Computer Generated Music Composition

[elektro80]

[ProEtContra]

It wasn’t a long time ago, with the appearance of computers...
"...computers will never defeat humans in the game of chess..."
"...Robert Moog: a dangerous anarchist out to destroy music as we know it..."
There's no area of art, science... that hasn't been compared in this or any other similar way.
Now, try to imagine musicians today who had never used the computer or chess analysis without comp assistance or...

So what is my point here you ask?

A couple of days ago, I have started in this, as well as in some other popular and well visited forums, a new thread titled "Computer Generated Music Composition". The topic for general discussion was a question "Pro Et Contra" on subject: programs for virtual composing / virtual music composers.

Too often the feedback is the same as in the sentences above mentioned.

So, do we have to wait again for several more years before we admit to ourselves that computers have become an inevitable and helpful tool while composing music. Whether we like it or not, it is reality.
The human touch will never be replaced, but that does not neccessarily mean that it can not be pushed and helped by IT. IT can do it also!

Probably you'll say now: He's one of them (developer, author...). Yes, I'm one of them. But, I have never mentioned any fact which would point to it. So, it is not an advertising. I'm just curious about this topic...

[Inventor]

Very curious stuff to read, folks! You've obviously been down this road before. On the same theme I'll mention this:

I have been writing some neural net programs in ChucK doing real-time music processing. They are simple feed-forward nets, fully interconnected and trained with backprop and rules. So far the computational limits have constrained me to less than 50 neurons, but that seems to be enough for simple tasks.

In the latest one I recorded myself saying "electro-music.com" on a 3 second wav file and trained the net hundreds of times, running backprop hundreds of times per iteration. It learns to mimic a vocoder-like version of what it hears. By mimic I mean it still requires my input to produce speech. Then I add noise to the source and the output degrades gracefully with increased noise.

I know that the program isn't doing much compared to say, one that could really compose music with neural nets, but it's a promising result given only 50 neurons I suppose. The reason I'm working with only 50 neurons is primarily that the real-time processing of the music is so compute intensive (FFTs and IFFTs) and the computer is kind of slow by today's standards. With non-real-time processing and a faster machine, one could use hundreds or even a few thousand neurons - and that's available today not tomorrow. With that many neurons perhaps the system would be capable of much more.

So as we've briefly touched upon in another recent thread, the emergence of AI music tools is upon us. My question is: what will we do with these tools? I'm open to suggestions and might try to code one up if inspiration strikes me...

[bachus]

[bachus]

[dewdrop_world]

[elektro80]

[elektro80]

[Sam_Zen]

I don't see much difference between a computer and the average musical instrument.
Both are tools to make sound.
But I call a computer, like any electronic device, an 'autophonic' instrument, because one doesn't have to hit or blow, etc.
So it becomes more a challenge to the 'muscles' of the brain.

Some aspect :
If an electronic device like the computer is used to compose, it's not only the range of possible sounds that has increased.
The possible ways to compose are increased as well. Different concepts even.
A work could be based on the ruling 'musical' things, but could also be based on a technical 'construction'.
Or some mathematics, like fractal algoritms or the golden section.

Another aspect :
The notation of the electronic composition. There has been hundreds of efforts in the course of e-history, all failed so far.
I mean as an option of playing the piece by someone else. How many 'covers' have been done in the e.m. world ?
Of course one could emulate wit some software the normal horizontal score like on paper, piano roll setting, etc.
But with the computer the original score also can be present in the form of a data-file.
Like the tracker-module format, complete with embedded instruments. Or a session-file of a multitracker-mix.

[iPassenger]

Completely agree with Eletcro80 on this one, re:decisions and intent.

Which is where any algorithmical compisition needs human input and involvement, it would be difficult for a computational process to take into account your current mood, planned objective, musical bias and so many other unique factors that may well influence the final set of decisions and ultimately the composition. I think it is great that we can get machines to build and create musical phrases and ideas based on set parameters be they mathematically or classically influenced.

[solyaris]

[x_x]

Computer-generated music is different, because the art is hidden in the code/algorithms ... Schopenhauer said that music is a sketch of will. Computer-generated music reflects an illusion of will, hence we're talking about sketching the illusion of will of the will.

I think, some of you are confusing computer-assisted composition with computer-generated music.

It's art out of science and art

Anyways there is an interesting book titled "computer-generated music" by de IEEE computer society press... it's somewhat old but it's interesting because it explains diverse frameworks that have been developed, if anyone interested ... I don't know if they sell it, but you can look for it in the library, that's how I found it.

[solyaris]

[dewdrop_world]

[solyaris]

[dewdrop_world]

[solyaris]