Why can I hear a difference between 32bit 384khz and 24bit 44.1khz?

[simplywyn]

He needs to check his Windows audio settings. (These will only apply when he does NOT use WASAPI output)

Quite possibly he needs to ,e.g., turn every Windows audio processing 'enhancement' to 'OFF', and e.g., configure his system correctly (Control Panel--> Sound-->Manage Audio Devices -->Playback-->[his device]--> Configure *and* Properties both have settings that may need changing )

Or he can just use WASAPI , and stop the ridiculous attempts at upsampling, unless there is some technical need for it.

Yea I just use WASAPI instead.

 

[pLudio]

Yea I just use WASAPI instead.

And make sure it is WASAPI in Exclusive mode and not Shared mode so you're not going through the Windows mixer.

 

[sq225917]

It's easy. Download a 384 file. Keep a copy and downsample it in Audacity to 16 44.

Then check their level is the same in Audacity and play them back into the dac at their native resolutions. If you still hear a difference then its down to the dac handling the files differently.

Because audibly the data is identical. Any difference is down to internal handling.

 

[Lambda]

It's easy. Download a 384 file. Keep a copy and downsample it in Audacity to 16 44.

It's not easy because there are many different down sampler, filter and dither algorithms.

 

[escape2]

It's easy. Download a 384 file. Keep a copy and downsample it in Audacity to 16 44.

The OP was doing it the other way, though. He was taking a source file of 24-bit / 44.1 kHz and upconverting it to 32-bit / 384 kHz. At least that's what I understood.

 

[Ennis]

You can hear the difference because there is a difference.

There's a harmonic content that needs to get recorded and played back with a higher sampling rate than 44,100/sec since there are hearable frequencies which superior harmonics are in the range of ultrasounds, and the presence of these affects their originating fundamental during playback (making it sound more natural and full). There's also a better reconstruction of the transient harmonics which are originated from the almost punctual intermodulations between different frequencies that are very much like sparse fragments of a sound waves but are very closely interrelated in timing and phase to the main fundamentals; and can be only captured properly with higher sampling rates than 44,10,0 samples per second.

(There are tons of softwares, theories and followers out there that seem to be convinced that the 20kHz frequency limit and the 2 samples per cycle is all about it.

There's also the fact that listening is like many other human abilities: there's people more or less gifted, and it needs to be trained. .. not everybody can listen the same and most people couldn't tell much of a difference like many people couldn't run a mile in X minutes).

How much any of this diminishes or not the enjoyment of music and recorded sound in general is just another story.

 

[SIY]

You can hear the difference because there is a difference.

There's a harmonic content that needs to get recorded and played back with a higher sampling rate than 44,100/sec since there are hearable frequencies which superior harmonics are in the range of ultrasounds, and the presence of these affects their originating fundamental during playback (making it sound more natural and full). There's also a better reconstruction of the transient harmonics which are originated from the almost punctual intermodulations between different frequencies that are very much like sparse fragments of a sound waves but are very closely interrelated in timing and phase to the main fundamentals; and can be only captured properly with higher sampling rates than 44,10,0 samples per second.

(There are tons of softwares, theories and followers out there that seem to be convinced that the 20kHz frequency limit and the 2 samples per cycle is all about it.

There's also the fact that listening is like many other human abilities: there's people more or less gifted, and it needs to be trained. .. not everybody can listen the same and most people couldn't tell much of a difference like many people couldn't run a mile in X minutes).

How much any of this diminishes or not the enjoyment of music and recorded sound in general is just another story.

So basically, you understand neither Fourier nor Shannon-Nyquist. But can certainly toss a word salad.

 

[tmtomh]

You can hear the difference because there is a difference.

There's a harmonic content that needs to get recorded and played back with a higher sampling rate than 44,100/sec since there are hearable frequencies which superior harmonics are in the range of ultrasounds, and the presence of these affects their originating fundamental during playback (making it sound more natural and full). There's also a better reconstruction of the transient harmonics which are originated from the almost punctual intermodulations between different frequencies that are very much like sparse fragments of a sound waves but are very closely interrelated in timing and phase to the main fundamentals; and can be only captured properly with higher sampling rates than 44,10,0 samples per second.

(There are tons of softwares, theories and followers out there that seem to be convinced that the 20kHz frequency limit and the 2 samples per cycle is all about it.

