[Benesyed]

1. EEG data has a lot of noise. Poor modeling can turn EEG noise into "signal".

2. the auditory cortex is hard to detect on EEG because it's not on the outer mantle of cortex. it's covered by other cortical tissue.

 

[mbphotox]

1. EEG data has a lot of noise. Poor modeling can turn EEG noise into "signal".

2. the auditory cortex is hard to detect on EEG because it's not on the outer mantle of cortex. it's covered by other cortical tissue.

sooo... who gets into a tube to listen to music while we analyse brain activity?

(cannot do headphones, bc of magnets.. but maybe some nice electrostats firing from a distance?)

 

[castleofargh]

I looked into the best methods to analyse the brain and now I’m going to have nightmares about all the primates with stuff in their heads.

 

[71 dB]

sooo... who gets into a tube to listen to music while we analyse brain activity?

(cannot do headphones, bc of magnets.. but maybe some nice electrostats firing from a distance?)

When I worked in the acoustics and signal processing laboratory, one of the projects some other people worked on was designing/developing a method to play sounds to the ears of patients in EEG with long plastic tubes from another room. The challenge was to filter the resonances of the tubes away for high quality sound. This was some 20 years ago.

 

[haci]

Not really. Firstly, the paper to which I linked demonstrated no excitation whatsoever of the auditory cortex (in response to filter ringing or >22kHz content). Therefore, there is no difference to hear, either consciously or subliminally.

Am I reading the paper wrongly? It mentions specificly auditory cortex for the MMN measure, and not the pathways before that. So it is not contradicting the papers cited, that supersonic effects could be there for example at Thalamus, ie subconcsciouss level, that comes before the cortical level in the path of sound signals

 

[gregorio]

So it is not contradicting the papers cited, that supersonic effects could be there for example at Thalamus,

Yep, I stated that it didn’t contradict the previous papers cited, except in terms of the potential/suggested benefits. It’s not impossible that something is registered at the Thalamus but there’s no evidence I’m aware of to support that hypothesis. Although something must be registering somewhere to alter the EEG, assuming that the EEG is actually altered and it isn’t just a “noisy” EEG misinterpreted as suggested above.

ie subconcsciouss level, that comes before the cortical level in the path of sound signals

If the assumption above is correct, which seems reasonable as various independent teams have verified the effect on EEG (and it seems unlikely they all made the same EEG misinterpretation mistake), then something must be occurring at the “subconscious level” (to alter the brain wave patterns). The suggestion of some earlier papers (such as Oohashi) was that this subconscious level effect *might* be perceivable, either by something registering in the auditory cortex which we might hear subliminally or some behavioural or psychophysical change which we could experience and therefore sense indirectly. However, the more recent paper demonstrates nothing is registered in the auditory cortex and no behavioural or psychophysical changes, so there’s nothing to hear (even subliminally) and no change to experience/sense (even indirectly).

So, what’s the difference between: A. No effect whatsoever and B. A purely subconscious effect which causes no differences that might be perceptible in some way (either directly or indirectly) and can only be detected with an EEG? Assuming we’re not all listening to music recordings while watching our EEG, there is no practical difference between A and B. In other words, if there is no effect we can somehow eventually become aware of, either directly or indirectly, the result is identical to “No effect whatsoever”.

The research expands our understanding but completely corroborates our prior knowledge of human perception thresholds and is corroborated by controlled listening tests. It seems conclusive that audiophiles are yet again clutching at non-existent “straws” in order to avoid the obvious, the proven effect of cognitive biases/placebo, just as they have in numerous other instances over the last half century or so.

G

 

[gregorio]

Just reading back, it seems this maybe the cause of confusion:

[1] The original signal lies along certain integers, no? [2] The dithered signal will lie along slightly different integers, no?

1. No! The original signal is constantly varying and the probability of that signal landing on an integer (represented by the bit depth) at any discrete points in time (the sample points) is small. The probability is increased if we use more bits/integers but it’s still relatively small. The original signal would only always “lie along certain integers” if we had either an infinite number of sample points or an infinite bit depth but obviously it’s impractical/impossible to get anywhere near an infinite number of either. So at any practical bit depth (or sampling rate), the original signal does not “lie along certain integers” and our integer values are going to be at least somewhat statistically wrong.

2. Yes, that’s the whole point! Dither slightly changes the integer values so that they are no longer “somewhat statistically wrong”, they are 100% correct, as explained by 71dB.

