[music-dsp] auditory patterns

[James Chandler Jr]

On Friday, June 6, 2003, at 08:17 PM, Joshua Scholar wrote:

> all this talk about hearing patterns in noise and seeing patterns in 
> TV snow
> (for God's sake), begs the question of whether any of these "patterns"
> really exist or if they're all hallucinations.

Hi Joshua

In the case of pseudo-random noise, yes the patterns "really" exist. 
Pseudo-random noise is often generated by a long XOR-tapped shift 
register. The feedback is configured so that it generates an "optimal 
length" pseudo-random pattern (loop-playing thru all possible values in 
a "randomly distributed" fashion). A long sequence of values repeating 
perfectly ad infinitum.

Hardware IC noise chips (as used in spectrum analyzers ferinstance), 
clocked at audio rates, typically run full-circle and repeat with a 
period of a few seconds.

If there was a reason to do so, you could home-brew much longer 
shift-register noise generators by stringing together many generic 
counter chips.

In software, it is easy to code up arbitrarily long repeating 
sequences. Lots easier than soldering a bunch of chips together.

Many common "built-in" computer system RND() functions have used 
shift-register techniques.

A spectrum analyzer will measure the repeating pattern perfectly flat 
(which is why pseudo-random noise makes such a good analog analysis 
source).

You can use a $1 shift-register noise IC and get "guaranteed flat 
response"...

Or you can reverse-bias a diode or transistor for fully-random 
non-repeating noise. But each individual noise diode or noise 
transistor, installed in each unit on the assembly line, would have 
different, unpredictable spectral distribution. This is not a good way 
to manufacture inexpensive, repeatable analyzers (GRIN).

Though the patterns "really" exist, humans don't notice the patterns 
with short-term exposure.

Under certain long-term conditions, humans can detect the patterns. I 
guessed that arbitrary room response filtering can make the patterns 
easier to hear, but that is just a guess.

Would be willing to bet that if you feed pseudo-random noise thru a 
narrow bandpass filter, you would identify the repeating pattern in 
only a few loops. But a room response is more subtle and complex, so it 
would take longer for the ear to detect the interaction between pattern 
and room.

Am not denying that there is a "hallucinatory" aspect to long-term 
noise listening.

If one hears repeating patterns in non-repeating "pure" noise, the 
detected patterns would surely be 100% hallucinatory? Perhaps the 
"tempo period" of such audio illusions would be related to the brain's 
perceptual loop time-constants?

I don't recall hearing "patterns" in silence, so perhaps noise is a 
pretty effective stimulator for auditory illusions?

> James Chandler "sees" things in TV snow that he's sure are somehow 
> really
> there, being so "obvious".

Heh! I never claimed the "TV snow" effect is related to "real" patterns 
in the snow. Had assumed that this phenomena is near 100% hallucinatory 
(or "optical illusion" if one prefers).

Though I THINK TV snow is pretty random, one can't just assume that TV 
snow is truly random... Color NTSC horizontal and vertical oscillators 
run at fixed ratios with a bit of slippage, and there is "beating" 
designed-in to various stages of the receiver circuitry. Also, adjacent 
"live" TV channels might "beat" against each other and leak into the 
receive channel. It could be that TV snow does have repeating elements. 
Dunno and would dare not guess (GRIN).

Have seen "TV snow" screen savers and Video Editor plugins, but would 
guess a typical "TV Snow" screen saver or plugin might be based on 
RND(), and not truly long-term random?

One might have to go to a bit of trouble to make a "truly random" TV 
snow test program, to rule out the possibility of "real" repeating 
patterns in the snow. Even going to such trouble, one might wonder 
about "patterning" accidentally introduced by interactions between the 
scan/update rates of the software vs video card vs display monitor?

> The obvious test is to record some white noise that's mathematically 
> perfect
> in some sense (average volume, for instance), loop it and check whether
> listeners can actully detect how often the sample loops.

Shift-register noise is "pretty good" from a 
long-term-fequency-distribution standpoint, but the instantaneous 
frequency distribution evolves over the period of the loop.

I'm mathematically ignorant, but is it possible to make a "random" 
signal where all frequencies always maintain equal instantaneous levels?

A truly random signal can be expected to occasionally hit fairly long 
stretches of repeated or similar values?

When a "truly random" signal hits the occasional string of repeated 
similar values, wouldn't the instantaneous high harmonics decrease and 
low harmonics increase? Similarly, when the random signal hits a 
stretch of unusually chaotic values, wouldn't the instantaneous high 
harmonics increase and low harmonics decrease?

If we force all harmonics to be always equal, aren't we restricting the 
random nature of the signal? Making it "less random"?

> A point, that I didn't really make clear in my last post is that you 
> if you
> think you detect a pattern in looped noise that's at a different rate 
> than
> the actual looping then the pattern probably doesn't exist except in 
> your
> nervis [sp] system, and that would be interesting to check for.

The patterns I heard in shift-register noise, always had the same loop 
period (two or three seconds, never bothered to time it). However, 
listening position in the room would change the perceived pattern 
within the invariant loop period.

James Chandler Jr.