Why 24/192 is a bad idea?

[Ethan Winer]

But if you came to the discussion with a bit more substance, you wouldn't need to start threads to dance in a circle with glee when his noisefloor is off by a bit.

It's not even a matter of "being off," because the level at which noise and other sounds can be heard varies greatly depending on the spectrum of the program versus that of the noise, correlated or not (THD versus IMD), and absolute SPL level. Masking is a huge factor, which is why you can't pin this down to a single number. People that attempt to do so, or expect me to do so, are truly missing how audio and perception work.

--Ethan

 

[FrantzM]

Thinly veiled? Completely unveiled, I'd say.

Audiophiles are their own worst enemies, often. To outsiders, we must appear exquisitely uncool. Nattering nabobs of negativity, indeed.

Agreed...

 

[jkeny]

Can't say I care for that thread. It strikes me as a thinly-veiled attack on Ethan. I understand that Ethan is a bit black and white in his views, that he doesn't mince words, that he can be rather demanding of those who attempt to attack his point of view without a well-formed and supported one of their own. I'm sure it's irritating; enough to make you just wait for him to make a mistake. But if you came to the discussion with a bit more substance, you wouldn't need to start threads to dance in a circle with glee when his noisefloor is off by a bit. You might actually have a viable position of your own. You might even find yourself occasionally agreeing with Ethan.

Tim

http://www.whatsbestforum.com/showt...o-transparency&p=109838&viewfull=1#post109838

 

[jkeny]

Oh yes, I understood what he posted. I also understood your post, which was nothing more than words to the effect of "I'm right, your wrong." What I don't understand is what it is which you are right about nor, for that matter, what it is about which he is incorrect. I also understand that is the capital A audiophile way: attack a person rather than the substance of the post. Fortunately here at WBF we try to focus on the post, not the poster, and one will be admonished to stick to this simple rule. So do you have anything of substance to add, e.g., reliable, repeatable, demonstrative proof, not opinion, that artifacts 100 dB down are audible?

http://www.whatsbestforum.com/showt...o-transparency&p=109839&viewfull=1#post109839

 

[jkeny]

It's not even a matter of "being off," because the level at which noise and other sounds can be heard varies greatly depending on the spectrum of the program versus that of the noise, correlated or not (THD versus IMD), and absolute SPL level. Masking is a huge factor, which is why you can't pin this down to a single number. People that attempt to do so, or expect me to do so, are truly missing how audio and perception work.

--Ethan

http://www.whatsbestforum.com/showt...o-transparency&p=109853&viewfull=1#post109853

 

[jkeny]

If you really want to get back to the subject of this thread it might be useful to read & discuss this paper from Cirrus "A New Perspective on Decimation and Interpolation Filters " http://www.cirrus.com/en/pubs/whitePaper/DS668WP1.pdf

In it you will find this statement

At the 75th convention of the AES in 1984, Dr. R. Lagadec presented a paper, “Dispersive Models for AD
and D-A Conversion Systems” [1], which discussed the potential degradation in audio quality that results
from the “pre-echoes” that exist in the impulse and transient response of FIR filters. This premise,
though of interest, apparently received little attention at the time. Now fast-forward to the mid-nineties, 96
kHz sample rates have become a reality, 192 kHz is on the horizon and SACD has been introduced.
There is general agreement that the evolution to higher sample rates offers an audible improvement over
either 44.1 or 48 kHz and, as with most human endeavors, we search to understand why. The obvious
and intuitive explanation is that the improvements result from the extended frequency range these formats
provide. However, it is generally accepted that the upper frequency limit of human hearing is in the neighborhood
of 20 kHz and both 44.1 and 48 kHz sample rates have frequency response that extends beyond
20 kHz. Apparently there must be something else and there has been considerable discussion, several
AES papers and magazine articles [2,3,4,5,6,7,8], since the mid 90's related to the audible improvement
of the higher sample rates. Much of this work has been a continuation of the work presented by Dr. Lagadec
in 1984. (It is also very interesting to note that Sony recently claimed on their website that it is the
removal of the digital decimation and interpolation filters, not the extended frequency range, that produce
the audible improvements offered by SACD over conventional PCM.)

 

[amirm]

It's not even a matter of "being off," because the level at which noise and other sounds can be heard varies greatly depending on the spectrum of the program versus that of the noise, correlated or not (THD versus IMD), and absolute SPL level. Masking is a huge factor, which is why you can't pin this down to a single number. People that attempt to do so, or expect me to do so, are truly missing how audio and perception work.

--Ethan

Masking -- a psychoacoustics factor where we don't hear faint sounds/distortions in presence of louder ones -- certainly helps to erase many of our sins in audio. To wit, we are able to compress most music at 128kbps and have many think nothing has been taken out despite throwing out more than 90% of the PCM samples.

Unfortunately what nature giveth, it takes away . Unless you can control what people listen to, you can't guarantee masking will always be there to bail you out. This is why most of us rip into lossless compression. It doesn't rely on masking to do its job. Likewise, if we want to say with confidence that distortions are not audible, then we have to do away with masking and make our call independent of that.

