How does the stereo image in digital audio shrink?

[Ethan Winer]

 

Let's assume you don't believe it's possible that the "lots of dithering with Pro Tools causing subtle monoization" could be true

 

 

That all makes sense as an extreme example of what might be possible. But it doesn't address the core issue of this thread, where it is claimed that a single conversion to digital causes shrinkage. This is what Beck and John Sayers are saying - that a digital recording of a tape or LP loses width.

 

Also, I would never suggest that guitar cables used with a passive guitar can't sound different. My examples always relate to speaker cables, and sometimes line level wire when connected to "normal" gear having a suitably low output impedance. There are also lousy cables that have too much capacitance, such as audiophile "flat" speaker wires with parallel flat conductors. In that case those are much worse than zip cord, which is what the high-end speaker wires claim they surpass.

 

--Ethan

[Anderton]

Zip cords dont' sound good until you unzip them, at which point they're uncompressed

[Rudolf von Hagenwil]

I recorded a full width stereo test file which permits to measure how by a stereo master tape transfer to digital the width get shrinked. I measured the original file.

So, if anyone records that file to his tape machine, then convert it to digital, then give the digitized file back to me, I can post how much narrower it is.

Anyone?

[MikeRivers]

 

I recorded a full width stereo test file which permits to measure how by a stereo master tape transfer to digital the width get shrinked. I measured the original file. So, if anyone records that file to his tape machine, then convert it to digital, then give the digitized file back to me, I can post how much narrower it is.

 

How do you measure width? And more important, how do you measure depth (or "three dimensionalness")?

 

If you're recording a stereo file with a completely independent source on each track, then looking for how much of one tracks's spectrum is on the other track, then you're measuring crosstalk. We haven't clearly established yet that this is the mechanism (and the only mechanism) that causes "digital shrinkage." I can't think of anything other than crosstalk that you could measure and express quantitatively, but I'm willing to hear your preposed experiment.

 

Now I KNOW that an analog tape deck will have crosstalk. If anyone who wants to give it a listen) says "this hasn't shrunk" then we can conclude that crosstalk is NOT the cause of the effect about which we're arguing.

 

So what's your threory?

[MrKnobs]

Well that was Craig's exact point - you have to KNOW the exact conditions that are known to create it. A lot of people do tests without actually knowing that, examples of which he illustrated in his post.

 

It's helpful to know the exact conditions that created an effect, but not always necessary. What's necessary is that it be reproducable so it can be tested.

 

"You say a poltergeist moved your mirror? How often does this happen? What, only the one time? Sorry ma'am, can't help you."

 

"You say your analog recording of a Leslie cab sounds less spacious when you convert to to CD format? And this happens every time? Interesting! May I have two copies of the recording, one with the 3D effect and one that has lost it after conversion? What, you only have the one without the effect?" :

 

 

Terry D.

[Rudolf von Hagenwil]

So what's your threory?

I recorded music which has an unambiguous localization. With the balance meter I can measure to which side the audio is biased. The correlation meter tells me how much similarity there is between the two channels. The phase scope (see image below) displays the bias. So we can compare how much narrower the stereo recording would be after an analog to digital transfer

[Jimbroni]

It's helpful to know the exact conditions that created an effect, but not always necessary. What's necessary is that it be reproducable so it can be tested. Terry D.

 

 

Wrong. Its pertinent in the scientific realm to understand the facts, the details, the exact conditions. Studying aeroacoustic effects on cars taught me that there just isn't a way scientificly to reproduce certain environments. But that doesn't change the fact people complain about whistling roof racks, no matter how much you pay lockhead to test them in the wind tunnel. Me personally I consider that a failure of science and not a failure of peoples observation. The same thing is going on in these discussions. I call it flat out lazy science that somehow thinks it knows it all, when it doesn't. And secondly the most necessary thing in the world of music is that it sounds fantastic.

[Jimbroni]

To get back on topic. I don't know that I buy the stereo width comments, I don't even see how its possible nor have I heard this difference. But what I have heard, and do buy into is the lack of depth in the digital realm.

The way the dynamic range is calculated in digital its broken up into chunks, and in the analog world its continuous. Even if the dynamic range of analog is less, within that range you have more flexibility that simply isn't there in digital. When you can totally alter the sound of a track by moving an effect or fader by 0.1, I question whats in between.

