Testing audiophile claims and myths

[bigshot]

The subjective sound experience is the same with recordings as real life. That is the baseline, not the sound to be tested.

 

[Vamp898]

I always wonder, how these mystical/magical things end up in the recording anyway.

They are recorded using average copper cables and an ADC. So anything magical that would have been there is lost at the moment where the signal travels through cheap copper cables into an ADC where it turns into Data.

I do own one of the best measuring sound studio equipment that is available on the market and the biggest downside is, that it only has 4 Channels.

Why? Because everything that has more than 4 Channels (no matter the price) starts to measure worse again. And trust me, out of convenience, most sound studios, if not all, will sacrifice the measured sound quality for more channels.

So one of the best measuring 4 Channel Studio DI you can buy has 114db SNR

And i can promise you that even the best and well adjusted recording room on this planet doesn't have an Noise Floor that is low enough to even use those 114db SNR. And this is just the SNR, look at the THD+N of such devices and so on. Look at the impulse response of the best studio Microphones that sell for a few thousand, look at their frequency response.

The gear we use to record the stuff is worse than the gear we use to listen to it back in almost every way.

So if our knowledge is not good enough to understand how humans actually hear sound which gives room for snake oil, we do not understand it when we're recording either. So the room for the snake oil is gone because all mythical and magical components are already lost in the recording stage.

 

[XTF1]

“There is no hope or point in quantifying anything for guys who think of themselves as an infallible sensory machine with infinite range. No amount of research will validate that delusion.”

Should be pinned as a must-read / must-ask for anyone posting on the science forum…

 

[2leftears]

If we can measure the physical sound isn’t the rest of the audio experience down to perception and obviously just preferences ?

That perception will vary from person to person with the listening experience for some people being dramatically more influenced by stimuli beyond the sound pressure waves entering their ears than others ?

But doesn't therein lie the rub?

Science has come up with some metrics for sound (or, more relevant perhaps, audio signals), but they are far from a complete descriptor. E.g. if total harmonic distortion is zero, we do indeed know that the signal is replicated perfectly. But if it is non-zero, there are many different permutations of the original signal possible that give the same non-zero THD value, but they are all different when examined in detail. The sound signal metrics that we use suffer from multiplicity; thus we can have two audio components that quantify identical in terms of frequency response, THD, IMD, noise levels etc, but that do produce different signal errors when examined in more detail, and may therefore sound different. The scientific knowledge as to how those different signal errors my affect perception in different individuals is as yet incomplete.

We can set some stochastically determined boundaries on what should/should not be audible to the vast majority of people, but until we know how our sound/audio signal metrics are misaligned with respect to the way humans process auditory information, absolutist statement re. perceived fidelity of sound reproduction are suspect, especially if we are looking at statistics that are close to the boundary of what is scientifically deemed perceptible and what isn't. (and since 16/44.1 PCM seems to come up again and again: that topology is very close to the boundary of what is scientifically deemed perceptible, so close that any flaw in its implementation may well be audible to a good number of people).

 

[Vamp898]

But doesn't therein lie the rub?

Science has come up with some metrics for sound (or, more relevant perhaps, audio signals), but they are far from a complete descriptor. E.g. if total harmonic distortion is zero, we do indeed know that the signal is replicated perfectly. But if it is non-zero, there are many different permutations of the original signal possible that give the same non-zero THD value, but they are all different when examined in detail. The sound signal metrics that we use suffer from multiplicity; thus we can have two audio components that quantify identical in terms of frequency response, THD, IMD, noise levels etc, but that do produce different signal errors when examined in more detail, and may therefore sound different. The scientific knowledge as to how those different signal errors my affect perception in different individuals is as yet incomplete.

We can set some stochastically determined boundaries on what should/should not be audible to the vast majority of people, but until we know how our sound/audio signal metrics are misaligned with respect to the way humans process auditory information, absolutist statement re. perceived fidelity of sound reproduction are suspect, especially if we are looking at statistics that are close to the boundary of what is scientifically deemed perceptible and what isn't. (and since 16/44.1 PCM seems to come up again and again: that topology is very close to the boundary of what is scientifically deemed perceptible, so close that any flaw in its implementation may well be audible to a good number of people).

This is exactly what we do and we do know how it works. Where does this sudden "Humans don't know science" in all areas come from? We are able to build the JWST and place it in the L2 but we do not how electricity works when it comes to audio...

 

[2leftears]

This is exactly what we do and we do know how it works. Where does this sudden "Humans don't know science" in all areas come from? We are able to build the JWST and place it in the L2 but we do not how electricity works when it comes to audio...

