Audiophile objections to blind testing - an attempt from a layman

[colonelkernel8]

We have been over this countless times, the DA converter output is a function of finite binary code, is it not?

I believe SACD dsd64 is 1 bit with 2.8 million samples, where the signal simply goes up or down a step between samples, unlike PCM where the bit depth stores unique sets of values that pertain to specific voltage, and SACD in 5.6, 11.2 megahertz, or PCM with higher bit depth and sample rate will give smaller and smaller measurements.



in my case, I have passed three blind level matched tests with significance. The trick for me was to do a due diligence task while taking the test(to take my mind off the test itself), listen for 10-15 minutes until I noticed the 24 bit “effect”, as described, above, if the effect never appeared it was 16 bit, only take 1 test a day, simply listen to X. But the key I think here is you have to know what the signature is of 24 but over 16, then listen for the effect. If you are going in with negative bias towards 24 bit, then I am not surprised you fail.

Unless you're listening to music that is recorded, mixed, and mastered at like -90dB (so you need to absolutely crank the volume), 24 bit for playback is simply useless. Your EARS have an SNR of about 70dB (at best, assuming normal, healthy hearing) given their own internal noise from things like nerves and blood flow (which you can HEAR that if you go in a very good anechoic chamber). 96dB > ~70dB. Recording in 24 bit makes sense so you can adjust levels/compression later on with less risk of quantization errors/information loss, but with playback, 16 bit is more than enough, period.

 

[2leftears]

would you make the claim as valid that 2 red wines taste the same or two fragrances smell the same since you failed a blind test?

No, but I would give the claim credence if hundreds of other people consistently 'failed' the same properly conducted blind test. But you are comparing apples with oranges. The way in which and time scales on which both taste and olfactory senses dull and recover are quite different from hearing, and it is really not a helpful analogy here.

Do you not first have to establish blind listening tests as effective for the means of what you are testing, digital audio formats, etc. and what thresholds are?

They have. And they don't have to be designed differently for testing digital audio formats or analog audio formats. Sound is sound and audio listening tests are audio listening tests. Of course you can mess up the design of the test in many other ways by e.g. not judiciously isolating the 16/24 bit factor from different mixing/mastering, but that's obvious.

 

[gregorio]

We have been over this countless times, the DA converter output is a function of finite binary code, is it not?

Shocking isn’t it, we’ve been over this countless times and yet you still don’t understand how digital audio works and apparently haven’t bothered to go and find out! The output of a DAC is not only a function of binary code, it also requires a reconstruction filter. However, your question does not address any of the points I made or the question put to you and is irrelevant given that I already stated the data would be more accurate. The point you’re missing, apparently deliberately, is that regardless of bit depth the accuracy of the output is effectively perfect as proven by Shannon, but your question indicates you have no idea what that “function of finite binary code” actually does or results in!

I believe SACD dsd64 is 1 bit with 2.8 million samples, where the signal simply goes up or down a step between samples, unlike PCM where the bit depth stores unique sets of values that pertain to specific voltage, and SACD in 5.6, 11.2 megahertz, or PCM with higher bit depth and sample rate will give smaller and smaller measurements.

Exactly, SACD only effectively provides an up or down step, while PCM gives orders of magnitude more accuracy by actually having values for each up/down step. So, why doesn’t SACD produce an output signal orders of magnitude less accurate than say 16bit PCM? You seem to be arguing against yourself here!

Can you provide peer reviewed research showing that the same set of cognitive functions are deployed when taking blind listening tests versus casually listening to music?

Ideal, you’re requesting peer reviewed research. As the “Burden of Proof” is yours because you’re the one claiming there are different cognitive functions, then you must do what you’ve requested or you will have demonstrated you’re a hypocrite. However, as already stated and you’re deflecting from, it’s irrelevant because even if this claim of yours is true and there is a “different set of cognitive functions”, it does NOT support your argument unless you provide reliable evidence that your supposed “different set of cognitive functions” actually produce more accurate and reliable results.

in my case, I have passed three blind level matched tests with significance. The trick for me was to do a due diligence task while taking the test(to take my mind off the test itself), listen for 10-15 minutes until I noticed the 24 bit “effect”, as described, above …

Again, it is trivially easy to pass a level matched ABX with 100% significance between 16 and 24bit, just not at reasonable listening levels. Secondly, if you’re doing something else and not concentrating on listening during a listening test, that’s the opposite of “due diligence”! Lastly and also again, the “24bit effect you described above” is NOT an effect of 24bit. The effect you described above is an effect of inaccurate level matching not of the difference with 24bit!
Patient: I’ve got a broken leg. Doctor: How do you know you have a broken leg. Patient: Because I’ve got a sore throat. Doctor: Then you’ve effectively proven you don’t have a broken leg because a sore throat is a symptom of something like a common cold, not a symptom of a broken leg!

