-

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

This site may earn a commission from merchant affiliate links, including eBay, Amazon, and others.
john12ax7 said:
From a technical perspective there are good reasons to go above 44.1k. It's not necessarily about capturing above 20k, but rather increasing the performance below 20k. The 10% margin,  while once maybe being necessary,  is in general poor engineering practice.  2-5x bandwidth would be more reasonable. 
That would be true in the context of traditional analog design, not so much in view of the common usage of digital filters.
I agree, though, that the transient response of sharp cut-off filters is a serious issue.
I would think that the choice of 44.1 k  results from a mythical 20kHz auditory limit; if a more reasonable choice of 16kHz BW whilst retaining 44k S, we may not have the same discussion. After all, I don't think any body seriously criticized 15k BW in broadcast.
 
JohnRoberts said:
If you think the transient response of A/D conversion is difficult, try compressors (but I'm sure you have done that too). Even power amplifier design requires rate of change consideration.
Diferent subjects IMO. In dynamics, the side-chain is the whole issue; the signal path is almost never a subject of discussion (possible exceptions for PWM gain-cell). For amplifiers, since they are dominantly Linear Phase, the main issue is slew-rate and associated IMD/TIM.

The million dollar question surrounding any discussion of transient response is "in what bandwidth". Back when I was active in audio design (I'm not anymore) I was a strong advocate for LPF any audio path as early as possible to prevent shenanigans from out of band signals that even analog electronics sometimes stumbles over.
Agreed. I was never a fan of DC-to-light frequency response; several customers of mine were quite shocked when they saw the actual frequency response of some of my gear, objective data being in contradiction with subjective evaluation. However, response beyond the -3dB points do count.
A typical example is the electric guitar; although the actual (-3dB) BW is quite limited (typically 3kHz for humbuckers, 4kHz for single-coil) a lot of the sonic signature resides in the harmonic content several octaves above. The ear is capable of discriminating formants that are way below the fundamental. Inserting a brickwall filter at 8-10kHz changes the perceived sound of the electric guitar.

Dangling a key chain in front of a mic would often reveal weak sisters who couldn't handle the 50kHz + stress.
Definitely. A very strong motive for LP'ing the insput stage.

PS: Perhaps more appropriate how many here can actually hear a (decent ) LPF at 30-50kHz....  I recall some long and passionate discussions on Geeksluts about this bandwidth issue years ago with little resolution.
All experiments I'm aware of or having participated in show that a low-order ultrasonic LPF having a good transient response (low-Q) cannot be detected, except when non-linearities occur in the signal path. In most cases, overshoot in the step response is a cause of saturation. This has been particularly researched in the study of phase audibility, where it showed that the main audible factor was distortion introduced by the increase of peak-level (which is not limited to the electronic path, but also to the transducers, the air and the ear. Interference between two paths having different phase-relationship is another issue.
 
80hinhiding said:
I heard a difference going from 44.1 or 48 to 96,
So did I

and wondered what I might hear with 192.
I understand that, and as I wrote earlier I didn't hear anything significant. But you are entitled (and should) experience it yourself.

  I don't think it's about trying to hear up to 192Khz frequency.
Indeed; nobody can make such a preposterous claim (although it has been demonstrated that some ultrasonics may be picked up by the retina cornea..?.

it's just seeing how the overall bandwidth is affected on a different setting.
I would think the frequency response as perceived by the ear doesn't make a difference when it comes to the 20kHz-80kHz range; the differences I perceive between SS and DS are in the way transients are handled. There is a very strong ultrasonic content in cymbals and various percussion instruments; how the filters effectively remove them and withstand them is paramount.

Maybe I'd like it less, maybe more.
I recently took a survey, comparing identical mixes done at SS and DS; the question was "which one has the more punch", not "what is the most transparent/true/pleasant", and I consistently chose the SS ones. Go figure...
 
john12ax7 said:
In your test isn't the 44.1k then getting upconverted at the output since it's only one digi 192?

YES, but then I said:

There were other versions of the test done, two different rigs with two sample rate sessions, different transfer A/D's etc.

If someone consistently gets 12 or 13 out of 20 that is actually not random but statistically significant.

