Logical Fallacies in Audiophile Discussions

[Part-Time Audiophile]

@WGScott: "In other words, you are assuming what you set out to prove, so you cannot logically conclude anything."

 

This was my point exactly, to Harley, when he wrote his self-serving attack on ABX testing. Another way to put this was, "You are begging the question." Rather common, actually.

 

Another version: "If your system were resolving enough [fill in the blank]". The assumption being twofold. One is that "my" system is better than "your" system, whatever "your" system might be. Two, if your system didn't suck, even a poor tin-eared bastard like you would hear "it". Again, this is guilty of begging the question.

 

On the other hand, while it is perfectly valid to call into doubt the validity of a particular empirical observation, such as challenging someone's claim that they heard a difference between two cables, that isn't the same as rejecting the observations as invalid or unsound. That is, "challenging" is not equivalent with "falsifying", and the burden of proof can be easily be met with additional observations or by altering the observational conditions. "Science Audiophiles" make this error (conflating a challenge with a falsification) quite frequently.

 

That said, offering an alternative covering theory (such as The Placebo Effect, or Cognitive Dissonance, or Drug Induced Hallucination), actually does invalidate all such personal observations -- but it does so only by changing the hypothesis. Note that in the case of the examples I mentioned, this also moves the argument from the realm of metaphysics ("what there is") into epistemology ("how we know what is").

 

This results in a neat logical fallacy called a "Category Mistake". A Category Mistake is when one side mistakes a thing (or a claim) to belong to one category when in fact it does not. In this case, the opposing sides hold different frames when arguing. One side is purporting to make claims about the world. The other side wonders if the first is possessed by an Evil Demon. Best result? No communication. Worst case? You and Descartes find yourselves with your brains in a glass jar.

 

When an epistemic challenge gets made, the burden of proof is to now show that the proposed/assumed bias does not in fact apply. Until that burden is met, all ontological claims are irrelevant and invalid. Not necessarily unsound, but merely invalid. Said another way, when your opponent is pulling the epistemic rug out from under you, you have to show "how you can know" before you can go on to make any claims about the world around you. This is why I always say, "Never bring a metaphysical knife to a epistemic gun fight."

 

[esldude]

"it would be great if someone could post links to files that the CA community could then test to confirm/dispel the fallacies/myths/etc. for example, 2 files from the same master - tweaking whatever variable is at issue - and ask if anyone hears a difference, and what it is, etc."

 

They have done such things many time at the www.hydrogenaudio.org which has many such tests related to various encoding schemes, using headphones usually. Lots of info here. Probably a general community that supports rational ABX testing for the most part.

 

[wdomeika]

in this insane conversation, that wgscott is actually wearing the aluminum foil helmet I described in a post earlier today (the spike is cutoff at the top of the picture).

 

"What's the frequency, Kenneth?

 

cheers,

 

Bill

 

[wgscott]

http://berkeley.intel-research.net/arahimi/helmet/

 

 

[wdomeika]

Your mother slaved as a powder girl in a wig store in Chelsea, to send her only son to MIT. So he could experiment with Foil...which he's likely been doing since he was 4 years old.

 

FYI...there's no Field's Medal for Foil.

 

cheers,

 

Bill

 

 

 

[sandyk]

"it would be great if someone could post links to files that the CA community could then test to confirm/dispel the fallacies/myths/etc. for example, 2 files from the same master - tweaking whatever variable is at issue - and ask if anyone hears a difference, and what it is, etc."

 

I did that in CA quite a while back.

I have also uploaded dozens of comparison .wav files for Rock Grotto members, and a few from DIYAudio.

The main problem is limiting the access to the comparison files to those who have suitably high resolution gear, and NOT through a network of any kind, or a server. The differences are fairly subtle, but through better than average gear they are very obvious.The differences are even more obvious with 24/96 .wav files, such as "Claire Martin-Too Darn Hot" from a Linn .flac DL conversion. Headphone listening is also an advantage.

Among the CA members who did hear clear differences between .wav files with identical checksums originally , were Peter St. , Claudius, Jkeny and Silverlight.

 

SandyK

 

 

 

[PeterSt]

Ok, for just one post I'll try to be on the serious side of things.

 

That sounds like a logical fallacy to me (i.e., that a 25 kHz signal can be decomposed into lower-frequency components).

 

Mind you, this is in the context of pure sines.

However, while wgscott put this as a fallacy I'd expect some responses to this. But (and maybe I missed it), instead there's something like this :

 

Just curious, does anyone know of any ABX testing done to see if a human listener can differentiate a 15khz sinewave from a 15khz square wave? Would not all of the additional energy that squares off such a wave be at 30khz and above?