G

 

[eq1849]

Unless you're listening to music that is recorded, mixed, and mastered at like -90dB (so you need to absolutely crank the volume), 24 bit for playback is simply useless. Your EARS have an SNR of about 70dB (at best, assuming normal, healthy hearing) given their own internal noise from things like nerves and blood flow (which you can HEAR that if you go in a very good anechoic chamber). 96dB > ~70dB. Recording in 24 bit makes sense so you can adjust levels/compression later on with less risk of quantization errors/information loss, but with playback, 16 bit is more than enough, period.

The thing is people focus on Dynamic Range of the bit depth, but ignore the voltage is more accurately quantized with a larger bit depth.

There are 256 quantization steps between each of the 16 bit integers in 24 bit.

The voltage is more accurately recorded with higher bit depth, it’s not just DR. And the range is more accurate DR is across a wider band of the signals amplitude.

Shocking isn’t it, we’ve been over this countless times and yet you still don’t understand how digital audio works and apparently haven’t bothered to go and find out! The output of a DAC is not only a function of binary code, it also requires a reconstruction filter. However, your question does not address any of the points I made or the question put to you and is irrelevant given that I already stated the data would be more accurate. The point you’re missing, apparently deliberately, is that regardless of bit depth the accuracy of the output is effectively perfect as proven by Shannon, but your question indicates you have no idea what that “function of finite binary code” actually does or results in!

The reconstruction filter, too, is built on finite binary code, no?

Exactly, SACD only effectively provides an up or down step, while PCM gives orders of magnitude more accuracy by actually having values for each up/down step. So, why doesn’t SACD produce an output signal orders of magnitude less accurate than say 16bit PCM? You seem to be arguing against yourself here!

Now let’s flips this argument around.

The theorem says 2X the highest frequency should be the rate.

Yet DSD samples millions of times per second.

Now WHY would it need to do that IF 1 bit is enough. Why not simply do 1 bit at 44.1khz?

Ideal, you’re requesting peer reviewed research. As the “Burden of Proof” is yours because you’re the one claiming there are different cognitive functions, then you must do what you’ve requested or you will have demonstrated you’re a hypocrite. However, as already stated and you’re deflecting from, it’s irrelevant because even if this claim of yours is true and there is a “different set of cognitive functions”, it does NOT support your argument unless you provide reliable evidence that your supposed “different set of cognitive functions” actually produce more accurate and reliable results.

The burden is on you as you have prescribed a method(blind listening tests), yet I don’t think you have offered peer reviewed evidence, that I’ve seen, that the method you advocate for has been validated as effective, determined thresholds, etc. for the purposes you prescribe(differentiation of amps, formats,etc.).

If you have the research feel free to post it.

 

[colonelkernel8]

The thing is people focus on Dynamic Range of the bit depth, but ignore the voltage is more accurately quantized with a larger bit depth.

There are 256 quantization steps between each of the 16 bit integers in 24 bit.

The voltage is more accurately recorded with higher bit depth, it’s not just DR. And the range is more accurate DR is across a wider band of the signals amplitude.



The reconstruction filter, too, is built on finite binary code, no?



Now let’s flips this argument around.

The theorem says 2X the highest frequency should be the rate.

Yet DSD samples millions of times per second.

Now WHY would it need to do that IF 1 bit is enough. Why not simply do 1 bit at 44.1khz?




The burden is on you as you have prescribed a method(blind listening tests), yet I don’t think you have offered peer reviewed evidence, that I’ve seen, that the method you advocate for has been validated as effective, determined thresholds, etc. for the purposes you prescribe(differentiation of amps, formats,etc.).

If you have the research feel free to post it.

Yeah, arguing with you is a waste of time. The bad faith/Dunning-Kruger is palpable.

 

[sander99]

The reconstruction filter, too, is built on finite binary code, no?

There is always a final analog stage in the reconstruction filtering. (Except in some effectively defective designs.)
In the case of a non-oversampling DAC the reconstruction filter is purely analog, and removes frequencies above half the sampling frequency.
In the case of a oversampling DAC there is a digital part and an analog part. The digital part removes the frequencies above half the sampling frequency of the original non-oversampled input signal in the new oversampled digital signal. The analog part removes the frequencies above half the sampling frequency of the new oversampled signal (and can be a very relaxed filter because typically there are several octaves space between 20 kHz and half the new sampling frequency).