I think the stereo width problem in digital has more to do with dynamic resolution, than the actual width. And while 24 bit give you more headroom to deal with this, it doesn't give any more resolution. I feel as though there needs to be sub bits, if only to make digital mixer less sensitive to small movements thus making them more natural.

[Dean Roddey]

Even if the dynamic range of analog is less, within that range you have more flexibility that simply isn't there in digital. When you can totally alter the sound of a track by moving an effect or fader by 0.1, I question whats in between.

 

 

In a 24 bit system, even if you are recording fairly low, there are probably close to a million increments between your peaks and wherever the bottom is. If you move the the fader by .1dB, there's a LOT of increments in that movement. So obviously the non-continuous digital representation is not the problem wrt to that. If you are peaking at -6dB, then you are down basically about 1 bit, so that would leave 1 million increments from there down.

[MrKnobs]

Wrong. Its pertinent in the scientific realm to understand the facts, the details, the exact conditions. Studying aeroacoustic effects on cars taught me that there just isn't a way scientificly to reproduce certain environments. But that doesn't change the fact people complain about whistling roof racks, no matter how much you pay lockhead to test them in the wind tunnel. Me personally I consider that a failure of science and not a failure of peoples observation. The same thing is going on in these discussions. I call it flat out lazy science that somehow thinks it knows it all, when it doesn't. And secondly the most necessary thing in the world of music is that it sounds fantastic.

 

Not sure where all that came from, my post wasn't intended to be arrogant.

 

This isn't rocket science (or even aerodynamics), it's signal processing. See Angelo's post above for an intelligent approach to the problem.

 

Again, this small, well-defined problem has little to do with music. The statement was made that spatial information is apparently lost when downsampling to 44.1/16. That's measurable.

 

If "Signal A" produces a certain effect and "Signal B" does not, then, barring external testing bias, Signal A and Signal B must be different. The differences may be very subtle, but they're there. The task then is to compare the signals and find the differences. Once that's done, then the differences are controlled and tested to find which are significant to the listener.

 

Which part of "not always necessary" wasn't clear?

 

I say that because in my line of work (and it sounds like it's the case in your line as well), we rarely have the luxury of knowing all the facts and conditions which have caused a problem we must fix. If we knew that going in, most of our work as researchers would already be done.

 

My points were (or were intended to be) you don't have to (a) understand the mechanism of something before you can test it empirically, (b) you have to reproduce the problem in order to study and eventually understand it. If you can reliably reproduce the problem you can then determine the contributing causations by careful experiment design.

 

Now I didn't say it would be easy. Almost every problem seems easy before you start working on it.

 

Terry D.

 

P.S. I'd really like to hear the before and after tracks, but it sounds like the before tracks are unavailable?

[Jimbroni]

Not sure where all that came from, my post wasn't intended to be arrogant. This isn't rocket science (or even aerodynamics), it's signal processing. See Angelo's post above for an intelligent approach to the problem. Again, this small, well-defined problem has little to do with music. The statement was made that spatial information is apparently lost when downsampling to 44.1/16. That's measurable. If "Signal A" produces a certain effect and "Signal B" does not, then, barring external testing bias, Signal A and Signal B must be different. The differences may be very subtle, but they're there. The task then is to compare the signals and find the differences. Once that's done, then the differences are controlled and tested to find which are significant to the listener. Which part of "not always necessary" wasn't clear? I say that because in my line of work (and it sounds like it's the case in your line as well), we rarely have the luxury of knowing all the facts and conditions which have caused a problem we must fix. If we knew that going in, most of our work as researchers would already be done. My points were (or were intended to be) you don't have to (a) understand the mechanism of something before you can test it empirically, (b) you have to reproduce the problem in order to study and eventually understand it. If you can reliably reproduce the problem you can then determine the contributing causations by careful experiment design. Now I didn't say it would be easy. Almost every problem seems easy before you start working on it. Terry D. P.S. I'd really like to hear the before and after tracks, but it sounds like the before tracks are unavailable?

I wasn't calling you arrogant. You've maintained a more open mind, then some others. And I agree Angelo's idea seems to be a good one, however the problem is a bit like using the fox to watch the hen house. Using Digital equipment that employs the same mathematical techniques to measure digital sampling is a biased test, IMO. To me if you sample something in digital, then turn around and analyze the signal with equipment that uses the exact same transforms, well of course it will not show a difference. Because the same techniques used to sample the signal with whatever shortcomings it may have are the ones being used to measure and analyze it. Like I said, Fox watching the Hen house.

