On Thu, Sep 5, 2013 at 4:20 PM, Jeff Hawkins <[email protected]> wrote:

> Patrick,
> I don't know what the equivalent would be for audition.  The dynamic range
> of any individual neuron is fairly limited, I think it is much less than
> the
> dynamic range you are able to perceive and detect.  So distributed
> representations are somehow making up for the limited dynamic range of
> neurons.


Well, there's a fascinating trick that the brain does with sound:  You can
express any waveform that you perceive as a single note as a bunch of sine
waves of different frequencies played together - harmonics.  We hear a
440Hz A-note from an oboe, but 880, 1320, 1760 and other higher multiples
are there - so in fact, a, say, triangle wave can be decomposed into a list
of sine waves of different frequencies.  Those other frequencies are what
make one instrument sound "brighter" than another.

What is interesting is that you can actually *take away* the 440 - the
fundamental note - entirely, but you will still perceive it to be there (I
read somewhere that some speaker systems that have surprising bass in a
small package actually synthesize *harmonics* of low notes that can't
physically be reproduced, and our brains fill in the low frequencies that
aren't actually there).

So for most sounds, you're tickling things that detect a lot more than one
frequency, and there's definitely some pattern recognition that allows for
a set of harmonics minus the fundamental to let the brain perceive the
fundamental note that's not there.

-Tim
_______________________________________________
nupic mailing list
[email protected]
http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org

Reply via email to