ORIGINAL: DonM

I would say the following as an analogy - If you took a person from a remote area of the planet and showed them a 4:3 tube NTSC analog TV and then showed them a Blu-Raw FPHD TV - they would possibly draw no distinction between what they just saw. I also suggest if you watched that old TV for a long enough time you'd get accustom to it's ability even if you currently are an FPHD viewer.

Good point. I seem to remember reading that when Edison (?) first demonstrated a wax cylinder recording playing along with a silent movie, people checked behind the curtains to see if there really wasn't a piano player there. I doubt anyone would be fooled by the sound of a wax cylinder today!

UnderTow

Wouldn't 24/96 be better for car stereo so you can over power the road noise?

ORIGINAL: John

I think a really important aspect of doing comparative work is spending a long time saturating yourself in the format before you go back. For me the quick A/B switch has taught me very little in life. I need to listen to speaker set A for several days with multiple mixes before I compare speaker set b - the same is with formats in my opinion. You'd be surprise what you hear when you spend the time.

-D

Now this is interesting because it is as apposed to a double blind approach as you can get. I do know what you're saying though I have experienced the same thing. I would never trust it though.

Not entirely. I've always considered training as an important part of double blind testing. The subjects need to familiarise themselves with whatever they are testing. Simply A/Bing unknown material with an unknown variable will not tell much but that doesn't mean that A/Bing in itself is useless.

UnderTow

ok, I want to see the list... so I can drool!

I don't get it.

A/B is how I make many decisions in recording. If you are specifically talking about sampling rates, and you A/B the same drum, with the same mic, with the only variable being sample rate, you are conducting the simplest of A/B tests. They either sound similar or not. And chances are that one of those two choices will sound better to you.

The question is answered with my ears. Do I like THIS or THAT?

And if I decide that I like THAT because it sounds better to me, I choose to use THAT. And I live with my decision.

Getting all hung up in "theory" or "math" seems awfully silly to me.

ORIGINAL: UnderTow

For instance, you could make a recording at 96khz, sample rate convert the recording to 48Khz with an offline SRC optimised for quality rather than performance. (Maybe one with a very complex and precise filter that takes an hour to convert a 1 minute recording). Then reconvert back to 96Khz. There will be no frequencies above 24Khz. Any of the so called advantages of recording at 96Khz have been removed.

Play the original and the reconverted 96Khz files back through the converter at 96Khz and you won't hear a difference. Now take the 48Khz file and the reconverted 96Khz file and play them back through a converter at 48Khz and 96Khz respectively. If you hear a difference, you know for a fact that the difference is caused by the (less than perfect) converter and has nothing to do with the intrinsic advantages, presumed or not, of recording and playing back at 96Khz. You ears alone could never tell you all this

Excellent example. Another aspect is that all of skulls recordings are frequency limited by the mics and instruments he is using(I doubt he is recording instruments with a 48Khz freq range). How can you hear a difference if there is no freq content above 20Khz?...the 48Khz and 96Khz files should sound identical. I think Bob Katz hits the nail on the head wrt 96Khz...it is converter filter quality. The less steep AA filters needed for 96Khz vs the steep AA filters needed for 44.1Khz are what really cause the difference in perception. A crappy 44.1Khz filter will make the 96Khz sound better when the *file* bit depth has nothing to do with it. Full disclosure: I have never spent any significant time experimenting with or recording with my gear at 96Khz for a variety of reasons(even though it is capable) and I respect the positions of experts like DonM here but I err on the side of science on this topic...(does it help that I have a degree in Applied Mathematics? )

ORIGINAL: DonM

For me engineering is Art and Science. Yin/Yang - they flow and certainly at different moments one is more 'dominant' than the other while a balance seems to remain.

I definitely indentify with that. And all music *is* math(frequencies, harmonic relationships. modes, etc).

I think I love the math (as much as the art) for two reasons - First - y'all know I teach grad and undergrad - there is this very 'academic' attitude those universities have...and math seems to play right into that attitude - the other reason is .... it seems everytime I understand the science or algorithmic aspect of digital audio it 'frees' me further to get a better artistic result - period.