 

[castleofargh]

Would you entertain that a given set of cognitive functions are possibly engaged when TAKING A TEST for format differentiation purposes and that a possibly different set of cognitive functions are possibly engaged when casually LISTENING TO MUSIC for enjoyment purposes?

Yes, that's why a listening test is a listening test, and casual listening isn't. Casual listening has much more than sound involved into creating our conscious experience of sound. You're correct about that, and we all agree, except some delusional people thinking themselves free of psychological biases. For example, when reading papers and books on sound localization, I often find vision listed as a relevant variable. Because vision is more precise for knowing where something is in space than the ears, and probably because our brain just has a bigger area dedicated to sight. The main idea for the relation simply is that the audio cues for localization are likely inferred/calibrated from how they correlate with the visual ones.
But it goes deeper than that. I have mentioned several times how when I listen to virtual speakers with headphones, seeing a pair of speakers in my field of view changes where I imagine the sound sources. I tend to anchor the virtual speaker to the real one I see that isn't ON and outputs no sound(I know that and still it has some impact). Until I move my head, if I don't also have head tracking the imaging breaks down for me. So clearly a multivariable and multisensory thing is involved here. And while my example is a little special, the impact of visual cues on sound localization is accepted and demonstrated.
Would any of that render me unable to tell in when 2 sounds are different in a blind test? Nope, that's irrelevant because I'd only test for sound at that time, so obviously I wouldn't see true or false indications of a possible sound source. And it's the same for any of those extra variables, we wish to remove them or control them because we agree that they could fool us into thinking we're feeling like we're "hearing" a change that would actually be, in this example, a visual change. Like maybe playing the same audio but moving the speakers that played no sound to a different place in the room. That won't affect me if I can't see said speakers. I won't get tricked into imagine a change in position from a cue I don't get, instead I'll have to really focus on the sound I really can hear and how where I imagine that, or if it's different from the other sample(depending on the question, maybe we ask to locate the sound, maybe we ask if the listener can tell which one is the same as sample X?).

I also read a bunch on psychological biases, obviously, and in particular, what seem to interest you(yeah for any recognition of bias!), multisensory interactions. I remember telling someone a while ago how for people working on it, those interactions are more and more considered the rule instead of the exception. Taste and flavor influenced by audio and visual information(not that hard to imagine. Perception of temperature influenced by sight, smell and taste(harder to imagine). Our local favorite, sight influencing sound with the McGurk effect. The other way around, sound influencing sight with the double flash illusion. The influence of vision on our sense of balance... You can find papers with one or more experiments on each of those, because it's after reading them that I came to know of those influences. I might still have a link or a PDF for some of those if someone cares and can't seem to find anything on his own.
And then we have some more specific interactions and actual battles for dominance when the information isn't coherent between senses, variables or areas of the brain, like with the also incredibly famous Stroop effect:

You have to say the color(the "ink's" color) for each word, not the written word. One column at a time. And on the third one, you should slow down because our brain can't help but read the word and notice it's a color while our task involves colors, so to the brain it's relevant under the circumstances(or something, we know it happens, not always why or how). And apparently we can't completely turn that off, it just ends up getting in the way as soon as the color and written color say something different. That's when everything starts taking massively more time. And ironically, wasting time and fighting contradictions is exactly the type of things the brain usually seeks to avoid by using the shortcuts and simplification that result in psychological biases and errors in judgement. Nothing is ever simple.

The first time I saw the Stroop effect was on a TV program where they had adults timed on that task and discussing the phenomenon. Then they got a kid who learned his colors but had yet to learn reading. Of course, he got no conflict from what the words spelled, so he completely humiliated the adults on the last column.




Now, none of that validates your many cases of false logic in support of sighted impressions against controlled test, or about the audibility of more than 16bit, or all the stuff in between where, I tell you, we all can see when logic leaves the room. I either can't be bothered to point it out because of how silly it is, or someone already did a good enough job, but not replying doesn't mean you get away with it.
It's fine to only care about the entire sighted experience. After all, it's how we'll get to enjoy music on a daily basis. It makes sense to care about that more than the results of some blind test. But obviously, pretending that the feelings of sound in casual impression correctly describe sound and only sound, that's silly. You yourself make the case of extra senses and variables affecting our impressions of sound, so who would you at the same time argue that said impression of sound tells you what you're really hearing? You can't have the cake and eat it too.