 

So first about the first quote above :

No, a 25 KHz SINE will not decompose into lower frequencies, because they just are not there.

So, we can't hear that, ever.

 

Or ?

 

Or do we maybe think that when smaller waves from wind on a larger surf won't influence that surf by any means ?

Aha. So, maybe we *will* perceive that pure 25KHz sine afterall. But then by the INFLUENCE it has on just other frequencies.

No, water waves are not the same as audio waves. But I think the effect will be the same (btw, long proven the other way around : take out bass and and (try to) see how high frequencies are influenced by it).

 

The second quote is far more interesting. It actually prooves how way unexperienced we all are. Why ? well because here too I see nobody respond, while everybody actually should. Well, at listening through, say, "decent systems". There is no, NO, (did I say no ?) way that this is not audible. However -and I don't want to start an OS vs NOS debate AGAIN)- it needs the right DAC to listen to. Ok, assuming DACs in order here.

 

It reminds me of a way long thread which ever emerged at (IIRC) Hydrogen about the udial phenomenon (6-7 years or so ago). Google for udial, and you'll find sufficiently enough of it.

I didn't look back, and maybe by now all is clear, but this is actually about the very same subject, while really exactly nobody could get what was actually going on. This, while at the time of reading it, me, myself and I could not get what was NOT going on ...

 

udial is about a composed tone of telephone dials (the 1 sounds differently from the 2 etc.), while in the mean time a sneaky 19.x KHz tone was in there too. The phone dials are rather soft, so people tended to turn up the volume to make them audible. And next ? next nothing much was audible, at least not from burned tweeters. Great.

But what happened ?

 

I tried it too. But I didn't try it *that* loud, because I was warned in advance. But also, I didn't like to play it loud because of the nasty tones all over. Which tones ? well, simply, the decomposed tones coming from the 19.xKHz SQUARE tone, which were quite very much audible. And quite some louder as the dials as well.

 

At the time I wondered how people could not perceive it, while I did so clearly. Some others perceived it too, but most did not. Nobody could get it.

At the end of the thread (reading took a day or so), I got it;

I was using a non oversampling DAC. Most did not. At least not at that time, when it was a kind of new-ish.

 

And today ? well, today we apparently need ABX-testing to see whether people can perceive 15KHz Square Tones. Yes, it really made me laugh.

Well, no worries. Although I never tried it, but even a 50KHz square wave will be audible. Just throughout the spectrum.

Also notice that a square wave -with a properly selected frequency- will have as much output (for level) towards DC (= 0HZ) as the original tone. Not because it has fundamentals there or so, but because frequencies can add up. Just try a slow square wave sweep, and you will see it.

 

To make it hopefully understandable ... heavy oversampling will turn squares into sines. This is how a 19KHz square will turn into a sine, next having all its energy in that one nice sine - that being way more high level than it ever was in digital in the first place (giving it some nice headroom) and that burning your tweeters ... because you'd never hear the sine.

But your audiophile's dog would be all over the place.

 

It is exactly here where I started to believe that OS couldn't be it. Synths wouldn't sound like intended, as would any square wave output (like the hairs of the bow of a violin). And this (once again ?) is how violins will start to sound like fl.... ah, never mind.

 

haha

Peter

 

[Miska]

heavy oversampling will turn squares into sines. This is how a 19KHz square will turn into a sine, next having all its energy in that one nice sine

 

OK, let's not talk about computer-generated tones where Nyquist sampling is violated. (like commonly is with today's digitally clipped pop-music)

 

But if you try to digitize 19 kHz square wave with 44.1 kHz ADC, you'll get 19 kHz sine or horrible aliasing. You won't be able to digitize 19 kHz square wave without heavy oversampling, and if you ever properly decimate it to 44.1 kHz it's gone... Digital scopes run sampling rates like 250 MHz or 1 GHz to accurately represent waveforms.

 

Regardless of oversampling or not, 44.1 kHz sampling rate cannot correctly represent 19 kHz square. Naturally NOS DAC could produce square waves, but only if it's not properly reconstructing the analog waveform... Any harmonic products above 22.05 kHz at 44.1 kHz sampling rate are called distortion.

 

(Traditionally put: at 19 kHz with 44.1 kHz sampling rate there's roughly one sample per cycle side, not enough to represent the difference between sine/saw/square waveforms)

 

Biggest problem with typical test tones is that they are mostly representing steady-state tones, usually even carefully ramping away any clear attack. Decay part is there usually due to carefully ramping the signal out.

