Are we just kidding ourselves?

[sandyk]

The young generation is happy with MP3 so they have to convince the old one that they can hear enough to keep the business going ¨-)

 

That is likely to soon change due to the efforts of YouTube and VEVO with their now usually 1080 resolution music video clips, as well as some 4K resolution MV clips, and the use of .aac audio instead of MegaPoop 3. At the bit rates used, .aac still isn't as good as LPCM or .flac, but for many it will still sound better than .mp3

[wgscott]

Please explain.

 

Fourier already did.

[kennyb123]

That is just wrong. In fact, it betrays a complete misunderstanding.

 

Quote from this article:

 

"In lab tests, people can distinguish between sounds as little as five milliseconds apart".

 

How can a sample rate of 44.1k be sufficient to capture the timing changes our ears are both capable of hearing and used to hearing when we hear live music?

[kennyb123]

Fourier already did.

 

Can you point me to where Fourier mentioned music sampling specifically? I missed that.

[REShaman]

I wouldn't buy a DSD DAC from those that doesn't believe in DSD: Benchmark

 

Walberg is Walberg, he is defending his own 24/96 recordings and I don't trust his ears, because I don't like his mediocre recordings.

 

Roch

 

Is Walberg Mark Waldrep?

 

Best,

Richard

[wgscott]

Can you point me to where Fourier mentioned music sampling specifically? I missed that.

 

Shannon and Nyquist developed the sampling theorem before digital music was invented. Fourier came before them. That is absolutely irrelevant, as you well know.

 

The main point is the claim you quote and endorse that the Nyquist theorem cannot apply to "complex" signals is incorrect. Anyone with a rudimentary understanding of the sampling theorem and what a Fourier Transform is would recognize this as bogus. If you don't have this understanding, that of course is absolutely fine, but it might therefore be problematic to dismiss it if you don't know anything about it.

 

Briefly, any signal (or indeed any function) can be expressed as a summation (or integral) of simple component sign and cosine functions. The sampling theorem then applies to each one of those components in exactly the same way as it would to one solitary periodic function. There is no difference, because a Fourier Transform is a linear summation. In addition, frequency and time are Fourier conjugate variables, so what applies in the frequency domain also equally applies in the time domain. In other words, any difference in the reconstructed impulse response would reside beyond the Nyquist frequency. Hence the stated concern about being unable to capture the information about the "velocity" of the sound wave (or more charitably, the impulse response) is unfounded.

[kumakuma]

Shannon and Nyquist developed the sampling theorem before digital music was invented. Fourier came before them. That is absolutely irrelevant, as you well know.

 

The main point is the claim you quote and endorse that the Nyquist theorem cannot apply to "complex" signals is incorrect. Anyone with a rudimentary understanding of the sampling theorem and what a Fourier Transform is would recognize this as bogus. If you don't have this understanding, that of course is absolutely fine, but it might therefore be problematic to dismiss it if you don't know anything about it.

 

Briefly, any signal (or indeed any function) can be expressed as a summation (or integral) of simple component sign and cosine functions. The sampling theorem then applies to each one of those components in exactly the same way as it would to one solitary periodic function. There is no difference, because a Fourier Transform is a linear summation. In addition, frequency and time are Fourier conjugate variables, so what applies in the frequency domain also equally applies in the time domain. In other words, any difference in the reconstructed impulse response would reside beyond the Nyquist frequency. Hence the stated concern about being unable to capture the information about the "velocity" of the sound wave (or more charitably, the impulse response) is unfounded.

 

Bill,

 

It sounds like you know what you are talking about but you know that your PhD in Chemistry doesn't carry any weight around here, especially with our friend from down under.

 

KK

[kennyb123]

The main point is the claim you quote and endorse that the Nyquist theorem cannot apply to "complex" signals is incorrect. Anyone with a rudimentary understanding of the sampling theorem and what a Fourier Transform is would recognize this as bogus.

 

I absolutely agree with you that Nyquist *can* apply to "complex" signals.

