AI music?

[Tennessee Leeuwenburg]

Crazy idea, just writing it down, not really planning to do anything.

I wonder if you could 'learn' to generate music... take a whole lot of (say, midi) files that you "like", then learn associations between say one bar and the next, make a markov chain or similar then randomly generate music. Maybe include multiple randomisation parameters so as to avoid getting stuck in short/repetitive loops to allow more drastic probablistic leaps.

Has it been done?

What other file formats parameterise music well? 

-T

[Petro Verkhogliad]

http://en.wikipedia.org/wiki/Algorithmic_composition

That seems like a good starting point. I've seen some relatively
successful attempts on this subject as well.

Petro

> --
> You received this message because you are subscribed to the Google Groups
> "Python Artificial Intelligence SIG" group.
> To post to this group, send email to ai...@googlegroups.com.
> To unsubscribe from this group, send email to
> aipy+uns...@googlegroups.com.
> For more options, visit this group at
> http://groups.google.com/group/aipy?hl=en.
>

[Guy K. Kloss]

On Thu, 02 Sep 2010 14:19:50 Tennessee Leeuwenburg wrote:
> Crazy idea, just writing it down, not really planning to do anything.

One guy from the NZ Python User Group (Douglas Bagnall) has done something
that emulates TV series. He's analysed with some code the general
picture/scene structure of hours of Shortland Street (NZ soap), and then
generates according to the captured heuristics blurred versions of "new
episodes" that mock overall geometry, scene changes, etc.

I could imagine that there's some similarity that can be used on music as
well.

Guy

--
Guy K. Kloss
Institute of Information and Mathematical Sciences
Te Kura Pūtaiao o Mōhiohio me Pāngarau
Massey University, Albany (North Shore City, Auckland)
473 State Highway 17, Gate 1, Mailroom, Quad B Building
voice: +64 9 414-0800 ext. 9266 fax: +64 9 441-8181
G.K...@massey.ac.nz http://www.massey.ac.nz/~gkloss

[Ian Ozsvald]

Is there a link to anything? This sounds mighty interesting!
Ian.

--
Ian Ozsvald (A.I. researcher, screencaster)
i...@IanOzsvald.com

http://IanOzsvald.com
http://MorConsulting.com/
http://blog.AICookbook.com/
http://TheScreencastingHandbook.com
http://FivePoundApp.com/
http://twitter.com/IanOzsvald

[Chris Babcock]

On Sep 1, 7:19 pm, Tennessee Leeuwenburg <tleeuwenb...@gmail.com>
wrote:

There are classifiers in CRM-114 that you can use for filtering.
Hyperspace would be best for binary data. I'm familiar with using
Markov chains and similar as classifiers, but that is a different
problem than search. Training an AI to recognize music you like is the
same order of difficulty as a spam filter. Creating music this way
would mean generating thousands or millions of songs algorithmically
and filtering for ones you like.

Compressed formats like MP3 probably destroy semantically relevant
information in compression, so look for verbose formats. For images,
vector formats like PostScript and SVG are handled well by most
classifiers while bit maps need classifiers made for binary data.

Chris

[Guy K. Kloss]

On Thu, 02 Sep 2010 21:38:26 Ian Ozsvald wrote:
> Is there a link to anything? This sounds mighty interesting!

This first one is on the Te Tuhi video game machine (making "one off" games on
demand in an automated way from crayon drawings of kids). It also mentions the
Shortland Street Emulator:

http://www.tetuhi.org.nz/exhibitions/exhibitiondetails.php?id=34

And this is a link to Douglas' Shortland Street Emulator:

http://halo.gen.nz/mimic/

Some of Douglas' techniques used for implementations can be (more or less
roughly) outlined in the Kiwi PyCon 2009 proceedings:

http://ojs.pythonpapers.org/index.php/tppm/issue/view/16

HTH,