Absolutely.

I also want to add - that I humbly appreciate the kind comments some folks make about my participation here - honestly - I am no expert - I am a learner with a deep passion for this work.

Well, you may be a "learner" in the sense that all of us, as we get older, realize that you never stop learning...but you are definitely a cut above IMO(and the risk of turning this thread into a hugfest...). Most of us arn't professors who actually *do* as well as teach. This is why your statement about "immersion then removal" and then hearing a difference regarding 96Khz intrigued me. I am assuming you have top of the line converters so the Bob Katz theory doesn't seem to fit for your use case. I need to experiement with 96Khz...

On the verge of the starting semester... my students are 'required' to read this forum - everyone of you folks contributes to their education and I appreciate that more than I can say - especially since you've all contributed to mine.

Poor kids... If I was in your locale I think I would try to take you classes just for the fun of it...

ORIGINAL: AndyW

ORIGINAL: UnderTow

For instance, you could make a recording at 96khz, sample rate convert the recording to 48Khz with an offline SRC optimised for quality rather than performance. (Maybe one with a very complex and precise filter that takes an hour to convert a 1 minute recording). Then reconvert back to 96Khz. There will be no frequencies above 24Khz. Any of the so called advantages of recording at 96Khz have been removed.

Excellent example.

Actually the claim "Any of the so called advantages of recording at 96KHz have been removed" presumes too much. The reason is that more than one sample was used for creating each new sample in the new audio clip. 96 kSamples/sec recordings contain redundancy if the frequencies are far below the Nyquist limit. This redundancy is usually considered a good thing, and it is not lost somehow; it is incorporated into the lower-rate clip during resampling. It is a serious misunderstanding to assume that because there is nothing above the new Nyquist limit, benefit is lost --- given that the coefficients for anything above the new Nyquist limit were zero (neglecting "noise") to begin with.

-----------------

My personal and general reason for working at 96/24 has nothing to do with what most people argue about (recording and playback, particularly listening) which I see as only muddying the real issues; rather it has to do with the processing that I use. I preserve the audio in the most accurate format I can stand, given CPU performance, storage limitations, and the unfortunately limited understanding of some audio programmers for as long as possible, then downsample. People differ on how much they "can stand" so I don't seek to convert anyone; however, I do take issue with the absurd notion that 96/24 has no advantages and am tempted to give the equally absurd response that 44.1/16 has no advantages.

Regards,
Dave Clark

ORIGINAL: DaveClark


Actually the claim "Any of the so called advantages of recording at 96KHz have been removed" presumes too much.

Perhaps. But I think it is more lilkely to be correct than not...see below.

The reason is that more than one sample was used for creating each new sample in the new audio clip. 96 kSamples/sec recordings contain redundancy if the frequencies are far below the Nyquist limit. This redundancy is usually considered a good thing, and it is not lost somehow; it is incorporated into the lower-rate clip during resampling. It is a serious misunderstanding to assume that because there is nothing above the new Nyquist limit, benefit is lost --- given that the coefficients for anything above the new Nyquist limit were zero (neglecting "noise") to begin with.

It is also a serious misunderstanding to assume more samples is necessarily better. For example, the vast majority of people can distinguish flicker in a 15fps video. Most cannot at 30fps(hence the current video standards). In video games, 60fps is commonly considered the most you need(ie: you cannot really tell the difference between 60fps and 120fps). Aural acuity is no different scientifically. Scientifically I do not believe it is possible to distinguish a 1Khz sine wave sampled at 22.05Khz, 44.1Khz or 96Khz. There have been tests that address this(athough I haven't looked recently) Basically it is beyond our aural acuity to distinguish between high audio sample rates *until* frequency response becomes the factor...and then there is a frequency response above which our aural acuity cannot distinguish either.

My personal and general reason for working at 96/24 has nothing to do with what most people argue about (recording and playback, particularly listening) which I see as only muddying the real issues; rather it has to do with the processing that I use. I preserve the audio in the most accurate format I can stand, given CPU performance, storage limitations, and the unfortunately limited understanding of some audio programmers for as long as possible, then downsample. People differ on how much they "can stand" so I don't seek to convert anyone; however, I do take issue with the absurd notion that 96/24 has no advantages and am tempted to give the equally absurd response that 44.1/16 has no advantages.