 

I think Dudley, whom I quoted, had something else in mind by the word "complex" and should not have used this word. His main gist was the point he made about velocity. I think he was suggesting complexity in the time domain.

 

Briefly, any signal (or indeed any function) can be expressed as a summation (or integral) of simple component sign and cosine functions. The sampling theorem then applies to each one of those components in exactly the same way as it would to one solitary periodic function. There is no difference, because a Fourier Transform is a linear summation. In addition, frequency and time are Fourier conjugate variables, so what applies in the frequency domain also equally applies in the time domain. In other words, any difference in the reconstructed impulse response would reside beyond the Nyquist frequency. Hence the stated concern about being unable to capture the information about the "velocity" of the sound wave (or more charitably, the impulse response) is unfounded.

 

I was with you right up until the last sentence.

 

Doesn't the sample rate selected for music need to account for more that just what's required by Nyquist? Doesn't it also have to be sufficient to pick up the changes occurring in time that we are capable of hearing in live music? If we're capable of distinguishing "between sounds as little as five milliseconds apart"' then wouldn't too low of a sample rate leave out some information that might contribute to convincing our ears that we are hearing the real thing?

[Jud]

Shannon and Nyquist developed the sampling theorem before digital music was invented. Fourier came before them. That is absolutely irrelevant, as you well know.

 

The main point is the claim you quote and endorse that the Nyquist theorem cannot apply to "complex" signals is incorrect. Anyone with a rudimentary understanding of the sampling theorem and what a Fourier Transform is would recognize this as bogus. If you don't have this understanding, that of course is absolutely fine, but it might therefore be problematic to dismiss it if you don't know anything about it.

 

Briefly, any signal (or indeed any function) can be expressed as a summation (or integral) of simple component sign and cosine functions. The sampling theorem then applies to each one of those components in exactly the same way as it would to one solitary periodic function. There is no difference, because a Fourier Transform is a linear summation. In addition, frequency and time are Fourier conjugate variables, so what applies in the frequency domain also equally applies in the time domain. In other words, any difference in the reconstructed impulse response would reside beyond the Nyquist frequency. Hence the stated concern about being unable to capture the information about the "velocity" of the sound wave (or more charitably, the impulse response) is unfounded.

 

First the other piece of what Kenny was wondering about, then a handful of mild caveats to the above.

 

Bill has explained nicely how Fourier math allows Nyquist to work on complex musical passages, not just simple sine waves (so long as none of the harmonics making up the complex piece of music has a frequency equal to or greater than half the sampling rate). But how does Nyquist allow us to be sure of what's between the samples?

 

Imagine a single point on a graph. There are an infinite number of musical signal waves that can pass through that single point. Now add a second point. Not all the signals that passed through the first point will also pass through the second, so you've reduced things a bit. As soon as you add a third point (as soon as your sample rate goes above twice the highest frequency of interest), there will only be one musical signal wave that can pass through all three points. So you have defined an entire wave all along its length, not just three points, and you can be quite certain of where that wave is at any moment. Therefore you do not need to sample at 100,000 times per second to determine where the signal is one hundred-thousandth of a second before or after the sample on a CD. The signal has been defined at every point along its length. Thus the intuitive objection ("I would need to sample everywhere to be sure of what the signal is doing everywhere, right?") is incorrect.

 

Mild caveats:

 

- Human hearing isn't modeled completely accurately by Fourier analysis, or to put it another way, the ear-brain system doesn't work completely in accordance with Fourier math. In particular, it appears from some academic research that (a) transients/impulses are processed in a different part of the brain than tones, and (b) humans may be able to detect transients with a faster rise time than a 20kHz sine wave.

 

- Fourier analysis and Nyquist work on harmonic signals. Many of the sounds in music are *inharmonic* (percussion, various vocal sounds such as plosives, string plucks, etc.). These can't be modeled exactly using Nyquist and Fourier analysis, though excellent approximations can be "built" out of harmonic signals. But in order to do this, one may need a sample rate adequate for greater than 20kHz since some of these inharmonic sounds will have faster rise times than a 20kHz sine wave. (See the first caveat above.)