 

Comparison is better done with complex waveforms where attack plays big role; like claves, glockenspiel, cymbals and jingling keys or similar. These also best show differences between different types of filters used to band-limit the signal.

 

 

 

[PeterSt]

All true.

 

(Traditionally put: at 19 kHz with 44.1 kHz sampling rate there's roughly one sample per cycle side, not enough to represent the difference between sine/saw/square waveforms)

 

Which is why -as I always say it- it needs a tradeoff between doing nothing (way bad) and doing way too much of oversampling (original sound is killed). Don't ask me where this tradeoff (of upsampling) precicely is, but something at the end of a square turning too much into a sine. Ehhm ... for audible frequencies, which ... ehh ... become audible when done wrongly (false harmonics blahblah).

 

Also, this is nothing we can anticipate upon (well, not that I know of), because (putting computer generated data aside), we wouldn't be able to see the difference between an original sine and an original square. Ok, it can be analysed allright (in real time) and put some heavy DSP on it to reconstruct it right, but it will be a too tough job in general (again, for me, or maybe for me at this moment :-). I guess some normal means of long enough filtering will come up with it automatically (hey, great), if only the lot wasn't "oversampled out" in the first place, if music were periodic, and when we'd like long ringing very much.

Or phase anomalies.

Which at least I don't.

 

... which doesn't make anything wrong in what you just said.

Peter

 

 

[wgscott]

"Regardless of oversampling or not, 44.1 kHz sampling rate cannot correctly represent 19 kHz square."

 

But what it can do is faithfully represent all the audible fundamentals that comprise the square wave.

 

(The trick here is to remember that the square wave is not a unique fundamental periodic function.)

 

[PeterSt]

Biggest problem with typical test tones is that they are mostly representing steady-state tones, usually even carefully ramping away any clear attack. Decay part is there usually due to carefully ramping the signal out.

 

Comparison is better done with complex waveforms where attack plays big role; like claves, glockenspiel, cymbals and jingling keys or similar. These also best show differences between different types of filters used to band-limit the signal.

 

You put this in later, and I can't agree more.

But what you won't agree upon is that you won't be able to show me any one good sounding situation from exactly this type of "music". That it IMO 100% equals what it is all about is interesting, or let's say, recognized that there the differences will be.

Btw, add bag pipes to it.

 

For a longer time I'm thinking it would be interesting and useful to meet. Maybe because there's so much to disagree upon, but all with good sense (as how I see it).

 

I also start to wonder what would be more difficult for education and a living : rocket science or stupid audio. It can't be *that* difficult to bring a rocket to Mars ... (I skipped the Moon ... too easy :-)

 

 

[PeterSt]

But what it can do is faithfully represent all the audible fundamentals that comprise the square wave.

 

Good one !

But I pass.

 

I guess the problem is in recognizing how the "shape" of the (square) wave is in the first place. Once that's done, your quote is in order.

 

I still pass, because I think (!) NOW it's time for those infinite number of small sines which won't fit in 22050 (someone mentioned it in this thread).

 

More coffee ...

 

[wgscott]

Even if you truncate the samping at 44.1 kHz, there are still an infinite number of possible sin functions between 0 and 22.05 kHz. What matters is whether you have sampled the one with the highest frequency sufficiently. All the others will be over-sampled.

 

If the cut-off of 44.1 kHz is not high enough, the distortions should become more audible as you go to higher and higher frequencies, in redbook recordings. Yet no one seems to make that claim. Why?

 

[Miska]

I guess the problem is in recognizing how the "shape" of the (square) wave is in the first place. Once that's done, your quote is in order.

 

Only way is to use high enough sampling rate to capture (ADC) all the needed harmonics without filtering anything out. For many cases even 192 kHz is not high enough.

 

If the information is missing from the captured data, it's missing and forever lost. No way to reconstruct something where the source doesn't tell what was there.

 

 

 

[Miska]

If the cut-off of 44.1 kHz is not high enough, the distortions should become more audible as you go to higher and higher frequencies, in redbook recordings. Yet no one seems to make that claim. Why?

 

High enough for what? It is enough for sines up to 22.05 kHz.

 

On ADC side, anything above 22.05 kHz is supposed to cut off, otherwise you have serious aliasing problem under 22.05 kHz. Definitely audible. Try doing non-oversampling ADC of 19 kHz square at 44.1 kHz without any antialias filtering (filterless NOS-ADC).

 

If you don't filter 44.1 kHz DA-conversion properly, you'll have serious distortions too, most likely also audible. That famous "digital glare" and edginess. Just play 19 kHz sine through filterless NOS DAC and measure level of tone at 25.1 kHz. Now the level ratio of that and 19 kHz tone is your distortion percentage.

