloom" for Iphone?

There are many admirable algorithmic and AI devices that emulate certain type of "reasoning" producing beautiful scores in many different styles.

The problem then, relies on the transposition of the score for music itself. Sometimes the music created is not very well played, with defficient sounds or very mechanical interpretations.

But even that part is getting better and better.

The programs that create synthesized ambient music are the best ones, since synthetic sounds, well, are synthetic! And already came from machines since ever, so they fit well. That's why I believe "bloom" is huge.

Programs that deal with acoustic instruments are not there yet. The midi lexikon sonate sounds crappy, but if you could imagine that midi sound, passing through a decent soundfont and a humanizer (like the one in Finale, for instance), then the result could be 100% better.

And then, you have voice. When it comes to voice music forget it. We are still miles away from convincing voices in real time.

So, It's possible. It's there. But we have some more years of work ahead, specially in the field not of composing, but on performing the composed music, to mimic human beings. And of course, regarding acoustic and organic instruments.

[Ogg Vorbis]

I suppose the real threat to humanity would be replacing the audience with computers, not replacing the writers.

Well, that's where David Temperley comes in: http://theory.esm.rochester.edu/temperley/

I looked through this but are you saying he's modeled human musical perception? That's an interesting opportunity! We could simply set up a network of computers to produce, perform, perceive, critique, catalogue and continue our artistic heritage for us so that we can finally get down to the serious business of de-evolving back into protohominids.

[xoxos]

nb gels much better with vocals (keep in mind that there are ten classes of passages, eg. verses, riffs, rejoinders, solos.. without solo parts atm. without vocals, the verses don't discretise as well) give breathcube 1 a few runs. occasionally it will dazzle. going back to genesis, a beat is an interval in time.

it is not to say i am a font of enlightened engineering. more that it's taken >40,000 lines of code to get it to this point. maybe some more gifted people should be doing what i'm doing with me, since i know nowt about programming or music theory.

what patterns are human patterns?

well, i'm not from this country, which is why i made what i made. because what i hear is a subset of what i know otherwise to be human.

i wish people could think like me some more, or more. cosa nostra cosa nostra.

[djanthonyw]

Haven't you heard of Micro soft ?

[Meffy]

No algorithm could ever match the genius of Mozart!

http://sunsite.univie.ac.at/Mozart/dice/

Well. Mostly. (Yes, it's an obvious cheat. But consider how long the idea's been around.)

[eduardo_b]

Given some of the popular genres of music, I'd say it's possible now. Not all human-based music is complex or even interesting, so computer emulation should not be difficult for some of it. Is that computer tarnce or human tarnce? Does it matter?

[Gamma-UT]

I suppose the real threat to humanity would be replacing the audience with computers, not replacing the writers.

Well, that's where David Temperley comes in: http://theory.esm.rochester.edu/temperley/

I looked through this but are you saying he's modeled human musical perception? That's an interesting opportunity! We could simply set up a network of computers to produce, perform, perceive, critique, catalogue and continue our artistic heritage for us so that we can finally get down to the serious business of de-evolving back into protohominids.

His group has written a number of analysers for things like metre, key and phrase structure that attempt to process them in a way analogous to what the brain does (based on existing theories of music cognition). There is already the Melisma music analyser: http://www.link.cs.cmu.edu/music-analysis/

All this is fairly basic compared to what the brain seems to do. However, as the models of how the brain understands music improve it's only reasonable to expect simulations to get better. If nothing else, they might be useful as teaching aids. Practica Musica, for example, has a counterpoint analyser. It's not going to tell you whether you can match Palestrina in ability but as the rules for species counterpoint were codified by Fux with pretty firm rules, it is relatively straightforward to write a program that works through them and tells you whether you're just doing it wrong.