 

- I continue to wonder whether the audible intermodulation effects of instrumental overtones above 20kHz are captured satisfactorily by sample rates adequate only for the audible range.

 

These caveats are completely independent from concerns arising out of the filtering used at various points along the recording and playback chain that I mentioned in a comment upthread.

[esldude]

I absolutely agree with you that Nyquist *can* apply to "complex" signals.

 

I think Dudley, whom I quoted, had something else in mind by the word "complex" and should not have used this word. His main gist was the point he made about velocity. I think he was suggesting complexity in the time domain.

 

 

 

I was with you right up until the last sentence.

 

Doesn't the sample rate selected for music need to account for more that just what's required by Nyquist? Doesn't it also have to be sufficient to pick up the changes occurring in time that we are capable of hearing in live music? If we're capable of distinguishing "between sounds as little as five milliseconds apart"' then wouldn't too low of a sample rate leave out some information that might contribute to convincing our ears that we are hearing the real thing?

 

The mistake your making is thinking the time between samples is the smallest time shift that can be portrayed by the digital signal. That is not the case. The sample values would be different for signal A vs signal B even if signal B is identical, but delayed in time by one microsecond. Which means the reconstructed waveforms like in right and left channel would show a phase shift of one microsecond. In fact properly operating redbook (44.1/16) can potentially portray phase shifts of as little as 55 picoseconds. Even lower if dither is properly applied.

 

Also I might mention that humans can with some test signals detect two signals different by as little as about 11 microseconds. Whether these signals are presented as analog or redbook digital 11 microseconds is possible.

 

If you haven't seen it, this excellent video shows very simply how PCM really works. It uses analog signal sources and analog scopes and FFT's to show what happens when an ADC/DAC conversion has occurred in the middle. Well worth 24 minutes of your time to view.

 

Xiph.Org Video Presentations: Digital Show & Tell

[Jud]

I just checked the Scheops web site, I couldn't find a single microphone that specified an upper frequency response above 20K. The same goes for AKG, every spec shows 20K. As for B&K (I assume you mean Bruel & Kjaer), they make specialized instrumentation microphones that extend to ultrasonics, but once again their studio microphones are speced to 20K. As the most popular studio microphones, the Neumann U67/U87, they are speced to 20K except for the old tube versions that are still much loved in certain circles that top off at 16K,

 

This is very much tangential to the main point, which has to do with filtering, not the ability to hear above 20kHz. But just so that no one is misled into thinking pro mics are all limited to 20kHz, here's an example of a very well thought of one that isn't: PM40

[kennyb123]

Fourier analysis and Nyquist work on harmonic signals. Many of the sounds in music are *inharmonic* (percussion, various vocal sounds such as plosives, string plucks, etc.). These can't be modeled exactly using Nyquist and Fourier analysis, though excellent approximations can be "built" out of harmonic signals. But in order to do this, one may need a sample rate adequate for greater than 20kHz since some of these inharmonic sounds will have faster rise times than a 20kHz sine wave.

 

Thanks Jud for that very nice explanation. I quoted the above as I think that gets to the crux of the point that I was trying to make and why I thought it worth quoting Dudley. You've just said it better than both he and I.

 

I find the areas you call out above are often greatly improved in high res. Cymbals, percussion, and plucked strings often sound far more realistic to me at higher sample rates. The more realistic reproduction of rise time contributes greatly to instruments sounding like the real thing.

 

That these "inharmonic" sounds exist in music seems to be overlooked by those claiming that 44.1k is sufficient.

[Jud]

Thanks Jud for that very nice explanation. I quoted the above as I think that gets to the crux of the point that I was trying to make and why I thought it worth quoting Dudley. You've just said it better than both he and I.

 

I find the areas you call out above are often greatly improved in high res. Cymbals, percussion, and plucked strings often sound far more realistic to me at higher sample rates. The more realistic reproduction of rise time contributes greatly to instruments sounding like the real thing.

