Computer Music (G6610, Fall 2011) week 5

Perceptual Models and Filter Theory

We started the class by revisiting the 'warped cat' fun we had using SPEAR data back in week 3, this time with the parentheses in the correct place! Following that, I showed a very simple neural-network perceptual model intended to classify timbre based on crude spectral (FFT) data. For kicks I worked up a max-patch based on my old friend Eliot Handleman's music/machine-perception exercise (the 'perceptual model' would be bored or afraid), using timbral complexes as the incoming data. It turned out to be a nifty way to make some interesting sounds, too.

We finished the class with a discussion of digital filter theory because filters play such a prominent role in shaping and understanding timbre.

Links

neural networks:

If you google "neural networks" or "connectionism" you will find many many many pages about these things. Here are a few links to get started:

Introduction to Artificial Neural Networks -- looks pretty decent. The older links I had used are all gone now. limited FFT display capabilities
class at Willamette University -- a bunch of interesting links at the bottom of the page.
Parallel Distributed Processing, Vol. 1 -- the Rumelhart and McClelland book. Start here if you are seriously interested (and get Vol. 2, also).
Music and Connectionism -- by Peter Todd and Gareth Loy. Although a bit dated, this is still an excellent collection of papers discussing the used of networks in music perceptual and compositional research.

filter theory:

As with neural networks, there are a billion pages on the web that deal with digital filter theory. Some are really basic, and some are ridiculously technical.

older class web page -- most of the links on this page are now dead, but there is a bit of discussion and older patcehs that may be useful.
Luke's filter web page -- nice tutorial written by Luke DuBois back when he was here.
Yehar's Digital Sound Processing Tutorial for the Braindead! -- funny name, actually really good explanation. By Olli Niemitalo. I don't know why he goes by "Yehar".
Digital Filter Tutorial -- by Julius O. Smith, one of the heavy-duty guys. Serious stuff. Almost all of Julius' work is on-line here, but it seems he has migrated a number of his publications to ths DSPRelated.com web site. If you click on the titles you still have public access to the on-line materials.
Miller Puckette's Filter Chapter -- from Miller's on-line DSP book, good info.

Class Downloads

catwarp-lisp.zip -- the LISP code showing how to mess with my cat's meow. See the week 3 class web page for how to run it. Again, this is not really "ready for prime-time", but the basics are there for anyone who would like to expand upon it.
nn-timbre.zip -- the source code for the simple neural-network timbre classifier we did in class. I left the executable files in the archive for those running on OSX machines and who don't want to mess with recompiling with the Makefile. Daniel Iglesia's max-patch version is also included (in the "iglesia-net" subfolder).
fear-boredom.zip -- the max patches for the Eliot Handleman fun. Send it spectral clusters, it will be afraid or bored! Whee! Plus it makes fun sounds while living it's two-state life.
filters.zip -- the simple max patches demonstrating basic filters I used in class.

I haven't included the LPC max-patch here, it is included as part of the [rtcmix~] help documentation.

Assignment

Yet Another Actual Assignment for next time!