There's also the fact that listening is like many other human abilities: there's people more or less gifted, and it needs to be trained. .. not everybody can listen the same and most people couldn't tell much of a difference like many people couldn't run a mile in X minutes).

How much any of this diminishes or not the enjoyment of music and recorded sound in general is just another story.

No. Humans cannot hear ultrasonics, regardless of whether the ultrasonics are fundamental tones or harmonics of audible-range fundamentals.

Sounds - transient harmonics or otherwise - within the audible range are not "better reconstructed" with higher sample rates. If that were true, then every single digital recording ever made would suffer from massive differences in sound quality within the audible range, regardless of sample rate - no matter your sample rate, 100Hz gets sampled 10x as much as 1kHz and 100x as much as 10kHz. So every recording would have much "better reconstruction" of bass drums, bass guitar, and other low-frequency sounds than of female vocals, snares, hi-hats, and other high-frequency sounds.

Finally, on the question of timing, the timing accuracy of 44.1kHz digital sampling is more than enough to accommodate the timing accuracy of human hearing - in fact, a key principle of digital sampling theory, which makes digital recording possible at all, is that the frequency of sampling really has nothing to do with the perceived "speed" or "timing accuracy" of transients or any other sounds. The idea that a transient might be "too fast" to be captured by 44.1kHz sampling is quite literally nonsense: the only sound "too fast" for 44.1kHz is a sound whose frequency is above 22.05kHz. The only sound that can be played or recorded that is "faster" than that is a sound at a frequency higher than that. Sampling frequency is about the highest-frequency sound that can be recorded and reconstructed. It is not about timing accuracy.

 

[BDWoody]

How much any of this diminishes or not the enjoyment of music and recorded sound in general is just another story.

That's pretty much all it is. People seem to like stories.

 

[aandres_gm]

You can hear the difference because there is a difference.

There's a harmonic content that needs to get recorded and played back with a higher sampling rate than 44,100/sec since there are hearable frequencies which superior harmonics are in the range of ultrasounds, and the presence of these affects their originating fundamental during playback (making it sound more natural and full). There's also a better reconstruction of the transient harmonics which are originated from the almost punctual intermodulations between different frequencies that are very much like sparse fragments of a sound waves but are very closely interrelated in timing and phase to the main fundamentals; and can be only captured properly with higher sampling rates than 44,10,0 samples per second.

(There are tons of softwares, theories and followers out there that seem to be convinced that the 20kHz frequency limit and the 2 samples per cycle is all about it.

There's also the fact that listening is like many other human abilities: there's people more or less gifted, and it needs to be trained. .. not everybody can listen the same and most people couldn't tell much of a difference like many people couldn't run a mile in X minutes).

How much any of this diminishes or not the enjoyment of music and recorded sound in general is just another story.

Or you could just read the thread to realise this had nothing to go with transients, supersonic content or golden ears, but with the oversampling process.

 

[krabapple]

You can hear the difference because there is a difference.

There's a harmonic content that needs to get recorded and played back with a higher sampling rate than 44,100/sec since there are hearable frequencies which superior harmonics are in the range of ultrasounds, and the presence of these affects their originating fundamental during playback (making it sound more natural and full). There's also a better reconstruction of the transient harmonics which are originated from the almost punctual intermodulations between different frequencies that are very much like sparse fragments of a sound waves but are very closely interrelated in timing and phase to the main fundamentals; and can be only captured properly with higher sampling rates than 44,10,0 samples per second.

So let's review your claim:
Unheard frequencies cause effects in the audible range. But when we record and play back what's in the audible range, the effect is gone.

Getting kind of tired of *Audio Science* Review participants promoting such nonsense.

 

[Ennis]

So basically, you understand neither Fourier nor Shannon-Nyquist. But can certainly toss a word salad.

Basically, I am a certified sound engineer. And (basically) I do not permit myself to believe I know something just because it makes me feel good about myself. Out of the many scientists, engineers, experts, specialists... in the sound world, as in any theoric science, there are only a handful that have a vague idea of anything... the rest are just reproducing some numerical fictions they're being given to play with.

And you -SYI- to have the guts to come with such a comment , you should be just a hobbist (who'd be very pleased about himself for understanding Fourier Transform and Nyquist Theorem. And I don't know why you see what I am saying as opposed to either).