 

That these "inharmonic" sounds exist in music seems to be overlooked by those claiming that 44.1k is sufficient.

 

These caveats are not intended as factual pronouncements but as potential sticking points. I have heard enough tremendously well recorded RedBook to be uncertain whether I can definitively hear differences between RedBook and higher res with inharmonic transients. (Right now I'm thinking of Brian Bromberg's CD "Wood," which has some amazing cymbal playing, as well as portraying Bromberg's bass with an almost "you are there" sense of reality.)

 

What I feel somewhat more confident about (though with just my ears I can't pretend to certainty) is the filtering stuff I talked about upthread. That is, I think the differences between better and worse - or really good and better; let's give filter designers credit for not giving us stuff bad enough to be classed as "worse" - filtering chains may well be audible.

[sandyk]

Extract from Jud's link to the PM40

 

With a frequency response of 9 Hz to 40 kHz, the mics deliver extraordinary impulse response

 

What's the point of using them for making piano recordings with ONLY 16/44.1 then, when RB CD is clearly incapable of doing them justice ?

[kennyb123]

These caveats are not intended as factual pronouncements but as potential sticking points.

 

Understood, but I do think that this is one contributor to why high res sounds better. The other, and probably more significant, is that filtering stuff you've mentioned.

 

I have heard enough tremendously well recorded RedBook to be uncertain whether I can definitively hear differences between RedBook and higher res with inharmonic transients. (Right now I'm thinking of Brian Bromberg's CD "Wood," which has some amazing cymbal playing, as well as portraying Bromberg's bass with an almost "you are there" sense of reality.)

 

I have a rip of that but have yet to get around to listening to it. I will do that tonight - thanks for the suggestion.

[kennyb123]

The mistake your making is thinking the time between samples is the smallest time shift that can be portrayed by the digital signal. That is not the case. The sample values would be different for signal A vs signal B even if signal B is identical, but delayed in time by one microsecond. Which means the reconstructed waveforms like in right and left channel would show a phase shift of one microsecond. In fact properly operating redbook (44.1/16) can potentially portray phase shifts of as little as 55 picoseconds. Even lower if dither is properly applied.

 

I did not know that. Thanks for offering this insight.

 

If you haven't seen it, this excellent video shows very simply how PCM really works. It uses analog signal sources and analog scopes and FFT's to show what happens when an ADC/DAC conversion has occurred in the middle. Well worth 24 minutes of your time to view.

 

Xiph.Org Video Presentations: Digital Show & Tell

 

Thanks I will check it out.

[Paul R]

Of course, you are simplifying here more than a little. Something like a simple transport delay is essentially linear, and delays all frequencies equally. That is not the case with PCM reconstruction, and is one reason why higher sample rates can, and often do, result in better reproduction.

 

The capability to resolve something does not automatically imply that the resolution is correct.

 

-Paul

 

 

The mistake your making is thinking the time between samples is the smallest time shift that can be portrayed by the digital signal. That is not the case. The sample values would be different for signal A vs signal B even if signal B is identical, but delayed in time by one microsecond. Which means the reconstructed waveforms like in right and left channel would show a phase shift of one microsecond. In fact properly operating redbook (44.1/16) can potentially portray phase shifts of as little as 55 picoseconds. Even lower if dither is properly applied.

 

Also I might mention that humans can with some test signals detect two signals different by as little as about 11 microseconds. Whether these signals are presented as analog or redbook digital 11 microseconds is possible.

 

If you haven't seen it, this excellent video shows very simply how PCM really works. It uses analog signal sources and analog scopes and FFT's to show what happens when an ADC/DAC conversion has occurred in the middle. Well worth 24 minutes of your time to view.

 

Xiph.Org Video Presentations: Digital Show & Tell

[mayhem13]

Extract from Jud's link to the PM40

 

 

 

What's the point of using them for making piano recordings with ONLY 16/44.1 then, when RB CD is clearly incapable of doing them justice ?