Contrary to what you say, we can compute the number if we look at threshold of audibility for noise/distortion and make sure our system matches or outperforms it. Indeed, this is precisely what Bob Stuart did in his paper, showing where 16, 18 and 20 bits quantizatoin noise lands relative to that, compensated for our ability to hear noise/distortion perceptually. Same practice is used by Hawksford, Dunn, etc.

 

[jkeny]

Furthermore in that same paper you will find:

The impulse response of a typical FIR filter for a 48 kHz sample rate is shown in Figure 1. Notice the ringing or “pre-echo” prior to the arrival of the impulse. This pre-echo creates what has been described as a time smear or energy dispersion of the original signal. Figure 2. shows the impulse response of the same filter at a 96 kHz sample rate. It is apparent that the pre-echo has shorter time duration and less time dispersion than at 48 kHz. It is this difference that many believe to be a primary source of the audible superiority of the higher sample rates. As a result of this, many industry experts have suggested that the additional bandwidth offered by the higher sample rates be used as a transition band to allow the use of less aggressive digital filtering, which minimizes pre-echoes and time dispersion even further. Julian Dunn [4] and Mike Story [7] have both written very nice discussions of this subject.

EDIT: How do I post these images - I tried a number of variations but no luck! Sorted

 

[Phelonious Ponk]

It's not even a matter of "being off," because the level at which noise and other sounds can be heard varies greatly depending on the spectrum of the program versus that of the noise, correlated or not (THD versus IMD), and absolute SPL level. Masking is a huge factor, which is why you can't pin this down to a single number. People that attempt to do so, or expect me to do so, are truly missing how audio and perception work.

--Ethan

I get that, Ethan. 16 bits, 17 bits, I'll bet most of these middle-aged guys can't hear the noise floor at 12 bits, but we dom't deal with we hear, in high end audio...well, unless,of course, what we measure and test defies the conventional wisdom. Then it's all about, only about what we hear. or think we do.

Tim

 

[Orb]

I get that, Ethan. 16 bits, 17 bits, I'll bet most of these middle-aged guys can't hear the noise floor at 12 bits, but we dom't deal with we hear, in high end audio...well, unless,of course, what we measure and test defies the conventional wisdom. Then it's all about, only about what we hear. or think we do.

Tim

So what is your thoughts on recordings where the signal for frequency range 2khz to 10khz (usually higher frequencies drop by another -20 to -40dbfs from review measurements in HifiNews) are in the -50dbfs to -90dbfs when looking at a digital recording?
Need to remember when it comes to digital the recording-signal is substantially lower than the 0dbfs, in the context of well recorded music (this is important as we know there are good and poor recordings so let us focus on the good ones for now please as they have most to lose).
Cheers
Orb

 

[Phelonious Ponk]

So what is your thoughts on recordings where the signal for frequency range 2khz to 10khz (usually higher frequencies drop by another -20 to -40dbfs from review measurements in HifiNews) are in the -50dbfs to -90dbfs when looking at a digital recording?
Need to remember when it comes to digital the recording-signal is substantially lower than the 0dbfs, in the context of well recorded music (this is important as we know there are good and poor recordings so let us focus on the good ones for now please as they have most to lose).
Cheers
Orb

Are you saying this is the norm? That digital recordings have a huge dip in amplitude response from 2k to 10k? This is news to me. My thoughts are that I need some more information or another explanation of what you're saying.

Tim

 

[rbbert]

Are you saying this is the norm? That digital recordings have a huge dip in amplitude response from 2k to 10k? This is news to me. My thoughts are that I need some more information or another explanation of what you're saying.

Tim

No, he's pretty clearly saying that good recordings (analog or digital) have a relatively low average and even lower minimum volume between 2k and 10kHz, and that frequencies above that are even lower in volume by a significant (20 - 40 dB) amount. Nothing at all controversial there, just simple observations.

 

[Phelonious Ponk]

No, he's pretty clearly saying that good recordings (analog or digital) have a relatively low average and even lower minimum volume between 2k and 10kHz, and that frequencies above that are even lower in volume by a significant (20 - 40 dB) amount. Nothing at all controversial there, just simple observations.

Hmmm....I'm gonna have to go back and re-read that one.

Tim

 

[Ethan Winer]

Unless you can control what people listen to, you can't guarantee masking will always be there to bail you out.

Of course. Background noise and hum / buzz are there all the time, and are more likely to be heard when the music isn't playing. But by and large we've been talking about the audibility of artifacts that affect the music. So at least some sort of masking is always available. Unless the music consists solely of a 100 Hz sine wave in the presence of treble-heavy hiss. Given the music some modern composers are writing, I suppose that's possible!

Remember, this started with the claim that bit depth affects more than just the noise floor, and not using enough bits can cause music to sound "bleached." So the context is definitely music, not noise at a static level. Not that I've ever once heard background hiss from a CD other than when the original recording is analog tape.

Likewise, if we want to say with confidence that distortions are not audible, then we have to do away with masking and make our call independent of that.