Then trying to argue that more bits make a more accurate signal, and later on, when it's inevitably pointed out that in practice all you can change with the extra bits is the signal(more likely noise), quieter than about -90dB, you take the sound audibility goalpost and move it to another planet. Now what matters is voltage, and trying to score half a point with straw man arguments and random statements, some correct, but sadly even those are entirely irrelevant to the matter of control testing and audibility of a sound, while the rest is... nearly entertaining. I mean, once you started to argue about DSD, you couldn't stop reading, great comedy, entirely irrelevant to blind testing and the audibility of bits beyond 16. And BTW, modern DACs convert the signal and nothing looks like PCM when it effectively gets reconstructed. It's much more like DSD in practice, fewer effective bits and massive sample rate.

Anyway, back to the topic, the idea that fewer controls make for more accurate results is pretty silly, no matter how we try to spin it.
The idea that a non test is more accurate than a test, certainly is a challenging one, and it would be nice to have some way to test that idea. Oh, wait...

 

[eq1849]

Yeah, arguing with you is a waste of time. The bad faith/Dunning-Kruger is palpable.

Ad hominem.

Yes, that's why a listening test is a listening test, and casual listening isn't. Casual listening has much more than sound involved into creating our conscious experience of sound. You're correct about that, and we all agree, except some delusional people thinking themselves free of psychological biases. For example, when reading papers and books on sound localization, I often find vision listed as a relevant variable. Because vision is more precise for knowing where something is in space than the ears, and probably because our brain just has a bigger area dedicated to sight. The main idea for the relation simply is that the audio cues for localization are likely inferred/calibrated from how they correlate with the visual ones.
But it goes deeper than that. I have mentioned several times how when I listen to virtual speakers with headphones, seeing a pair of speakers in my field of view changes where I imagine the sound sources. I tend to anchor the virtual speaker to the real one I see that isn't ON and outputs no sound(I know that and still it has some impact). Until I move my head, if I don't also have head tracking the imaging breaks down for me. So clearly a multivariable and multisensory thing is involved here. And while my example is a little special, the impact of visual cues on sound localization is accepted and demonstrated.
Would any of that render me unable to tell in when 2 sounds are different in a blind test? Nope, that's irrelevant because I'd only test for sound at that time, so obviously I wouldn't see true or false indications of a possible sound source. And it's the same for any of those extra variables, we wish to remove them or control them because we agree that they could fool us into thinking we're feeling like we're "hearing" a change that would actually be, in this example, a visual change. Like maybe playing the same audio but moving the speakers that played no sound to a different place in the room. That won't affect me if I can't see said speakers. I won't get tricked into imagine a change in position from a cue I don't get, instead I'll have to really focus on the sound I really can hear and how where I imagine that, or if it's different from the other sample(depending on the question, maybe we ask to locate the sound, maybe we ask if the listener can tell which one is the same as sample X?).

We are stuck in this never ending loop of a person claiming all soda tastes the same, and that he can prove it by failing another taste test. The other says sodas do taste differently, just not when gulping them blindly back to back, but rather drinking in isolation.

 

[VNandor]

The thing is people focus on Dynamic Range of the bit depth, but ignore the voltage is more accurately quantized with a larger bit depth.

There are 256 quantization steps between each of the 16 bit integers in 24 bit.

The voltage is more accurately recorded with higher bit depth, it’s not just DR. And the range is more accurate DR is across a wider band of the signals amplitude.

You are wrong and people already tried to explain why a higher bit depth won't yield a "more accurate" recording after a certain point.
I will link this one last time although you always ignored it in the past. Maybe read and digest it this time:

"In an analog system, the signal is continuous, but in a PCM digital system, the amplitude of the signal out of the digital system is limited to one of a set of fixed values or numbers. This process is called quantization. Each coded value is a discrete step... if a signal is quantized without using dither, there will be quantization distortion related to the original input signal... In order to prevent this, the signal is "dithered", a process that mathematically removes the harmonics or other highly undesirable distortions entirely, and that replaces it with a constant, fixed noise level."

At least pretend you try to understand digital audio before you keep posting your misconceptions in this forum. I'm not expecting you to understand bit depth, quantization error, dither and how they relate to each other (as you never did or made an effort to do so), so I will give you one last example that you might be able to grasp before I accept that some people just never learn.