Oh, we do know how electricity works when it comes to audio, we know it very well indeed.

But where do you get the impression that there is a "Humans don't know science" trend? I think the question is more one of "does science fully understand humans?"

E.g. for 100 years after Japanese scientists first identified umami, Western science thought with absolute certainty that it knew all about our sweet/sour/bitter/acid taste receptors until they finally realised the Japanese had been right all along about umami and glutamate receptors. Likewise, history is littered with examples of science denying the fact they don't know all there is to know yet, or even worse, discouraging the questioning of established scientific understanding.

The idea that sound signal quality/fidelity can be adequately described by frequency curve/THD/IMD/noise levels etc. is a sign of hubris I think. We humans can perceive more complex sound features than that, so two "perfect" measuring amps can still sound different to experienced ears.

The question is what do you do in this situation? Maybe admit furthering scientific knowledge/research is needed / write off people's experiences as subjective nonsense / or start selling snake oil products by inventing pseudo-science?

If someone tells me a hi-res audio file sounds better to their ears than a 16/44.1 audio file, I'm not going to call them deluded because I know full well that our scientific understanding of audio signal processing, and perhaps more important, our commercial and practical implementation of it, are far from perfect.

 

[FunkyBassMan]

But doesn't therein lie the rub?

Science has come up with some metrics for sound (or, more relevant perhaps, audio signals), but they are far from a complete descriptor. E.g. if total harmonic distortion is zero, we do indeed know that the signal is replicated perfectly. But if it is non-zero, there are many different permutations of the original signal possible that give the same non-zero THD value, but they are all different when examined in detail. The sound signal metrics that we use suffer from multiplicity; thus we can have two audio components that quantify identical in terms of frequency response, THD, IMD, noise levels etc, but that do produce different signal errors when examined in more detail, and may therefore sound different. The scientific knowledge as to how those different signal errors my affect perception in different individuals is as yet incomplete.

We can set some stochastically determined boundaries on what should/should not be audible to the vast majority of people, but until we know how our sound/audio signal metrics are misaligned with respect to the way humans process auditory information, absolutist statement re. perceived fidelity of sound reproduction are suspect, especially if we are looking at statistics that are close to the boundary of what is scientifically deemed perceptible and what isn't. (and since 16/44.1 PCM seems to come up again and again: that topology is very close to the boundary of what is scientifically deemed perceptible, so close that any flaw in its implementation may well be audible to a good number of people).

This is really fantastic. Thanks for it.

 

[FunkyBassMan]

The idea that sound signal quality/fidelity can be adequately described by frequency curve/THD/ID/noise levels etc. is a sign of hubris

Precisely.

 

[2leftears]

This is really fantastic. Thanks for it.

The whole "16/44.1 is sufficient" debate I have followed on and off for years, and a lot seems to get lost in translation between the people who understand the underlying mathematical theory, and those who just pick up on the headline statements. Many extend the ideas put forward in Nyquist/Shannon too far beyond what the theory was aiming to prove.

The gap between the theory and what is practically possible is that Nyquist/Shannon proves that in context of Fourier decomposition of the orginal signal versus its PCM sampled counterpart no relevant information has been lost, and that therefore perfect reconstruction of the original signal from the PCM data samples is indeed possible. However whilst it encompasses the Whittaker-Shannon interpolation formula for reconstruction it does not address more practical methods of reconstruction, and it also requires perfect band filtering in the frequency (Fourier) domain. The conventional sample-and hold DAC reconstruction with analogue low-pass filters in the time domain as found in early digital audio is theoretically flawed in this respect, although in practice it results in quite acceptable performance. Oversampling filters with multiple coefficient parameterisation improve matters quite a bit.

The 16/44.1 proponents (and TBH, it works perfectly well enough for me) should have a look at how the same theory panned out in the world of digital image capture. Mathematically the two are quite comparable (compare audio frequency vs. spacial frequency, our upper hearing limit vs. visual acuity, etc.) In practice it turns out the most practical solution to improve image quality in digital photography is to work with a higher pixel count than visual acuity would suggest as sufficient. This reduces the need for an aggressive image-softening anti-alias filtering in front of the digital sensor and post-capture software filtering processing (its audio equivalent is low-pass filtering of the audio prior to sampling, and post DAC filtering.) Likewise, the most practical way to improve audio quality is to increase the sampling rate to one above what Nyquist/Shannon stipulates as sufficient, e.g. 48kHz or 96kHz.