 

[voodooless]

Basically, I am a certified sound engineer.

Great! Than surely you can tell us what a sound engineer does with ultrasonics? And how do you verify that what you did to them has any effect?

 

[tmtomh]

Basically, I am a certified sound engineer. And (basically) I do not permit myself to believe I know something just because it makes me feel good about myself. Out of the many scientists, engineers, experts, specialists... in the sound world, as in any theoric science, there are only a handful that have a vague idea of anything... the rest are just reproducing some numerical fictions they're being given to play with.

And you -SYI- to have the guts to come with such a comment , you should be just a hobbist (who'd be very pleased about himself for understanding Fourier Transform and Nyquist Theorem. And I don't know why you see what I am saying as opposed to either).

What you are saying is opposed to Nyquist because you are claiming that sample rates above 2x the highest needed frequency make for "better reconstruction" of those frequencies. That is directly opposed to Nyquist, and if it were true, all digital recording - not to mention pretty much all signal transmission involving digital sampling - would either not work, require far more bandwidth than it does, or sound and measure markedly different than it does.

 

[SIY]

Basically, I am a certified sound engineer. And (basically) I do not permit myself to believe I know something just because it makes me feel good about myself. Out of the many scientists, engineers, experts, specialists... in the sound world, as in any theoric science, there are only a handful that have a vague idea of anything... the rest are just reproducing some numerical fictions they're being given to play with.

And you -SYI- to have the guts to come with such a comment , you should be just a hobbist (who'd be very pleased about himself for understanding Fourier Transform and Nyquist Theorem. And I don't know why you see what I am saying as opposed to either).

It would seem that you don’t know enough to be “opposed” to Shannon-Nyquist and Fourier. Likewise, you do not seem to understand what a theorem is.

 

[Rottmannash]

IME it is possible to get audible differences between files due both to the way the DAC handles the data it receives and also by there being a difference between the mastering of the files on offer.

It would seem to be good marketing practice for a vendor to make sure the file which costs more to be of a better sounding master, even if it is only 0.3dB louder, if clients are able to compare them themselves before purchase.

Each DAC will have its own way of dealing with an incoming file before it converts it, to suit the technology of chipset used if nothing else. That means different sample rate files will go through different manipulation pre-conversion and maybe some DACs are not transparent in the manipulation or use a different reconstruction filter for different sample rates.
I have certainly experiences sound change due to an early downsampling software converting a file to 16/44.1 from 16/48, for example and some reconstruction filters are audibly different to a classic one.

In my own evaluation of the audibility of "higher" res I used a good re-sampler to convert a 24/96 file to 16/44.1 then back again to 24/96. This meant that the DAC manipulated both files the same way, since it saw the same type of file but the second file had any of the data above 22.05kHz and lower than 16-bit (if there was any) removed. So I was certainly only listening for difference due to loss of data rather than other artefacts.
When I did that I could not hear any difference between the files.
FWIW
I did the Dr Aix on line investigation too, and couldn't hear any difference there either.

I did the Aix as well and scored horribly... to my chagrin.

 

[BDWoody]

And (basically) Out of the many scientists, engineers, experts, specialists... in the sound world, as in any theoric science, there are only a handful that have a vague idea of anything... the rest are just reproducing some numerical fictions they're being given to play with.

Which of those two groups are you in? We have quite a few in the first category about the place, so those in the second sometimes have a hard time here.

 

[Ennis]

So let's review your claim:
Unheard frequencies cause effects in the audible range. But when we record and play back what's in the audible range, the effect is gone.

Getting kind of tired of *Audio Science* Review participants promoting such nonsense.

A vinyl record, for example, can record up to 15kHz-17kHz. .. Because of the tip of the recording stylus is not small enough to make a clear 'drawing' of frequencies above this. But it does record harmonics of higher frequencies as -periodical- small notches (what they become when projected into the surface of a vinyl master) that when played back ad substance to the sound.

Tape does record these higher frequencies naturally and it does adds them during the playback.

Harmonics occur as part of a sound wave and audio signal in a different fashion (although explainable using the same principles). When a wave occurs it is because it carries certain amplitude which is sufficient to create the next cycle of the same in the longitudinal sense of propagation. Waves do not isolate their energy in a certain portion of the medium but rather they propagate in all direction. A secondary oscillation that occurs from a central one, will carry some of the energy of the main wavelength and will have a very similar frequency value. In the space of the medium between these two, if these carry enough amplitude, some superior harmonics will form. Most sound waves indeed carry higher harmonics. Lower harmonics, always. The difference and correlation of the phase and amplitude between any fundamental and it's harmonics is very specific and no current digital systems are able to capture it. Although this can be remediated to a point using higher sampling rates and more bit depth.