[Jimbroni]

In a 24 bit system, even if you are recording fairly low, there are probably close to a million increments between your peaks and wherever the bottom is. If you move the the fader by .1dB, there's a LOT of increments in that movement. So obviously the non-continuous digital representation is not the problem wrt to that. If you are peaking at -6dB, then you are down basically about 1 bit, so that would leave 1 million increments from there down.

 

 

I'm not sure what you're getting at here. If you move a fader by .1 dB and that in turn affects the signal enough to be quite audible, then well the dynamic resolution is not all that fine, IMO. In fact, I find this to be the most annoying thing of working with digital.

[Lee Flier]

 

And I agree Angelo's idea seems to be a good one, however the problem is a bit like using the fox to watch the hen house. Using Digital equipment that employs the same mathematical techniques to measure digital sampling is a biased test, IMO. To me if you sample something in digital, then turn around and analyze the signal with equipment that uses the exact same transforms, well of course it will not show a difference. Because the same techniques used to sample the signal with whatever shortcomings it may have are the ones being used to measure and analyze it. Like I said, Fox watching the Hen house.

 

 

Yeah, this was something I wondered about as well, and one of the reasons I asked Terry how he would quantify the stereo field and what equipment he would use to measure it. It would seem that if you used a digital scope to measure an analog tape master you would be defeating the purpose... you may well not be able to measure that which gets "lost" in the translation to digital because the scope would already have translated it and thus already "lost" whatever might be getting lost.

 

Of course, Angelo might well have an analog scope meter.

[blue2blue]

This thread cracks me the hell up.

[MrKnobs]

Yeah, this was something I wondered about as well, and one of the reasons I asked Terry how he would quantify the stereo field and what equipment he would use to measure it. It would seem that if you used a digital scope to measure an analog tape master you would be defeating the purpose... you may well not be able to measure that which gets "lost" in the translation to digital because the scope would already have translated it and thus already "lost" whatever might be getting lost. Of course, Angelo might well have an analog scope meter.

 

It really wouldn't matter which you use, but we can try both.

 

I've been super busy out on the road of late, or I'd have answered your question in depth earlier. The weather has been conducive to recording road noise (yay! ) and in my biz you gotta make hay while the sun shines.

 

Anyway, my initial plan (the initial plan never survives contact with the enemy, as they say) would be to use computer software to do a bit comparison of the two files, the one that has the "3D" sound and the one that's lost it somehow do to downsampling. The program would calculate the differences sample by sample and then display those in various graphic ways. This is a standard practice we use for many things. You may recall Craig posting some downsampling accuracy plots for various DAW software a while back.

 

Now, note that my plan has never been to say, "Oh, look! The bits are the same between the two files, therefore you are all deluded and don't really hear anything, case closed. My plan has been to find the differences in the files that you're hearing. I hope you see the difference.

 

Next, I think a vectorscope is very good, because one would expect that differences in perceived "3d" soundscape would have to come in the time domain. The vector or phase display is designed to do exactly that, and every mastering lab I've ever used has one and uses it religiously. And yes, they used to have analog scopes to do this, and now they mostly have digital computer programs.

 

To suggest that a high quality digital vectorscope gives a different answer than the analog version is a huge claim, one that has a very low probability of being true. These things are and have been in use for many years in television, audio, and labs and not only was this tested rigorously when they were designed but someone would have noticed the problem long ago.

 

I can't vouch for any specific brand of DAW/mastering software that provides this function, these change constantly and it's outside my field.

 

Again, though, the focus is not to use technology to disprove anyone's hearing experience, it is to understand it.

 

Terry D.

 

P.S. Funny story from the road yesterday: my assistant and I traveled to Waco (about 100mi from Austin) with the assigned goal of testing the noise on some old concrete pavement that had been diamond ground to restore the original smoothness. Our sponsor wanted to know if the noise was also reduced.

 

We have some permeable pavement test sections there nearby on the same Interstate that we tested last in 2006, so we figured why not retest those while we're on location and see if they've become louder with wear.

