Slashdot Mirror

IMHO this is yet another example of how academic projects are judged by the amount of attention they attract, rather than on whether they advance the state of the art.

Every Spring semester at Michigan State University's Computer Science department, the capstone class (taken by seniors to graduate) did a project and had a "poster competition" to see who did the best project.

The team that won the year I saw them was the team that wrote a program that graphed a song's FFT over time. That's it. They went on to babble about how you can recognize a song based on how it looks, visual recognition, and it did some ill-conceived 3D stuff that, by making the song data fit into even less space on the screen, was even more impossible to see. (I think you were supposed to eventually pick the song you wanted to hear by looking at this tiny, tiny representations.... at the risk of potentially offending one of the authors, who may conceivably read this, that's stupid! If they just seriously tried it once, they'd have seen how poorly this worked.) (See here for an example of a guy playing around with that kind of graph; note most songs look NOTHING like that in an FFT graph. ;-) )

The fact is, it's a neat idea but it doesn't work. All songs in a particular pretty much look alike in an FFT graph. The differences are pretty minimal. Making it smaller doesn't help at all. The program looked really cool on a poster, using one song, but use it on six or seven real songs and ask even yourself to distinguish them and you can't; you don't "see" and "hear" that way.

IIRC a dot-com was founded based on this idea, AFAIK indepedently derived.

What does this have to do with your post? I thought about half of the other posters deserved the prize over this project, in that they were useful, interesting, or potentially even groundbreaking, in the small way that a semester project can be. But they didn't have a Beatles song graphed out on their poster. They lose.

Even college professors aren't immune to judging on surface appearences and glitz, rather then real value.

Even if you can't point out a true aural difference between CDDA and a 256kbs MP3 (I'm not too familiar with WMA and how the quality scales), the fact of the matter is, you are not getting the original material as the artist intended it. I ran across a cool link on /. a couple weeks ago about about the Aphex Face. Granted, its a highly specific example, but any kind of lossy compression at any bitrate, whether you can "hear" the difference or not, would destroy that aspect of the sound wave. To me, that is inexcusable for a final product and is inherently inferior.

Furthermore, even if human ears can't hear the difference between a CDDA source and a compressed version in a lossy compression scheme, further lossy compression using the WMA/mp3/whatever as a source, *will* make a noticable difference, as more and more "unimportant" parts of the sound are thrown away. This is why I will never purchase music in a lossy format--the flexibility of the file is severely limited. This seems like a very likely scenario for music bought from Buy.com's store. Since you can't redownload what you buy if you lose it, if your only permanent copy of the music is a Redbook audio cd, any mp3s you make from this source will be noticably inferior to the orignal WMA.

Well, if it was Aphex Twin music, I'd be very impressed that their songs were runnable on a C64 :)

Aphex Twin?

It turns out you get a clearer picture of the face if you use a logorithmic scale, as described on this page. They've also got a few more neat pix for other tracks and a link to a program to make your own.

Find more info about the Aphex face, as well as some software to do same thing yourself
here.
Additionally, more information on it can be found here