You have a device that can be set to output any voltage between 0 and 9 with perfect accuracy. This device is completely ideal, if you set it to output 5V, it will generate exactly that, not something between 4.9V and 5.1V. It's 5.000... volts to infinity and beyond (like a perfect microphone combined with a perfect sound source).

You also have an ideal multimeter not bound by physics that can measure voltage perfectly. However, it still have to show the measured voltage in a typical 8 segment LCD display, effectively quantizing the result. Let's say you set the ideal voltage generator to output exactly 5.4321V. How many digits (bits/steps) would you need for that if you weren't allowed to lose precision? You would need to quantize to 5 digits (0.0001 steps) because 5.4321 fits on a 5 digit display. Having more digits would not make the result more accurate as 5.432100 is still the same.

What happens if you pick a trickier output voltage like the square root of 2 or something like pi? The voltage measurement before the display is still perfect but unless you want to use an ...extremely large LCD display, you will have to settle for a finite number of digits and some precision will be lost due that, because the display will quantize the result to some digits. In this case, adding more digits would always make the result more and more accurate.

However, voltage measurements are bound by physics, so let's say you can measure the voltage with a certain amount of precision, like +/- 0.1%.
So if you set the perfect voltage generator to output pi volts (3.14159265359 and so on), the measured voltage can be anywhere between ~3.1385 and ~3.1447.
What do you think, how many of the digits are meaningful in this case, how many digits do you need to not lose the 0.2% precision? The value of the 10th digit is about 0.00000029% compared to pi (I might be off by a 1-2 zeros, it does not matter at all). This combined with the fact that the measurement's precision itself is limited to 0.2% makes displaying anything beyond the fifth or so digits completely pointless. The reading is not more accurate, the numbers will be completely random after a certain point (called "noise" in certain contexts) because the measurement already lacks precision. The only way to make the result more precise would be to increase the measurement's precision itself from +/- 0.1% to something better, displaying more numbers won't help.

The same rules apply when recording (measuring) a signal and recreating (generating) a signal. Both of these steps have a finite precision. Depending on how precisely these steps are done, more bits won't add anything of value because at the recording side, it just encodes random fluctuations coming from the studio, mic preamps even from the ADC itself (noise). Likewise at the generating side, it doesn't matter if the bits tell the DAC to create 1000.001mV if the DAC has a noise floor of 0.1mV, then everything after the first digit of the decimal point will be random. On top of that, the driver itself might not even respond to such a small change, and if it did, the tiny movement in air that it might create should reach the ear instead of getting obliterated by some other random sound coming from PC fans, fridges, passing cars, even people. 24bit is pointless for listening because all in all, neither step comes even close to utilizing the last couple of bits.

 

[castleofargh]

We are stuck in this never ending loop of a person claiming all soda tastes the same, and that he can prove it by failing another taste test. The other says sodas do taste differently, just not when gulping them blindly back to back, but rather drinking in isolation.

Who cares what one believes he can do when nothing can verify his claim? If you have the bottle you picked at the store in front of you, how is that an experience strictly about taste? You know, before you touch a glass, what to expect and what the answer is. Your mistake is to keep wanting to draw conclusions about taste from it(or that logic moved to sighted listening).
Experiments are sorted with dependent, independent, and controlled variables. The more uncontrolled variables you introduce, the less conclusive your experiment is. That's a simple, well understood approach that does not deserve to be ignored just when the results don't follow your guts. You want to know if changing the soda(independent variable), has an impact on how it tastes to you(the dependent variable), then we have to set up a test where you don't know which soda you're given, and we have to remove anything that could give you that information while tasting, plus anything that is susceptible of affecting your judgement or the actual taste(controlled variables) beside the independent variable(swapping the sodas). Adding salt is a big no no. We could consider giving both drinks in a bottle of Pepsi, but that would be an experimental mistake because:
-How much of the carbonated gas will be loss while transferring the Coca-Cola? How much effort would be needed to make sure that bottle isn't contaminated by remaining Pepsi or water or any cleaning solution if one was used? How much psychological influence would that bottle have on the subject? He might be able to tell the sodas apart in isolation, but now because he's seeing the bottle, a part of his experience isn't coming from taste but from his brain's expectation of the Pepsi taste, making him predict and experience a more Pepsi "taste". That bottle could be considered priming for the brain. So we probably have to get rid of the bottles entirely and give you glasses with both sodas.
Some choices will be made for that test, and because of those choices, the result might be altered. And each time this happens, the fix is to do our best to control whichever variables we didn't control enough. It could be a statistical impact because after Y successive tasting your taste buds are somewhat numb. And maybe different experiments with a controlled, but different number of successive trials would help reveal that issue and suggest more tests to narrow down the optimal number. Maybe the difference between experiments was simply that one test always started with Pepsi while the other test always started with Coca-Cola. If it is found to have an impact, the trials would need to be more spread out in time and the initial drink randomized.
So on and so forth, many things will become an influence for the test asking if we can tell those drinks apart by taste. There is a paper suggesting that in general, we can achieve better discrimination in an ABX test if we go A, X, B instead of A, B, X. There have been trials dismissed because a letter or a color was put on the samples to identify them and that alone became a psychological bias.