Maybe for you, but for me I would want a consistent 19 or 20 right guesses from all the participants to be sure they were hearing a difference.

john12ax7 said:
Maybe I'll give 192k another spin too.

Well I'm glad that after I took the time to explain in detail, a very convincing rigorous test I did on professional audio engineers in a professional listening environment, proving that in a controlled test no one could tell the difference between sample rates...It became even more possible in your mind that you COULD tell the difference.

John the tests you sited were very vague and not too convincing. And only done by you on yourself.

I would still like to hear about someone else's test (in detail) that prove people can actually hear the difference. Test's that have a pool of participants. I know I don't have hard data to show, so I don't expect people to 100% believe what I'm saying, this is a forum on the internet after all...But man this is almost as bad a the brewery, people just believe what they want to believe...







 
bluebird said:
I would still like to hear about someone else's test (in detail) that prove people can actually hear the difference. Test's that have a pool of participants. I know I don't have hard data to show, so I don't expect people to 100% believe what I'm saying, this is a forum on the internet after all...But man this is almost as bad a the brewery, people just believe what they want to believe...

I believe and have been saying pretty much that...

I also make a point of not arguing with people on the WWW about what they say they can hear... It is impossible to prove and generally irritates them.

JR

PS: Many political arguments are equally impossible to prove.
 
bluebird said:
Maybe for you, but for me I would want a consistent 19 or 20 right guesses from all the participants to be sure they were hearing a difference.
I believe you misunderstood this point; my understanding of it is that 12-13 out of 20 is pretty close to half-right and half-wrong, not too dissimilar to the results one would expect if the panel answered at random, but still, the number of samples is high enough to make it statistically acceptable (within reason).
 
You talking to me? That was my point, 12-13 out of 20 is random in my mind and not statistically significant as John12ax7 stated in this case. In fact some guys guessed 8 or even 5 right out of the 20 guesses. All I'm trying to say, is people could not tell the difference.  I as well as the other studio techs involved, wanted to be as scientific as possible and get a solid answer to report to Universal Music.  I was not biased in any way because I wasn't trying to sell anything or convince anyone I had super human hearing. I just wanted a solid answer. They wanted to know what the best sampling rate was to achieve tape to and we were tasked at giving a good answer to that question. What drives me crazy is (the completely nontechnical) executives decided to go with 192K anyhow, "just in case". We all threw up our hands and wondered why we went through all the trouble!

Interesting thing is, when I cut from hires files sent from Warner Brothers they are delivered in 96K 24bit format.  So somehow they're archival department decided on 96K.  Which I think is a better (level headed)  format for tape archiving than 192K.

Note: I'm not advocating for tape archival to be done at 16bit 44.1K. I was just explaining one of the many tests we did involving tape machine models, A/D converters, and sample rate during that time.  I only touched on that particular test, because of the claims that the difference in sample rates could be heard. I was hoping for a little more evidence to prove it. Not so I can be right, but so I can be educated. I'm a professional mastering engineer and knowing people can actually hear the difference would help to advance my techniques and ultimately provide a better product. But I won't change the way I work on hearsay alone.
 
bluebird said:
YES, but then I said:

There were other versions of the test done, two different rigs with two sample rate sessions, different transfer A/D's etc.

Oops sorry,  I totally missed that part.

bluebird said:
Maybe for you, but for me I would want a consistent 19 or 20 right guesses from all the participants to be sure they were hearing a difference.

But that's not how statistics works.  You do not need 19 or 20 from everyone to prove there is an audible difference.

bluebird said:
Well I'm glad that after I took the time to explain in detail, a very convincing rigorous test I did on professional audio engineers in a professional listening environment, proving that in a controlled test no one could tell the difference between sample rates...It became even more possible in your mind that you COULD tell the difference.

John the tests you sited were very vague and not too convincing. And only done by you on yourself.

I would still like to hear about someone else's test (in detail) that prove people can actually hear the difference. Test's that have a pool of participants. I know I don't have hard data to show, so I don't expect people to 100% believe what I'm saying, this is a forum on the internet after all...But man this is almost as bad a the brewery, people just believe what they want to believe...