 

When we arrived and drove the loop, my partner said, "Damn, they milled only half the stretch of old concrete and left half of it crappy like it was. Worse, they covered over half our permeable test sections with brand new permeable pavement and left half of it like it was."

 

I don't know what he was thinking, he's usually a really bright guy. But what a researcher's dream! The DOT for their own reasons, had created a perfect experimental test bed of before and after sections that were contiguous and exposed to the same traffic. Best of all, they were all in a 6 mile loop we could drive over and over again in a circle taking measurements without even leaving the vehicle.

 

True, we had to stop at Home Depot for some paint to mark all the new sections, and we used the GPS to find our old, buried sections, but then it was all cake as we spent the next two hours collecting a ton of data. And it didn't even rain on us!

 

So, before and after - just like the Leslie file downsampled to 44.1.

[Dean Roddey]

I'm not sure what you're getting at here. If you move a fader by .1 dB and that in turn affects the signal enough to be quite audible, then well the dynamic resolution is not all that fine, IMO. In fact, I find this to be the most annoying thing of working with digital.

 

 

I don't even know how to respond to this without being potentially offensive really.

 

A) It's very unlikely you can actually hear a .1dB difference, and are probably just convincing yourself that you are.

B) It has nothing to do with digital having limited dynamic resolution either way. It has resolution far beyond the human ear to hear the difference between adjacent levels, clearly.

C) If you can in fact hear such a thing on a digital system, it would be because of greater accuracy in the digital system, not the other way around. A .1dB difference is the same difference in either system. Given that it would represent probably a change of tens of thousands of sample levels, obviously it's not because you suddenly moved from one big, chunky sample level another.

 

So I just have a problem with your whole premise really.

[Lee Flier]

I've been super busy out on the road of late, or I'd have answered your question in depth earlier. The weather has been conducive to recording road noise (yay! ) and in my biz you gotta make hay while the sun shines.

 

Hehe... no problem... hey, we're all busy. I owe several other people responses in this thread myself.

 

To suggest that a high quality digital vectorscope gives a different answer than the analog version is a huge claim, one that has a very low probability of being true. These things are and have been in use for many years in television, audio, and labs and not only was this tested rigorously when they were designed but someone would have noticed the problem long ago.

 

Be that as it may, let's consider the specific claim being made here: that transferring a recording from analog to digital "loses" something (in this discussion, the "something" being width and depth in the stereo field). Given that, digitizing the "before" analog test track at any point would seem to be a very fundamental error in the test.

 

I'd be curious to see the results with both types of scopes though.

 

Of course if you use an analog scope, I presume you'd have to use it on the decoded signal in the case of the digital track, in which case the DAC becomes a factor as well.

[John Sayers]

I suppose we could always ask Han to get out his Hammond and Leslie and repeat the original analogue recording style. Then he could sample it in 24/192 (what Rupert Neve reckons is the closest digital will get to analogue), then 24/96 and finally 16/44.1.

[Han]

I suppose we could always ask Han to get out his Hammond and Leslie and repeat the original analogue recording style. Then he could sample it in 24/192 (what Rupert Neve reckons is the closest digital will get to analogue), then 24/96 and finally 16/44.1.

 

 

The problem is that a Hammond doesn't produce a very complicated sound. It doesn't contain any high freq.

 

You'd better get one of these Sheffield Labs DC vinyl records, get a really good recordplayer and listen to it very carefully. The James Version by Harry James is the best DC I've ever heard.

 

And next convert it to whatever digital format you have at hand and listen very carefully again. That will make discussions like these not neccesary.

[MrKnobs]

Of course if you use an analog scope, I presume you'd have to use it on the decoded signal in the case of the digital track, in which case the DAC becomes a factor as well.

 

Now you're giving me scientist deja vu!

 

In the research community (as is sort of the case here), everyone has theories and ideas but no one wants to share their data.

 

T.

[Beck]

Interestingly no one has commented on the organ. I did mention how it was miked and what you should expect to hear. When the organ reaches it's first sustained note you are hearing a stereo recording of a single note produced through a rotating horn. It should therefore circle your head. The original master it did. The vinyl version did. A cassette dub did. The CD version doesn't. It was transferred using the latest digital gear at one of Sydney's top mastering studios.

Nice work John,

Great experiment and outstanding work on the graphics so everyone can visualize what's going on.

The results are no surprise to me.