Yes, again this is distortion, and masking is always present to some extent.

--Ethan

 

[Orb]

Hmmm....I'm gonna have to go back and re-read that one.

Tim

Tim,
Rbbert has it spot on.
So what are your thoughts on those recordings?
Because it seems you feel recordings start at -0dbfs and only go down to -50dbfs, where in reality good recordings will have music-signal much lower than that as I mentioned.
Thanks Rbbert for your post.
Cheers
Orb

 

[Soundproof]

What's being pointed out here is something that it's worth making a note of - the energy level in music drops off with an increase in frequency, and the ability of the sound to travel through the medium (air), is likewise reduced with higher frequencies. (Small and frequent waves don't make as much of an impact as large ones that come rolling towards you, is a serviceable image).

One can, of course, also tweedle something up in the batsonics-sphere, to make it appear there is something we can hear there.

Like in this HD-Tracks high-res release of Norah Jones, which also illustrates the point about falling energy levels the higher the frequency, until the 22.5kHz point, when the tweedling begins.

http://www.computeraudiophile.com/a...27-norah-jones-come-away-me-hdtracks-four.jpg

 

[Orb]

That is just one aspect of what I am mentioning.
But continuing on with this aspect emphasised by Soundproof; how many have listened to a good digital recording with all frequencies above 5khz, then 10khz,etc equalised out and then compared them to the original?
Ideally I would prefer this to be done at or as part of the studio process and keep same sampling rate-filters-etc, for various technical reasons that "may" have a contribution.
Would be interesting to try this for both a native high-rez recording and then its CD version to see if both tests give the same results, but the complexity here is that we do have technical differences that may throw up some anomolies.
Cheers
Orb

 

[Phelonious Ponk]

Tim,
Rbbert has it spot on.
So what are your thoughts on those recordings?
Because it seems you feel recordings start at -0dbfs and only go down to -50dbfs, where in reality good recordings will have music-signal much lower than that as I mentioned.
Thanks Rbbert for your post.
Cheers
Orb

I don't think I've said, here or anywhere, that good recordings start at 0 dB. I'm sure many digital recordings, lacking the noise problems, have been recorded much lower. My thoughts on them? I don't think I've ever heard the noise floor of a good digital recording. Then again, I din't think I've listened for one in decades.

Tim

 

[bdiament]

0dBFS

A bit off topic but since it has been mentioned, some information:

In the old days of digital, it was thought resolution was maximized by ensuring the loudest sounds peak at 0 dBFS ("0 dB Full Scale"), the loudest that can be captured by a digital recording system.

With every ~6 dB (actually, more like 6.02 dB) representing one bit in the digital "word", lower level signals are captured with ~1 bit less resolution for every 6.02 dB below 0 dBFS. Translated to English, this means, with uncompressed audio, where the average level can be 20 dB or more below the peak level, the average level parts of the audio signal would be captured with >3 bits less resolution than the loudest peaks, assuming the level of the loudest peaks was 0 dBFS. With a 16-bit system, that means the average level sounds in a recording would be captured using the bottom 12 or 13 bits of resolution. Other sounds in an uncompressed recording may be well down in level from there and end up being captured by considerably fewer bits. Astute listeners will hear the coarsening of harmonic structure that results from the lower resolution, as well as the defocusing, if not complete obliteration of spatial information. (I talked about this in a little more detail in post #25 on page 3.)

They key I want to mention in this post is that most of my colleagues have come to avoid having the maximum peak at 0 dBFS and today, will leave anywhere from 0.3 dB to a dB (sometimes more) headroom. The current thinking is to see 0 dBFS as the equivalent of an overload and in fact, the meters on some A-D converters will now go "into the red" when the signal reaches 0 dB, rather than waiting until that level is exceeded.

It turns out a number of D-A converters, particularly cheaper ones, can "stick" when the signal gets to 0, producing artifacts, sometimes manifesting as low level clicks, sometimes worse. Moreover, it is possible for the peaks in the reconstructed analog signal to be higher than the values represented by the samples on either side of the peak (hence, they are called "intersample peaks"). Avoiding max peaks at 0 dBFS eliminates these problems.

Best regards,
Barry
www.soundkeeperrecordings.com
www.barrydiamentaudio.com

 

[amirm]

It turns out a number of D-A converters, particularly cheaper ones, can "stick" when the signal gets to 0, producing artifacts, sometimes manifesting as low level clicks, sometimes worse. Moreover, it is possible for the peaks in the reconstructed analog signal to be higher than the values represented by the samples on either side of the peak (hence, they are called "intersample peaks"). Avoiding max peaks at 0 dBFS eliminates these problems.

Best regards,
Barry
www.soundkeeperrecordings.com

Good post Barry. The above is the reason digital audio measurements are done at -3 dbFs to -6 dBfs. Word length clipping may occur in the DAC's digital processing due to ringing and such and push us over the edge. Sad that pop music comes near 0 dBfs and as such gets audibly distorted on a number of devices I have listened to.