But throughout trying to mitigate or solve those new-found variables, one simple concept remains. It's the non-controlled variables that can mess up the result of a test. If you have a different opinion, you better come with more than half logic and opportunistic analogies, because the entire community of researchers disagrees with you for good reasons supported by long lists of experiments.
Casual listening is a nest of uncontrolled variables and variables, giving you hints to get the answer you want in advance and throughout the non test. From the standpoint of experimentation and answering a clear question with independent variable(24bit becoming 16), and dependent variable(audible difference), a sighted impression is not just a bad test full of uncontrolled variables, it's not even a test. You try to give meaning to that anyway because you want your feelings from it to be right, but that's a terrible justification.

I'd be tempted to say that I can tell when it's Pepsi, because I've had a few times when I asked for Coca-Cola, got presented with a glass and no bottle(maybe they serve it like beer, more likely they use a big bottle to save money, maybe they only have Pepsi and hope you won't notice or complain). And sometimes I felt like the sort of aspartame taste(how I call it, not sure if it has anything to do with actual aspartame, it's a sweetener thing IMO) gave Pepsi away. Now when I call the waiter out on it he admitted it. Done deal, I can tell, /me so stronk!. Right? Well, maybe, maybe not. I never got a statistically significant amount of that happening. And again, we're back to uncontrolled variables. Maybe I got served Pepsi several times and did not notice, so I only remember the times I did, but what percentage is that? Maybe the waiters couldn't care less, they have one of those "the client is king" policy, so they take the stuff away, do nothing, bring it back in the original coke bottle that I mistook for Pepsi?
Until those variables are controlled, my experience isn't conclusive. I sure can believe anything I want and build up my little cult with people of similar tasting ability, or testing inability, that agree with my opinion. Still not worth in total as much as one better controlled experiment.
And yes, that obviously applies to tests on audibility and any test worth its salt.

It's complete nonsense to try and defend the conclusiveness of a sighted casual impression. You can have all the opinions in the world, they won't turn into anything else because you have no mean to even know for a fact when you're right or not.

And I think I'm done with this. 2 long posts to mostly state the obvious and easy to google, psychology and experimental requirements.

 

[danadam]

The thing is people focus on Dynamic Range of the bit depth, but ignore the voltage is more accurately quantized with a larger bit depth.

No one ignores it because it is the same thing. The outcome of this "lesser accuracy" (and the only outcome) is higher noise and therefore less dynamic range.

 

[gregorio]

The thing is people focus on Dynamic Range of the bit depth, but ignore the voltage is more accurately quantized with a larger bit depth.
The voltage is more accurately recorded with higher bit depth, it’s not just DR. And the range is more accurate DR is across a wider band of the signals amplitude.

Thank you, I can’t think of a more perfect confirmation of my assertion that “your question indicates you have no idea what that “function of finite binary code” actually does or results in”, priceless! A “function” of that finite binary code that you clearly do not know exists is called a “dithering quantiser”: There is always some amount of error when discretely quantising a continuous signal, more bits produce a smaller amount of error (more accuracy) but still there is always some (quantisation) error. A required part of the process is therefore a “dithering quantiser”, which effectively converts the error to a perfectly random signal (white noise), the result being an effectively perfectly accurate signal plus white noise. Therefore, with fewer bits we have a perfectly accurate signal with more noise, while with more bits we have a perfectly accurate signal with less noise. So in BOTH cases we have effectively a perfectly accurate signal, the ONLY difference is the amount of noise and of course, noise defines the “noise floor” and therefore the dynamic range.

Hence why your assertion is utter nonsense! We are focusing on dynamic range precisely because we ARE considering “the voltage is more accurately quantised with a larger bit depth”, which is the exact opposite of your false assertion the we’re ignoring it! Not to mention that your assertion is utter nonsense just purely on simple logical grounds; I have stated more than once in responses to you that yes, the data/quantisation would be more accurate with more bits and obviously, I can’t be stating something repeatedly and ignoring it at the same time!

