I tried to attach two 10-second wav files to this message but I couldn't
because I couldn't compress them under 750k or so, if you want to play
with them, I'll send them off-list. Basically they were the first few
seconds of a song, but each file had half the samples of the original,
one was the odd samples and the other the even samples. The point was
that the files sounded pretty much the same.
These files are completely different! Check them out side by side in a
hex-editor, after the first few bytes of header information they're
completely different! Not any similarity whatsoever!
Sure, the correct thing to do is FFT both files and filter them until
the amplitude and frequency distributions show pretty much the same
music. (I split a single wav file into two lower-resolution ones, with
the first file having the odd samples and the second having the even).
But what I want to get into is an idea that's been percolating in the
back of my mind for a week or so. It's kinda a reflection on one of the
several neural codes in the brain and how it is distinct from all
conventional approaches.
All conventional computing is based on words. In all conventional neural
network simulations that I'm aware of use a vector of discreet values.
Basically, each machine word being a discreet unit. What if you designed
your computation around a continuous bit-stream like:
http://en.wikipedia.org/wiki/Direct_Stream_Digital
I'm sure that if you converted my files of discreet samples into
bit-streams, the underlying similarities would again appear, but in a
much more generalizable way... This seems similar to the approach the
brain actually uses. =P
This is as close as a digital system can get to a Steve Richfield
approach. ;)
Conventional CPUs are probably The Suck at this type of bit-twiddling,
but an FPGA could probably do it reasonably well.
Anyway, being an imbecile, I haven't been able to progress very far
towards a practical algorithm with this idea. =\
--
E T F
N H E
D E D
Powers are not rights.
-------------------------------------------
AGI
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-c97d2393
Modify Your Subscription:
https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-2484a968
Powered by Listbox: http://www.listbox.com
