Is Tom Scholz (Boston) correct?

[Waveform]

In the current issue of S & V, Tom Scholz of Boston says:

"Digital sounds a lot different than the original source. Things are further compromised when you decrease the resolution to 16 bits from 24 bits or higher. The combination of only 16-bit resolution and only 44.1-kHz sampling rate absolutely demolishes any part of the signal above 10k. If you put a 12k tone, which most people can hear, through a 16-bit, 44.1k sampling at digital conversion, you will be shocked at what that waveform looks like coming out the other end. You can put a pure tone in, and it comes out looking like some monstrous garbled thing......And if you changed that 12kHz signal just a percentage point, the signal that spits out will look entirely different. So every time somebody hits an "s" or a cymbal, or plays a delicate violin or even a raunchy distorted guitar with lots of high frequencies, the high-frequency end of that spectrum is completely mangled into something different. That's why people speak about strange sibilance, or things sticking out or not sounding "right," when they listen to CDs. It isn't right. It's completely different than the original recording."

The interview is with Mike Mettler, the current editor of S&V, who lets this stand without comment, indicating that he may agree with this.

 

[chadnliz]

I dont know if everything he says is correct but you have to believe a guy who has been in the music and recording industry both in bands and behind the scene for over 30 yrs has a bit more knowledge then most enthusiast's on web forums.

 

[croseiv]

I don't think it's 'completely' different from the original recording, because a sound would become unrecognizable the way he describes it. At least in my experience sounds seem to be pretty consistent. My guess is there is a little exaggeration going on here.

 

[MDS]

The guy has been playing music for 30 years but that quote illustrates that he has absolutely zero knowledge of how sampling works. A 44.1 kHz sampling rate demolishes any part of the signal over 22.05 kHz NOT 10 kHz.

 

[mtrycrafts]

In the current issue of S & V, Tom Scholz of Boston says:

"Digital sounds a lot different than the original source. Things are further compromised when you decrease the resolution to 16 bits from 24 bits or higher. The combination of only 16-bit resolution and only 44.1-kHz sampling rate absolutely demolishes any part of the signal above 10k. If you put a 12k tone, which most people can hear, through a 16-bit, 44.1k sampling at digital conversion, you will be shocked at what that waveform looks like coming out the other end. You can put a pure tone in, and it comes out looking like some monstrous garbled thing......And if you changed that 12kHz signal just a percentage point, the signal that spits out will look entirely different. So every time somebody hits an "s" or a cymbal, or plays a delicate violin or even a raunchy distorted guitar with lots of high frequencies, the high-frequency end of that spectrum is completely mangled into something different. That's why people speak about strange sibilance, or things sticking out or not sounding "right," when they listen to CDs. It isn't right. It's completely different than the original recording."

The interview is with Mike Mettler, the current editor of S&V, who lets this stand without comment, indicating that he may agree with this.

He is out to lunch on this; perhaps on other things too but haven't followed his writings on them

He must have and still is sleeping or purposely not electing to read the most recent comparison of Red Book and hi-rez audio, under controlled conditions:
http://www.bostonaudiosociety.org/explanation.htm

This was also published in JAES but this is available on the net now.
A lot of participants, zero the number who could differentiate the hi res from Red Book. Lunch over

Maybe he will wake up from his snooze

 

[Waveform]

He is out to lunch on this; perhaps on other things too but haven't followed his writings on them

He must have and still is sleeping or purposely not electing to read the most recent comparison of Red Book and hi-rez audio, under controlled conditions:
http://www.bostonaudiosociety.org/explanation.htm

This was also published in JAES but this is available on the net now.
A lot of participants, zero the number who could differentiate the hi res from Red Book. Lunch over

Maybe he will wake up from his snooze

I was wondering why an ostensibly rational publication (i.e. one that does not do cable and interconnect listening tests, for instance) would at least not refer him to Nyquist.

 

[mtrycrafts]

I was wondering why an ostensibly rational publication (i.e. one that does not do cable and interconnect listening tests, for instance) would at least not refer him to Nyquist.

No idea, but if he is one of their writers, I guess he has some freedoms without being censored or corrected. Maybe they will get a bag full of mail on his article showing his mistakes

 

[mr-ben]

The guy has been playing music for 30 years but that quote illustrates that he has absolutely zero knowledge of how sampling works. A 44.1 kHz sampling rate demolishes any part of the signal over 22.05 kHz NOT 10 kHz.

Technically, it's not so simple. A 11khz tone sampled at 44khz means that one cycle must be defined by only 4 data points. How accurately can you represent a sine wave using only 4 points? The four points are evenly spaced apart, you don't know where the wave starts and ends, and it's constantly changing and mixed with other frequencies. A 22.05 khz frequency would need to be represented by two data points. Add in the fact that you only get 16 bits of accuracy in these points and I'm amazed that it works as well as at does.