The reconstruction filter, too, is built on finite binary code, no?

Already answered by @sander99

The theorem says 2X the highest frequency should be the rate.
Yet DSD samples millions of times per second.
Now WHY would it need to do that IF 1 bit is enough. Why not simply do 1 bit at 44.1khz?

That’s good, you’re asking questions rather than just spouting nonsense, although unfortunately, I have the feeling you’re just being facetious but I’ll take the chance you’re actually being genuine and explain it anyway:

The Sampling Theorem does not state that 2x the highest frequency should be the rate, it states that the sampling frequency must be more than 2x the highest audio frequency. The reason that 1bit at 44.1kHz is not enough is because of that function of binary code and quantising that you are ignorant of; the “dithering quantiser” produces a perfectly accurate signal plus noise, the fewer bits the more noise. This (dither) noise can be dealt with by spreading it out and moving it into areas of the audio spectrum where we can’t hear it, this is called “noise shaped dither” and has been employed since the late 1980’s. With only one bit there is consequently a huge amount of noise but a sampling rate of 44.1kHz only provides an audio frequency band up to 22.05kHz, so there’s not much space to move and spread this noise out without impacting the audible freqs. Hence why much higher sample rates are required with only 1bit, a sample rate of say 2.8MHz (SACD) provides a frequency bandwidth up to 1.4MHz, plenty of space to move and spread out such a huge amount of noise. This is why with SACD/DSD we always see a significant increase in noise above about ~25kHz.

The burden is on you as you have prescribed a method(blind listening tests), yet I don’t think you have offered peer reviewed evidence, that I’ve seen, that the method you advocate for has been validated as effective, determined thresholds, etc. for the purposes you prescribe(differentiation of amps, formats,etc.).

If you have the research feel free to post it.

Common sense dictates that concentrating/focusing on something provides more discrimination ability than not doing so, what would the point of it be otherwise? You think maybe we concentrate our attention on something in order to avoid noticing details about it?

However, you’ve asked for specific peer evidence of specific things, such as 1. A superior testing methodology and 2. That short term listening is superior:
1. There’s masses of reliable evidence that double blind tests are more accurate and reliable, going back over a century. In fact, it’s so standard/well established science we teach it to school children. Nevertheless, you’re asking specifically about listening tests for differentiation, presumably unaware that the same principles apply, so, try: Standardising Auditory Tests - Munson and Gardener, JASA, 1950.
2. There’s also a huge wealth of science regarding short term/echoic memory, dating back over a century but really taking off in the 1960’s and ‘70’s, why don’t you just look it up on Wikipedia? Nevertheless, here again is one peer reviewed paper (amongst many): Attentional strategies and short-term memory in dichotic listening - Bryden, Cognitive Psychology, 1971.

Now it’s your turn. Provide peer reviewed evidence that long term casual listening provides more accurate and reliable results than controlled DBT/ABX listening tests. If you don’t, you will have demonstrated that you’re both a hypocrite AND just spouting BS!

We are stuck in this never ending loop of a person claiming all soda tastes the same, and that he can prove it by failing another taste test. The other says sodas do taste differently, just not when gulping them blindly back to back, but rather drinking in isolation.

Firstly, no one is claiming “all soda tastes the same”, so stop with the strawman arguments! Secondly, yes, the other guy “says the sodas do taste differently”, the problem is, he can never reliably prove it and worse still, he will commonly say the sodas “do taste differently” even if he is unknowingly tasting exactly the same soda!

You seem to be deliberately missing the whole point of DBTs, even though it’s taught at school and is trivially easy to go and look up, and your apparent lack of understanding the basics of digital audio also seems deliberate, because it’s been explained to you more than once and again is relatively easy to research for yourself. So at this point, either we have to believe you have a serious lack of intellectual ability or, that you’re just trolling!

G

 

[eq1849]

No one ignores it because it is the same thing. The outcome of this "lesser accuracy" (and the only outcome) is higher noise and therefore less dynamic range.

And LESS RESOLUTION.

 

[danadam]

And LESS RESOLUTION.

Whatever you want to call it, the only outcome is higher level of broadband noise.