No intent to argue here either. "Most accurate method" is a good argument...the debate to me has always been the costs of that additional accuracy and whether that accuracy was worth it. Kinda like grenades...does it matter if it was one inch or two inches from the target?

Regards,
Dave Clark

ORIGINAL: mike_mccue

Just a comment about game FPS. You may not visually see a difference between 60 and 120fps, but you will experience much better gaming on the 120fps system. All your input and the response of the gaming environment will be enhanced and you will perceive a much smoother interaction with the game. I used to do some developing for race games... the difference between 60fps and 120fps might for example be the factor that lets you control the vehicle and keep it on the track while slower rates might take your input and send you off the track before you ever get a chance to correct it.

In most cases, if you Vsync the game to 60 or 75Hz those other things will still be as responsive(because I understand what you are saying). The problem with average frame rates in games is that it is just that...average. Faster CPU's and GPU's will allow you to have higher average fps therefore more responsiveness, yet, if you Vsync the video to 60Hz there is little to no difference in playabiliy and perception.

It's not just about persistence of vision.

Well, pick your metaphor...I am talking strictly about visual acuity because it is what is most relevant to audio acuity. Throwing "responsiveness" into it involving video games just muddies the water.

I learned this while wondering why anyone would need a game frame rate beyond 24-30fps... but now that I've experienced it I appreciate the higher fps.

Game fps rates are actually a suitable metaphor for my thoughts about more data gives you more accurate filtering.

I'm a gamer...I understand. your methaphor is really about system capability...better capability, better filtering, better responsivness. But the better capability also results in faster fps. fps and responsiveness in games are coorelated but not necessarily causal to each other, to borrow statistical terms. Something more related to the current topic would be gaming mouse dpi...can you really tell the difference between a 2400dpi and 4000dpi mouse? same discussion, different topic group.

ORIGINAL: DaveClark

Actually the claim "Any of the so called advantages of recording at 96KHz have been removed" presumes too much. The reason is that more than one sample was used for creating each new sample in the new audio clip. 96 kSamples/sec recordings contain redundancy if the frequencies are far below the Nyquist limit. This redundancy is usually considered a good thing, and it is not lost somehow;

What do you mean by redundancy? Can you clarify because this doesn't make any sense to me.

Anyway, more than one sample is used either way. This is always the case with any modern ADC.

Whether the AD converter reduces the number of samples by taking the oversampled signal and decimating it straight down to 48Khz or whether you use an intermediate 96Khz step and use an offline SRC to do the last step, the principles remain the same. The advantage of adding the last step as a separate process is that you can throw a lot of DSP at the problem and are not limited by the realtime (and cost) constraints of an ADC.

My example in the previous post is a way to demonstrate that the main issues with 1 FS sampling (as a target rate for the ADC) are those realtime and cost constraints.

it is incorporated into the lower-rate clip during resampling.

How so? All intermediate samples have been discarded and anything above 1/2 Nyquist is gone.

It is a serious misunderstanding to assume that because there is nothing above the new Nyquist limit, benefit is lost --- given that the coefficients for anything above the new Nyquist limit were zero (neglecting "noise") to begin with.

What do you mean by "coefficients"? Coefficients of what? This is an ambiguous term so I don't know what you mean.

If you make a claim like "It is a serious misunderstanding" you really need to explain yourself.

My personal and general reason for working at 96/24 has nothing to do with what most people argue about (recording and playback, particularly listening) which I see as only muddying the real issues; rather it has to do with the processing that I use.

There are good reasons to do processing at higher sampling rates, agreed. Still, talking about the theory is only muddying anything if people misunderstand what is being discussed.

I preserve the audio in the most accurate format I can stand, given CPU performance, storage limitations, and the unfortunately limited understanding of some audio programmers for as long as possible, then downsample.