 

[FunkyBassMan]

The whole "16/44.1 is sufficient" debate I have followed on and off for years, and a lot seems to get lost in translation between the people who understand the underlying mathematical theory, and those who just pick up on the headline statements. Many extend the ideas put forward in Nyquist/Shannon too far beyond what the theory was aiming to prove.

The gap between the theory and what is practically possible is that Nyquist/Shannon proves that in context of Fourier decomposition of the orginal signal versus its PCM sampled counterpart no relevant information has been lost, and that therefore perfect reconstruction of the original signal from the PCM data samples is indeed possible. However whilst it encompasses the Whittaker-Shannon interpolation formula for reconstruction it does not address more practical methods of reconstruction, and it also requires perfect band filtering in the frequency (Fourier) domain. The conventional sample-and hold DAC reconstruction with analogue low-pass filters in the time domain as found in early digital audio is theoretically flawed in this respect, although in practice it results in quite acceptable performance. Oversampling filters with multiple coefficient parameterisation improve matters quite a bit.

The 16/44.1 proponents (and TBH, it works perfectly well enough for me) should have a look at how the same theory panned out in the world of digital image capture. Mathematically the two are quite comparable (compare audio frequency vs. spacial frequency, our upper hearing limit vs. visual acuity, etc.) In practice it turns out the most practical solution to improve image quality in digital photography is to work with a higher pixel count than visual acuity would suggest as sufficient. This reduces the need for an aggressive image-softening anti-alias filtering in front of the digital sensor and post-capture software filtering processing (its audio equivalent is low-pass filtering of the audio prior to sampling, and post DAC filtering.) Likewise, the most practical way to improve audio quality is to increase the sampling rate to one above what Nyquist/Shannon stipulates as sufficient, e.g. 48kHz or 96kHz.

Great. And presumably bit depth as well?

 

[2leftears]

Great. And presumably bit depth as well?

As long as the information density of the data stream is reasonably consistent, then in theory bit depth and bit rate/sampling rate are somewhat interchangeable using e.g. dithering and noise shaping technologies. After a fashion the 1-bit DSD very high bit rate topology is an extreme example of that (only 1 bit, but very high bit rate, enabling very similar if not better performance than 16/44.1). Technics had a half-way house MASH technology.

 

[FunkyBassMan]

As long as the information density of the data stream is reasonably consistent, then in theory bit depth and bit rate/sampling rate are somewhat interchangeable using e.g. dithering and noise shaping technologies. After a fashion the 1-bit DSD very high bit rate topology is an extreme example of that (only 1 bit, but very high bit rate, enabling very similar if not better performance than 16/44.1).

Good point.

 

[bigshot]

if total harmonic distortion is zero, we do indeed know that the signal is replicated perfectly.

Wrong. It doesn’t have to measure zero to sound perfect, it just needs to fall below the threshold of audibility, which with human ears is a long way from zero. I don’t think you’ve bothered to research what is audible and what isn’t, so you’re assuming that everything is audible. That just isn’t true. There is a point where error is too small to matter. If you don’t understand that, you have no way of predicting how something sounds.

If a piece of gear produces a signal that is audibly transparent, it sounds exactly the same as any other audibly transparent piece of gear, even if it measures different. Most amps and DACs that aren’t defective and are used the way the designer intended them to be used are audibly transparent.

Sound doesn’t have to measure perfectly to sound perfect. It just has to have less error than your own ears can perceive. Throw out your absolutism, and you’ll understand. There’s an element of OCD that makes audiophools think that sound has to be perfect. It doesn’t. It just has to be good enough. This same OCD is behind the argument, “We can’t know everything, so we can’t know anything.” Science does know how to make a great sounding stereo system. It’s not a mystery. Optimize the output of your transducers. That will have a positive effect on sound quality. Real world physical acoustics can have so many variables that it becomes difficult for the layman to predict, but you really can go to Walmart or Amazon and buy a digital player and be confident that the DAC is producing a signal that is audibly transparent. It isn’t rocket science. No need to over complicate it to the point that you lose sight of the goal… to play great sounding music in the home.

As we point out over and over again in this forum, images can not be used as analogies for sound. You listen to sound at reasonable listening levels, which presents a relatively fixed perspective. Images can be examined with a magnifying glass to examine details more closely, but you can’t turn up the volume of sound beyond a comfortable volume to examine details without incurring hearing damage. Sound has a relatively fixed optimal target resolution. Images don’t because they are viewed in a wider variety of ways than sound is listened to. Apples and oranges.

 

[bigshot]

This is really fantastic. Thanks for it.

Precisely.

Good point.

“me too”

 

[FunkyBassMan]

me too

Glad to hear it.