[Beck]

The thread has mutated a bit since I last posted, but I wanted to respond to a couple things from a few days ago nonetheless. I

[Beck]

LOL, you forgot frequent breakdowns, expensive replacement parts (if you can even find them), and all the time and skill it takes to align the heads and electronics before every session. Digital users don't have to waste their time trying to overcome the medium's limitations because there aren't any!

 

 

Spoken like a true believer. I mean it really is double dog reverse LOL because you once again failed to notice the big white digital elephant in the middle of the room: :poke:

 

- Hard drive failures

- Power supply failures

- RAM failures

- Motherboard failures

- Monitor failures

- Graphics, audio card/system failures

- OS crashes/reinstalls

- I scream, you scream, we all scream for Blue Screen

- Patches, bugs and fixes

- Compulsory software upgrades due to new OS incompatibility

- Patch to fix buggy patch that supposedly fixed previous bug

- Compulsory hardware upgrades to accommodate buggy new resource hogging OS

- Dump old favorite software because it just won

[Han]

That was quite some post Tim! All I can add is that I bought a new Tascam MSR24S in 1992 that worked day in day out, sometimes seven days a week and it still works every now and then when I run out of tracks on the Otari 2". In 1996 the head was amost completely gone and the funny thing was that one of the tracks in the middle, like 11 or 12 were having dropouts.
So it got a new head and that's all, it never have let me down, not for a split second, even when I was recording on location, it always worked like a charm and still does.

Some 11 years ago I bought this Otari MX80 from a private studio, very little used and technically almos a new machine. This machine had only one little problem once when a cpacitor was blown on one of the audio PCB's. The machine is known for not having that typical 'analog' sound, what you put into it will come out for 98%, well at least 95%.

The machine is very reliable and very steady, aligning is a piece of cake and this is a real workhorse of a machine. I have never had to cancel any session in my 20 years of running a studio. The very old Fostex B16 which I've used for the last time in mid 1992 is still in good working order.

That's my analog gear, my digital gear: I have five DAT recorders, only one is still working, four are broken. I have two standalone (very expensive) CD recorders, both are broken, the one that has cost most gave up after some 100 CDR's. My cellar is full of computers that are obsolete or broken, I have spent much more money on digital gear than on analog gear so far. All analog gear that I've bought during the years is still working, even my more than 50 years old Telefunken M10 and my 50+ years old full tube Philips PRO 52 master machine, my old Studer machine got a few new caps recently and sounds better (IMHO) than any of the DAT, CDR or PC based gear.

My Hammond/tube Lesley has never failed once while it's almost 50 years old. My Rhodes works like a dream, all tube guitar amps didn't have problems.

All the microphones still work while some are 50+ years old. The problems I've had were all digital related and somtimes it drives me really mad.

Besides all this, digital gear sounds good, really good, no doubt about that, but never brilliant like hi end analog gear can. Modern (young) people don't know what they're missing.

[Jimbroni]

I don't even know how to respond to this without being potentially offensive really. A) It's very unlikely you can actually hear a .1dB difference, and are probably just convincing yourself that you are. B) It has nothing to do with digital having limited dynamic resolution either way. It has resolution far beyond the human ear to hear the difference between adjacent levels, clearly. C) If you can in fact hear such a thing on a digital system, it would be because of greater accuracy in the digital system, not the other way around. A .1dB difference is the same difference in either system. Given that it would represent probably a change of tens of thousands of sample levels, obviously it's not because you suddenly moved from one big, chunky sample level another. So I just have a problem with your whole premise really.

 

 

Well working with automotive subjectives, I know that the average person can tell a 3 dB difference and some can tell a 1 dB difference. I don't believe I can tell .1 dB difference, that'd be impossible really, but I can infact tell these differences on digital mixer. Which makes me think something is fishy. I know its not just me, because I often do mix tweaks with my other band members, to get another set of ears on it, they can hear these tiny little tweaks as well. Knobs maybe onto something with the fader resolution thing.

 

As far as your last point, its kinda off the mark. If the resolution allows for such small movements to have an audible effect thats not accuracy, thats missing something. Now my point isn't limited to faders, but eq's, compressor make up gains, spectralizers, you name it. These things are too grainy. Is that caused by bit depth or DAW and plugin design or just the nature of digital I don't know. but I suspect alot of these differences between analog and digital are rooted in this in someway.