Increasing the sampling rate in itself does not increase storage accuracy within the audible range! It can increase accuracy (for instance, due to the lower constraints put on the filter design) but it is not a given.

however, I do take issue with the absurd notion that 96/24 has no advantages

I didn't see anyone claiming that there are no advantages to increasing the sampling rate. The only discussion is what those advantages are and whether there are not better ways to deal with any of the issues that increasing the sampling rate is addressing.

UnderTow

Hi Andy - Undertow,

It could be that we are just talking past each other --- for example, using different terminology for the same thing --- but I think there may still be some misunderstanding of technicalities. At this risk of creating even more unnecessary confusion (isn't this the thread in which I excused myself, not wanting to get bogged down???):

ORIGINAL: AndyW
It is also a serious misunderstanding to assume more samples is necessarily better.

I never claimed that and, in fact, studiously avoided saying that because it is too often not true. The topic is how accurately the downsampled version represents all of the information in the original, not how good or bad the original is. I would agree, though, that some folks may not realize that higher sampling rates can be worse under certain circumstances, so it's probably good that you (Andy) mentioned it.

On redundancy: If, in the frequency representation, the upper half of the coefficients (*) are zero, there is redundancy present. More values are present in the time domain than are required to exactly represent the signal (per the sampling theorem).

(*) I myself often use the term "coefficients" for the frequency representation based on the Fourier expansion. I believe many others do this, also, but I haven't really counted heads because nobody has ever requested clarification on this before. I apologize for any confusion.

ORIGINAL: Undertow
All intermediate samples have been discarded....

No, they are not (simply) "discarded." They are utilized in making new estimates of values. It is crucial to understand this difference. I see that you (Undertow) do understand that several values are used in each estimate, but I see you using the word "decimate" and "discarded." I'm not sure you fully appreciate the distinction I'm drawing, so am pointing it out once again. At the very least, this is a poor choice of words, IMO --- even worse than my use of the term "coefficients!"

ORIGINAL: Undertow

Increasing the sampling rate in itself does not increase storage accuracy within the audible range! It can increase accuracy (for instance, due to the lower constraints put on the filter design) but it is not a given.

Not by itself, correct. To fully appreciate the advantages of higher sampling rates one must understand the process as a whole, not focus in on listening tests alone, recording alone, etc. as is often done in discussions on the merits and demerits of using higher sampling rates. Too many folks simply record something at two rates, then have a group of people listen who turn out to be incapable of distinguishing between the two recordings. The "investigators" then claim that they don't see any advantage in using higher rates. The advantages of using higher sampling rates are at best quite small at each step, accumulating over the entire production, so this method of proof (disproof) is pretty much invalid.

To really understand the origins of the advantages of higher sampling rates (not guaranteed as Andy said, but probabilistic), I think it helps to have taken courses in numerical analysis (or studied it independently), statistics, signal processing, and who knows what else....

It also helps to have actually worked extensively with higher rates. I think DonM's description --- on this thread, I believe --- was very true. While working at the lower rates, I really am not bothered by them because, as Don says, one becomes acclimatized. However, later when reviewing the end results of many projects at the same time, and forgetting which is which, the ones utilizing lower sampling rates invariably don't sound good to me by comparison --- over many genres, styles, instruments, etc. As I said previously or elsewhere, I have asked myself, "What is WRONG with these?" before recalling what I had done.

ORIGINAL: Undertow

I didn't see anyone claiming that there are no advantages to increasing the sampling rate.

Not explicitly here in this thread, correct.

Regards,
Dave Clark

Andy don't feel alone in this I see it much as you do. I am sure there are many others that also share the same view.

Hi Dave,

ORIGINAL: DaveClark

On redundancy: If, in the frequency representation, the upper half of the coefficients (*) are zero, there is redundancy present. More values are present in the time domain than are required to exactly represent the signal (per the sampling theorem).

(*) I myself often use the term "coefficients" for the frequency representation based on the Fourier expansion. I believe many others do this, also, but I haven't really counted heads because nobody has ever requested clarification on this before. I apologize for any confusion.

Aha! There was no mention of Fourier in your previous post!