 

[FatStrat85]

Technically, it's not so simple. A 11khz tone sampled at 44khz means that one cycle must be defined by only 4 data points. How accurately can you represent a sine wave using only 4 points? The four points are evenly spaced apart, you don't know where the wave starts and ends, and it's constantly changing and mixed with other frequencies. A 22.05 khz frequency would need to be represented by two data points. Add in the fact that you only get 16 bits of accuracy in these points and I'm amazed that it works as well as at does.

I think that was Tom's point. That still doesn't mean that you could tell the difference between a CD and something sampled at higher than 44.1 kHz, but the sound gets technically less and less accurate as you increase in frequency towards 22 kHz, regardless of whether or not we can hear the difference. I doubt Tom (or anyone else) could pass a double-blind test.

 

[MDS]

Technically, it's not so simple. A 11khz tone sampled at 44khz means that one cycle must be defined by only 4 data points. How accurately can you represent a sine wave using only 4 points? The four points are evenly spaced apart, you don't know where the wave starts and ends, and it's constantly changing and mixed with other frequencies. A 22.05 khz frequency would need to be represented by two data points. Add in the fact that you only get 16 bits of accuracy in these points and I'm amazed that it works as well as at does.

It seems that everyone points to the small number of samples at high frequencies and then concludes that it cannot possibly accurately represent the original analog waveform. How can you accurately represent a sine wave using only 4 points? Perfectly - in fact you need only two.

The reconstruction filter (or 'anti-aliasing') is going to filter out all frequencies above Fs/2 and 'connect the dots' so to speak. Regardless, claiming that audio data above 10 kHz is decimated and that anyone can hear it is way off base.

 

[mr-ben]

How can you accurately represent a sine wave using only 4 points? Perfectly - in fact you need only two. The reconstruction filter (or 'anti-aliasing') is going to filter out all frequencies above Fs/2 and 'connect the dots' so to speak.

Can you please expand on this, or point to a good explanation somewhere? I've never been able to understand how this is possible. It seems like two points could be translated into virtually anything, and a certain amount of surrounding context is necessary. Something like cymbals would look relatively random, and the DAC isn't aware of the frequencies it's looking for.

 

[MDS]

There is an article on this site called 'Digital Audio Myths' or something like that that explains it pretty nicely without the math behind the sampling theorem or fourier analysis.

 

[FatStrat85]

http://www.audioholics.com/education/audio-formats-technology/exploring-digital-audio-myths-and-reality-part-1/?searchterm=digital audio myths

Very interesting read. Thank you, MDS. I wish "Figure 4" wasn't missing though...

 

[mtrycrafts]

Can you please expand on this, or point to a good explanation somewhere? I've never been able to understand how this is possible. It seems like two points could be translated into virtually anything, and a certain amount of surrounding context is necessary. Something like cymbals would look relatively random, and the DAC isn't aware of the frequencies it's looking for.

This is the first link Googling 'sampling theorem'

http://en.wikipedia.org/wiki/Nyquist–Shannon_sampling_theorem

have fun.
There are pages and pages of explanations why and how it works perfectly well.
But then, a recent DBT comparing hi res sampling to 44.1 showed no audible differences.

Not sure what will convince you, Mr Ben Maybe you can tell us what will?

 

[mr-ben]

Thank you everyone for the links - they were really helpful. Unfortunately I'm a big fan of experimentation over theory, and so until I can try this myself I can't be sure about it, but I can see how it would work now.

I'm not really familiar with S&V, so the fact that the editor didn't comment may be due to politeness, laziness, or just the way they operate, not necessarily agreement. Chris Tham makes a point of stating in the audioholics article "By the way, the above is a screen capture of Adobe Audition, which is the only digital audio editing tool I have encountered that actually correctly draws the analog reconstructed waveform in between sample points.", and so I can imagine that Tom Scholz is basing his observations on the display of some inferior software, rather than on actual measurements. He's a musician, not a recording engineer, and may not have enough experience looking at this type of software to recognize it.

 

[skers_54]

If anyone has access to MATLAB (doubtful) it can draw some very nice figures illustrating this point. We used it extensively in my signals and system analysis class last semester. The theory (Nyquist and Fourier, among others, which are not easily explained without detailed graphical examples unfourtunately) I learned flies directly in the face of the claims maid by Tom Sholz. The interpolation formula describes how to reconstruct a signal from sampled values without error provided no aliasing occurs. There is some validity in the claim, as no real signal can truely be band-limited and completely avoid aliasing. By including frequencies above the range of human hearing, the aliasing is limited to inaudible portions of the signal and has minimal, if any, affect on the sound. If it did, I would suspect the Redbook sampling frequency would be higher than 44.1kHz