 

These days you also have tons of distortions on RedBook recordings because the signal is pushed to digital clipping and produce all kinds of nasty square waves.

 

 

[PeterSt]

If the cut-off of 44.1 kHz is not high enough, the distortions should become more audible as you go to higher and higher frequencies, in redbook recordings. Yet no one seems to make that claim. Why?

 

I stared at this until my screen saver jumped in.

 

Can't you rephrase this somehow ?

E.g. who would you like to make exactly what claim ?

 

[PeterSt]

High enough for what? It is enough for sines up to 22.05 kHz.

 

No no no. It is enough to MAKE sines up to 22.05.

Whether they were that (sines) originally or not.

 

I think that is a difference.

 

[Miska]

No no no. It is enough to MAKE sines up to 22.05.

Whether they were that (sines) originally or not.

 

Yes, that's something we agree on. Waveform became limited to what is representable as set of sines under 22.05 kHz, already at ADC stage when antialiasing filter ripped everything else off...

 

If it didn't, there would be spurious tones (aliases) all over the 22.05 kHz frequency band.

 

First (3rd) harmonic making up 19 kHz square wave would appear at 12.9 kHz, if I did the maths correctly. For 15 kHz square, at 900 Hz.

 

 

[esldude]

Sorry, removed comment.

 

[sandyk]

esldude

Do you have good playback capabilities of 24/96 .wav files ?

If you do, send me a PM.

SandyK

 

[James1776]

"I was told there would be no math required on CA."

 

No, the only math free zone (and economics free zone) is under the dome in Sacramento.

 

Bada Boom

 

[Jeff In San Diego]

Actually, it's pretty simple math in this case.

Square wave is composed via additive synthesis of all of the odd harmonics.

 

[*]The first harmonic is the fundamental (19khz in your example)

[*]the next is third harmonic, which is simply 3 X 19khz or 57khz.

[*]for a decent approximation of a square wave we would also need the 5th harmonic, which is 5 X 19 or 95khz.

[*]The next one (7th harmonic - 133khz) would already be well above the Nyquist frequency for 24/192 recording

 

 

Of course, a square wave is a pretty nasty sounding waveform, and not something we sit around listening to.

 

Still, something like the top string on a violin probably requires quite a bit of above-20khz spectrum to make a Strad sound different from a grade-school learner violin. OTOH, maybe we form our opinion of the quality of the instrument from the lower strings and our brain fills in the rest?

 

Also, as has been mentioned here already, the transients that a waveform starts and to some degree ends with can have a steep up or down slope, and those require substantial harmonic energy to produce.

 

[Jeff In San Diego]

Maybe this points to another fallacy. We describe people who can more readily "hear" a difference made by some seemingly minor part of a sound system as "golden ears".

 

Could it be, instead, that they are "lazy brains" (no disrespect intended)? Perhaps they are folks who more quickly revolt when made to do the extra mental processing required to fill in the missing stuff.

 

What we describe as an ability to hear better might in fact be a difference in willingness to or awareness of the brain processes needed to fill in or otherwise "photoshop" the musical experience.

 

[esldude]

Yes, fairly good anyway. I have a Tact that will handle anything up to 192 khz. I have a computer for a source to it. The sound card in it will put out SPDIF up to the same frequency.

 

However, I looked at some postings, and did find where you talked about the various files. Quite respectfully, I simply disagree with your apparent premise. That bit perfect copies of digital files will sound different depending on the original source of those bit perfect copies. There simply is no rational explanation for that which jibes with known physical, electrical properties. It would have to be highly apparent for me to be slightly interested as it seems simply illogical.

 

As you have described something not all that easily heard, it would be my opinion some people are fooling themselves. If I heard extremely minor differences, under the circumstances I would conclude I was fooling myself. Were it as obvious as a 15 khz sine and square wave differ then I would with high curiosity investigate the results.

 

I have done some ripping, encoding and such. Never have I found losslessly compressed files or uncompressed files even from multiple copies to sound different. So I think your supposition leaves me thinking I would be wasting my time to try it.

 

Regards,

Esldude

 

[Miska]

What we describe as an ability to hear better might in fact be a difference in willingness to or awareness of the brain processes needed to fill in or otherwise "photoshop" the musical experience.

 

Brain can learn in terms of listening. You can teach it to hear things you haven't heard or recognized before.

 

I used to be a passive sonar guy, teaching people to hear and recognize things with passive sonar (underwater thingie), and it was kind of amazing how much people's listening skills can improve.