I still find the term confusing though. In my terminology the coefficients are what are used to achieve the results. So, I would much prefer using terms like "levels in the upper bands" or "energy in the upper bands" or similar.

Anyway, your statement still doesn't make sense to me. You wrote "It is a serious misunderstanding to assume that because there is nothing above the new Nyquist limit, benefit is lost --- given that the coefficients for anything above the new Nyquist limit were zero (neglecting "noise") to begin with." One of the main reasons to use a higher sampling rate is to allow a shallower filter curve in the anti-aliasing filter so there would have been energy above the new Nyquist limit.

In other words, a shallower filter curve would result in there being energy above 24 Khz in a 96 Khz sampling rate.

ORIGINAL: Undertow
All intermediate samples have been discarded....

No, they are not (simply) "discarded." They are utilized in making new estimates of values. It is crucial to understand this difference. I see that you (Undertow) do understand that several values are used in each estimate, but I see you using the word "decimate" and "discarded." I'm not sure you fully appreciate the distinction I'm drawing, so am pointing it out once again. At the very least, this is a poor choice of words, IMO --- even worse than my use of the term "coefficients!"

Dave, once the file is converted to 48Khz, those samples will have been discarded regardless of anything else! (Regardless of whether they were used or not). This is not an incorrect use of the term.

Also I believe that this was all made clear both in the original example that you quoted and responded to and in the follow-up post. As I said, either route you take, more than one sample was used to achieve any single sample value at the final sample rate target which, as far as I can see, invalidates the redundancy argument. When going straight to 48Khz from 6.144 Mhz (or whatever), there are more samples then when you go from the intermediate 96Khz rate.

I still fail to see what advantages there are left for sampling in my particular example. Can you clarify?

ORIGINAL: Undertow

Increasing the sampling rate in itself does not increase storage accuracy within the audible range! It can increase accuracy (for instance, due to the lower constraints put on the filter design) but it is not a given.

Not by itself, correct. To fully appreciate the advantages of higher sampling rates one must understand the process as a whole, not focus in on listening tests alone, recording alone, etc. as is often done in discussions on the merits and demerits of using higher sampling rates. Too many folks simply record something at two rates, then have a group of people listen who turn out to be incapable of distinguishing between the two recordings. The "investigators" then claim that they don't see any advantage in using higher rates. The advantages of using higher sampling rates are at best quite small at each step, accumulating over the entire production, so this method of proof (disproof) is pretty much invalid.

Did you skip some of the posts in this thread or did you read the whole discussion? That might explain some of the confusion. I believe this has all been said.

UnderTow

EDIT: Broken Italic

ORIGINAL: DaveClark

Yes, I've read it. Yes, I agree. Wonderful book. The point I tried to make is that the results of simple listening tests (record/playback) are not a "proof" or "disproof" of benefits of working at higher sampling rates for an entire project, but people use them as such.

I agree...but I viewed that the discussion here was about *audible* benefits to working at higher sample rates, which to me is comparing "identical" recordings, one at 96Khz, one at 44.1Khz.

Too often I have heard people say, "If it can't be measured, it isn't truly different" when it's inappropriate. The fact that a group of listeners cannot hear a difference in one or many experiments does not equate to "can't be measured" and should not be offered as such. Furthermore, results that are inaudible to many people as a single event are not necessarily inaudible when compounded. If a 0.1 db reduction is done enough times, anyone who is not deaf can eventually hear a difference.

But it is appropriate here because I am talking not just statistical measurement...but *scientific* measurement.

In case there is any confusion, I should also mention that when I do my accidental comparisons discussed either above and elsewhere, they are ALL either MP3 files at 320 kbits/sec or CD's burned at 44.1/16, i.e. equivalent format. Also, many of these projects use samples that were originally at 44.1/24 or even 44.1/16, but upsampled by me. So for my case, it's neither A/D nor D/A. Either I'm imagining things or there is an audible difference between working at higher versus lower rates in the end product.

If it goes through converters so you can hear it, it is D/A...and there is good evidence that it is the *filtering*, not the source sample rate that is the issue. I believe you hear an audible difference...I just disagree on *why* you are hearing an audible difference.