G

 

[eq1849]

Just reading back, it seems this maybe the cause of confusion:

1. No! The original signal is constantly varying and the probability of that signal landing on an integer (represented by the bit depth) at any discrete points in time (the sample points) is small. The probability is increased if we use more bits/integers but it’s still relatively small. The original signal would only always “lie along certain integers” if we had either an infinite number of sample points or an infinite bit depth but obviously it’s impractical/impossible to get anywhere near an infinite number of either. So at any practical bit depth (or sampling rate), the original signal does not “lie along certain integers” and our integer values are going to be at least somewhat statistically wrong.

2. Yes, that’s the whole point! Dither slightly changes the integer values so that they are no longer “somewhat statistically wrong”, they are 100% correct, as explained by 71dB.

G

1. Absolutely. I'm with you on this.

2. The modified signal cannot be more accurate than the original unmodified signal statistically is my point.


Back to 71's example of something measured at 7.8 rounding down to 7(instead of going to the closer integer 8) 20% of the time.

7.8 constant noise burst over 5 samples:
signal: 7.8, 7.8, 7.8, 7.8 ,7.8

signal after rounding: 8,8,8,8,8
error: (8-7.8)*5=1

dither modified signal after rounding 8,8,8,8,7
error: (8-7.8)*4+(7-7.8)=0


error is cancelled out, sure. But now it is on integer 7 on the 5th sample of the dithered signal. It should be integer 8.

After rounding:
original 88888
dithered 88887

Statistically there is trading some integer accuracy there. There's no way around it.

I know you have a wealth of technical knowledge, so if I'm off somewhere let me know. I debate with you because I respect your knowledge

 

[bigshot]

The thing I really want to know is, how many angels can dance on the head of a pin?

 

[GearMe]

The thing I really want to know is, how many angels can dance on the head of a pin?

3...

just like the answer to the age-old question...

How many licks to the center of the Tootsie Pop

 

[sander99]

@eq1849: this video might be illuminating, it has been referenced in this forum before but maybe you missed it:

 

[bigshot]

It's in my sig file, but no one looks at that...

 

[71 dB]

error is cancelled out, sure. But now it is on integer 7 on the 5th sample of the dithered signal. It should be integer 8.

The numbers "should" of course be 7.8 7.8 7.8 7.8 7.8, but we don't have that luxury when using integers.

After rounding:
original 88888
dithered 88887

Statistically there is trading some integer accuracy there. There's no way around it.

original is 7.8
truncated to integer without dither is 8
truncated to integer with dither is 8 (80 % of the time) or 7 (20 % of the time)

Without the dither you have just 8 all the time and no way of telling how much it is off. Was the original value 7.9 or was it 7.8? Perhaps 8.03? You can't tell. With dither you can do statistical analysis of how much you were off. The percentages of values 7 and 8 tell you what the original value must have been. That's how you effectively know the original signal, but the price of knowing it is the dither noise you added. The dither is a problem or it isn't. In 16 digital audio it isn't, because it is quiet enough to be inaudible.

 

[castleofargh]

Even if it was audible, I would rather have tiny uncorrelated noise than tiny correlated whatever(I want to say noise for quantized noise, but isn’t the fact of correlation with the signal making it distortion?)

 

[eq1849]

The numbers "should" of course be 7.8 7.8 7.8 7.8 7.8, but we don't have that luxury when using integers.

original is 7.8
truncated to integer without dither is 8
truncated to integer with dither is 8 (80 % of the time) or 7 (20 % of the time)

Without the dither you have just 8 all the time and no way of telling how much it is off. Was the original value 7.9 or was it 7.8? Perhaps 8.03? You can't tell. With dither you can do statistical analysis of how much you were off. The percentages of values 7 and 8 tell you what the original value must have been. That's how you effectively know the original signal, but the price of knowing it is the dither noise you added. The dither is a problem or it isn't. In 16 digital audio it isn't, because it is quiet enough to be inaudible.

Even if it was audible, I would rather have tiny uncorrelated noise than tiny correlated whatever(I want to say noise for quantized noise, but isn’t the fact of correlation with the signal making it distortion?)

Mathematically, it's a trade off. less noise for less accuracy. But it's brilliant math, as it does solve the issue from a listening standpoint of not hearing digital noise hiss. So I'm with you...I would dither in this case.

But at the same time...if I have the option where I can get a more accurate signal, with even less noise,...why would I not go with that option?