 

[eq1849]

Thank you, I can’t think of a more perfect confirmation of my assertion that “your question indicates you have no idea what that “function of finite binary code” actually does or results in”, priceless! A “function” of that finite binary code that you clearly do not know exists is called a “dithering quantiser”: There is always some amount of error when discretely quantising a continuous signal, more bits produce a smaller amount of error (more accuracy) but still there is always some (quantisation) error. A required part of the process is therefore a “dithering quantiser”, which effectively converts the error to a perfectly random signal (white noise), the result being an effectively perfectly accurate signal plus white noise. Therefore, with fewer bits we have a perfectly accurate signal with more noise, while with more bits we have a perfectly accurate signal with less noise. So in BOTH cases we have effectively a perfectly accurate signal, the ONLY difference is the amount of noise and of course, noise defines the “noise floor” and therefore the dynamic range.

Hence why your assertion is utter nonsense! We are focusing on dynamic range precisely because we ARE considering “the voltage is more accurately quantised with a larger bit depth”, which is the exact opposite of your false assertion the we’re ignoring it! Not to mention that your assertion is utter nonsense just purely on simple logical grounds; I have stated more than once in responses to you that yes, the data/quantisation would be more accurate with more bits and obviously, I can’t be stating something repeatedly and ignoring it at the same time!

Already answered by @sander99

That’s good, you’re asking questions rather than just spouting nonsense, although unfortunately, I have the feeling you’re just being facetious but I’ll take the chance you’re actually being genuine and explain it anyway:

The Sampling Theorem does not state that 2x the highest frequency should be the rate, it states that the sampling frequency must be more than 2x the highest audio frequency. The reason that 1bit at 44.1kHz is not enough is because of that function of binary code and quantising that you are ignorant of; the “dithering quantiser” produces a perfectly accurate signal plus noise, the fewer bits the more noise. This (dither) noise can be dealt with by spreading it out and moving it into areas of the audio spectrum where we can’t hear it, this is called “noise shaped dither” and has been employed since the late 1980’s. With only one bit there is consequently a huge amount of noise but a sampling rate of 44.1kHz only provides an audio frequency band up to 22.05kHz, so there’s not much space to move and spread this noise out without impacting the audible freqs. Hence why much higher sample rates are required with only 1bit, a sample rate of say 2.8MHz (SACD) provides a frequency bandwidth up to 1.4MHz, plenty of space to move and spread out such a huge amount of noise. This is why with SACD/DSD we always see a significant increase in noise above about ~25kHz.

I believe the 2.8 million samples are required to capture the wave accurately in DSD, as the next sample cannot immediately move across the range as in PCM, but a linear fashion, so the millions of samples are needed not just for noise purposes.

Show me a proof please that dither under the constraints of binary code is perfectly recovering the signal.

Common sense dictates that concentrating/focusing on something provides more discrimination ability than not doing so, what would the point of it be otherwise? You think maybe we concentrate our attention on something in order to avoid noticing details about it?

However, you’ve asked for specific peer evidence of specific things, such as 1. A superior testing methodology and 2. That short term listening is superior:
1. There’s masses of reliable evidence that double blind tests are more accurate and reliable, going back over a century. In fact, it’s so standard/well established science we teach it to school children. Nevertheless, you’re asking specifically about listening tests for differentiation, presumably unaware that the same principles apply, so, try: Standardising Auditory Tests - Munson and Gardener, JASA, 1950.
2. There’s also a huge wealth of science regarding short term/echoic memory, dating back over a century but really taking off in the 1960’s and ‘70’s, why don’t you just look it up on Wikipedia? Nevertheless, here again is one peer reviewed paper (amongst many): Attentional strategies and short-term memory in dichotic listening - Bryden, Cognitive Psychology, 1971.

Now it’s your turn. Provide peer reviewed evidence that long term casual listening provides more accurate and reliable results than controlled DBT/ABX listening tests. If you don’t, you will have demonstrated that you’re both a hypocrite AND just spouting BS!

https://pmc.ncbi.nlm.nih.gov/articles/PMC5285336/

the effects may be in the subconscience.

Firstly, no one is claiming “all soda tastes the same”, so stop with the strawman arguments! Secondly, yes, the other guy “says the sodas do taste differently”, the problem is, he can never reliably prove it and worse still, he will commonly say the sodas “do taste differently” even if he is unknowingly tasting exactly the same soda!

You seem to be deliberately missing the whole point of DBTs, even though it’s taught at school and is trivially easy to go and look up, and your apparent lack of understanding the basics of digital audio also seems deliberate, because it’s been explained to you more than once and again is relatively easy to research for yourself. So at this point, either we have to believe you have a serious lack of intellectual ability or, that you’re just trolling!

G

I don’t think someone failing a negatively biased test is proving too much Gregorio..