Another part of the story is that there's no feasible way for any current digital system to match their sampling clock or microprocessor clock to a certain value of phase for any audio frequencies. Let's say that you have a tone of a constant frequency coming out of a violin. .. This tone is A 440. This tone is going to acquire any 360 degrees of phase 440 times in one second. No digital system can tell if they started recording at any given value of phase or another. Or, in other words, when the ADC generates a code value, they are not delivering the phase value of the voltage of the audio signal.

(Certainly you can see in many DAW's a beutiful graphic which is somehow an approximation of the voltage value and phase of any signals. But the margin of error is huge, depending how you wanna calculate it, it can be as huge as 600-1200% or more).

Most sound fields are composed by different tones of different frequencies, at different amplitudes. When these frequencies travel together in the same medium and when they reach together the element of a microphone, they are going to superpose to each other 'transiently' at interger points and create numerous intermodulations.

Similarly to the effect of 2 equal tones cancelling or reinforcing each other, any two simultaneous frequencies will alter each other in some way even if they are far apart in the spectrum. And the values of the resulting intermodulations are very phase-specific (and not always a factorial of 2x2x3x3x5x5x7x7).

Oftentimes this transients harmonics as well as the natural harmonic could be 'seen' in a graphic as appearing for a couple of miliseconds every some five or six, ten or more miliseconds.

And the precise reproduction of this harmonic content is what makes a recording to sound more natural. Most digital systems will cut the frequency spectrum at 20kHz. No digital systems are able to deliver a neat reproduction of the harmonic content of a recording. Although, it does not means it's all lost. Loudspeakers and preamps will -to some extent- re-create artificially the lost harmonic content by reinforcement of fundamentals.

And anyways the sound quality is not everything. I prefer to listen to my favorite song at a listenable level of compression than a recording I don't like nevermind what level of resolution.

 

[Ennis]

Which of those two groups are you in? We have quite a few in the first category about the place, so those in the second sometimes have a hard time here.

Apoligies for entering in such a smug and arrogant domain. I hope the couple of comments I made help you break your cycle of self redundancy.

I do have things to do and no time for people being hard of each other.

Sweetest goodbye!

 

[Ennis]

No. Humans cannot hear ultrasonics, regardless of whether the ultrasonics are fundamental tones or harmonics of audible-range fundamentals.

Sounds - transient harmonics or otherwise - within the audible range are not "better reconstructed" with higher sample rates. If that were true, then every single digital recording ever made would suffer from massive differences in sound quality within the audible range, regardless of sample rate - no matter your sample rate, 100Hz gets sampled 10x as much as 1kHz and 100x as much as 10kHz. So every recording would have much "better reconstruction" of bass drums, bass guitar, and other low-frequency sounds than of female vocals, snares, hi-hats, and other high-frequency sounds.

Finally, on the question of timing, the timing accuracy of 44.1kHz digital sampling is more than enough to accommodate the timing accuracy of human hearing - in fact, a key principle of digital sampling theory, which makes digital recording possible at all, is that the frequency of sampling really has nothing to do with the perceived "speed" or "timing accuracy" of transients or any other sounds. The idea that a transient might be "too fast" to be captured by 44.1kHz sampling is quite literally nonsense: the only sound "too fast" for 44.1kHz is a sound whose frequency is above 22.05kHz. The only sound that can be played or recorded that is "faster" than that is a sound at a frequency higher than that. Sampling frequency is about the highest-frequency sound that can be recorded and reconstructed. It is not about timing accuracy.

Our modern science does not know the exact efficiency of the air or any mediums for each frequency. Although a superior harmonic can be identified or in theory found as as x1.5 , x2 or x3 times the frequency but the same amplitude as the fundamental (in theory anywhere from 1.1875 and on ... times the frequency, and still you can find harmonic noise in the range from 1 to 1.1875).

And you don't hear the ultrasonic harmonics, what you hear is the harmonic modulation of the fundamental in the hearable range, making it sound wider and fuller.