 

It's not Redbook that's the bottleneck.....it's the physical limitations of your ears.

[mayhem13]

Thanks Jud for that very nice explanation. I quoted the above as I think that gets to the crux of the point that I was trying to make and why I thought it worth quoting Dudley. You've just said it better than both he and I.

 

I find the areas you call out above are often greatly improved in high res. Cymbals, percussion, and plucked strings often sound far more realistic to me at higher sample rates. The more realistic reproduction of rise time contributes greatly to instruments sounding like the real thing.

 

That these "inharmonic" sounds exist in music seems to be overlooked by those claiming that 44.1k is sufficient.

 

Are you claiming that Redbook or 16/44 has increased odd order harmonic distortion over Hi-res?.......which one.....3rd? 5th?........how much of an increase or closer to the fundamental?.......single tone bursts at specific points in the FR or across the entire passband?

[gmgraves]

No, it doesn't mean that every textbook on human hearing is wrong. It simply suggests that most need updating to reflect more recent research.

You also keep dismissing the large number of reports about people hearing further improvements with the very latest DSD format.

Theory will eventually show that these people are correct. It just takes a while to catch up with the real world.

 

 

Most ultrasonic microphones are instrumentation/calibration mikes and not recording mikes. Point is, they're not that rare.

[mayhem13]

These caveats are not intended as factual pronouncements but as potential sticking points. I have heard enough tremendously well recorded RedBook to be uncertain whether I can definitively hear differences between RedBook and higher res with inharmonic transients. (Right now I'm thinking of Brian Bromberg's CD "Wood," which has some amazing cymbal playing, as well as portraying Bromberg's bass with an almost "you are there" sense of reality.)

 

What I feel somewhat more confident about (though with just my ears I can't pretend to certainty) is the filtering stuff I talked about upthread. That is, I think the differences between better and worse - or really good and better; let's give filter designers credit for not giving us stuff bad enough to be classed as "worse" - filtering chains may well be audible.

 

While you theorize about the filter's impulse or step response and it's ability or lack there of to faithfully reproduce dynamic rise and decay of a signal, best not to forget the limitations on the acoustic side of things where we're dealing with mechanics....and sadly things are far worse........except for plasma. Maybe add a bit of sulfer hexaflouride to your Vandy's enclosure to stabilize that ringing passive radiator?.......at least until it leaks out and further impacts global warming.

[mayhem13]

Most ultrasonic microphones are instrumentation/calibration mikes and not recording mikes. Point is, they're not that rare.

 

Please don't antagonize him as accepted science has little weight against his theoretical positions. You'd do better to stick toothpicks under your fingernails.

[kennyb123]

Are you claiming that Redbook or 16/44 has increased odd order harmonic distortion over Hi-res?.......which one.....3rd? 5th?........how much of an increase or closer to the fundamental?.......single tone bursts at specific points in the FR or across the entire passband?

 

You quoted me, but I didn't claim any of those things.

[bmoura]

+1 .

 

After 40 years in this hobby, I agree.

 

I listen to classical music and have just got into digital because of Qobuz and NativeDSD.

 

I downloaded two free files from NativeDSD to discern differences between Native 128dsd and 256dsd. I could discern no difference in quality although the 256dsd file was twice as large.

 

However, their was a difference in quality between the files I have ripped from cd's and those that I downloaded from NativeDSD. NativeDSD is vastly superior. Now, whether this is due to the recording / mastering process or the technology, I don't think I'll ever know.

 

My results match yours.

 

Some files ripped from SACDs can come close to the original in sound quality but others are quite a bit less convincing. And when you're offering downloads created from the DSD Edit Master, as you find on NativeDSD.Com or DSDFile.Com for example, you really are very close to the source - much closer than an optical disc or optical disc rip can provide.

 

It's a good time to be a music fan!

[sandyk]

It's not Redbook that's the bottleneck.....it's the physical limitations of your ears.

 

Then according to you, 24/96.24/192 and DSD is a big con job, and anybody who hears a benefit from these higher resolution formats is delusional ?