"sCrAmBlEd?HaCkZ!" - TranceAddict Forums

TranceAddict Forums (www.tranceaddict.com/forums)
- Music Discussion
-- "sCrAmBlEd?HaCkZ!"

Posted by nefardec on Apr-10-2007 20:35:

"sCrAmBlEd?HaCkZ!"

http://www.popmodernism.org/scrambledhackz/

Posted by nefardec on Apr-10-2007 20:38:

postscript (for some reason the embedded video was covering this text up in the original thread)

This is related to a previous thread I opened called "music is obsolete". In it we bickered about a shitload of things. somewhere in there I started talking about dissecting samples into tiny granules and running spectral analysis on them and then resynthesizing using this information -

well looks like it's been done now - and with video too! This guys is into all that pop culture/kitsch/illegality/digital rights shit, but I see a major use for software like this with original music, basically creating original music from original music in this sort of computer generated way.

obviously it's crude now but it's interesting, no?

Posted by 4clubber on Apr-10-2007 22:55:

that is AWESOME! and what's even more exciting is that I have no problem understanding the mathematics and calculation behind it (the 5 190-something-dimensional vectors, and finding the point, i.e. sample bit, in the library closest to an argument sample bit)

and what a coincidence...i just did a project in a scientific computing class dealing with face recognition - to summarize, given a sample of let's say 20 images of different faces (those images are represented as matrices), a different face can be approximated pretty closely using only the data from the 20 image dataset - this is essentially the same concept as scrambled?hackz!
(for more info look up "eigenfaces")

and nefardec, it only seems crude in the video, and for all we know that may be because the sample music/video they are using is limited. in the eigenface analogy, a new face is only very roughly approximated if only let's say 5 images are in the dataset. given a large database of samples, the technology will be less and less crude. theoretically, if the database is large enough, a new sample introduced will be perfectly approximated, i.e. the approximation will equal the original.

imagine the implications this may have on copyrighting...if any track can be represented as a (huge) collection of infinitesimal samples from other sounds. future copyright checks may be linear algebra calculations that tell us how close the matrix representation of one musical piece is to that of another...

Posted by 4clubber on Apr-10-2007 23:19:

Also, another reason it is crude right now is because, as the guy says in the video, it is geared towards live performances, which means the calculations must occur in real-time and not take extremely long. That puts a limit of how large the database can be and how much time is spent looking for the closest approximation of a sample. In the demonstration, there were also cases where the approximation was delayed a few seconds after he gave an argument sample, which leads me to believe the technology already has parameters to set how good of an approximation is desired. If time is not an issue and the dataset is pretty large, then the approximation would be much better.

Posted by tibberino on Apr-11-2007 00:56:

Absolute bore o shite thread!

Posted by 4clubber on Apr-11-2007 01:04:

quote:
Originally posted by tibberino
Absolute bore o shite thread for simple minded fools like me!

fixed...any more useless input?

Posted by nefardec on Apr-11-2007 05:52:

how could you not like HAMMER TIME?

Posted by kadomony on Apr-11-2007 17:02:

amazing!