I appreciate the detailed response and I'm not trying to dismiss it.  But you would really need to do a proper statistical analysis to prove things one way or another. What was the confidence level? Were the results consistent as time went on? Did trials1-5 have the same distribution as 15-20? Why not extensive hard data after such a painstakingly thorough test?

For me I've found it very easy to get fatigued in this type of test.  It is also important,  for me,  to have proper source material that you are intimately familiar with.

I try to keep an open mind in the realms of audibility. Wanting to reexamine 192k is out of my own curiosity, previously I didn't hear a benefit of QS vs DS, is this still true? It is not a dispute of your tests.
 
bluebird said:
You talking to me?
Yes.  :)

That was my point, 12-13 out of 20 is random in my mind and not statistically significant as John12ax7 stated in this case.
After re-reading his post I'm confused. Indeed 12-13 /20is random, but gettinng consistently random answers is statistically viable.

In fact some guys guessed 8 or even 5 right out of the 20 guesses. All I'm trying to say, is people could not tell the difference.
No doubt about it.

What drives me crazy is (the completely nontechnical) executives decided to go with 192K anyhow, "just in case". We all threw up our hands and wondered why we went through all the trouble!
Much worse decisions have been made according to the "umbrella" principle; think fallout shelters  ::)

Interesting thing is, when I cut from hires files sent from Warner Brothers they are delivered in 96K 24bit format.  So somehow they're archival department decided on 96K.  Which I think is a better (level headed)  format for tape archiving than 192K.
Agreed; in the current state of scientific knowledge 96/24 has a significant advantage over tape in all known aspects of performance.
 
Thanks John, and I want to make it  known that I respect your points of view on this forum and know your a smart dude. That's kind of why I got a little irritated at your initial  response, because I felt you did not give me an equally scientific and complete argument to back up your views.

john12ax7 said:
I appreciate the detailed response and I'm not trying to dismiss it.  But you would really need to do a proper statistical analysis to prove things one way or another. What was the confidence level? Were the results consistent as time went on? Did trials1-5 have the same distribution as 15-20? Why not extensive hard data after such a painstakingly thorough test?

You are absolutely right.  There was no hard data in the end and I didn't write a book about it or anything. I definitely don't want anyone to take my experience as gospel, but I do think it was an intensive enough test for me to share with other people in hopes that they will become more educated and less gearslutty about the subject.  8)

Like I said before, I would like to hear about a test close or even more intensive to what I did to educate me in the other direction. If there is another direction... I definitely don't want to start an argument just for the sake of who's right or wrong!

Much respect!



 
abbey road d enfer said:
Yes.  :)
After re-reading his post I'm confused. Indeed 12-13 /20is random, but gettinng consistently random answers is statistically viable.

Thanks for clarifying Abby! Believe or not, I always mope around for the rest of the day when you or JR (or PRR for that sake) put an educated lid on something I say...which has happened more than not! (since 2004)

 
bluebird said:
Thanks for clarifying Abby! Believe or not, I always mope around for the rest of the day when you or JR (or PRR for that sake) put an educated lid on something I say...which has happened more than not! (since 2004)
Thanks for the nice compliment, but please don't lay it too thick; I wouldn't want to be mistaken for a guru.  ;)
 
Maybe I should clarify my statistics position. Suppose you do many (this is important) trials of 20 with three different groups:

Group A consistently gets 12 of 20
Group B consistently gets 10 of 20
Group C consistently gets 8 of 20

So the total is 50% correct, but is this random? Actually no.  You have actually proved that both Group A and Group C can hear a difference, even if it's a subtle not easily obvious one.

It seems your test was an A/B, "pick the 192k" correct? So even a consistent 8 of 20 shows an audible difference too,  it's just that the preconceived notion of 44.1k vs 192k was wrong.  I'm not necessarily concluding this is the case,  as I don't know all the details,  just a point to consider.

I really feel you need a control.  A/B/C, live tape vs 44.1k loop through vs 192k loop through,  "pick which is closest to the source". This eliminates preconceptions of what should be better.

Side question: how transparent is SRC (I use samplitude / sequoia) ? Would a 96k file converted to 44.1k then back to 96k, then compared to the original, is this a valid comparison?

Things online don't always come across the right way. I appreciate that things on here tend towards a civil discourse  :)
 
john12ax7 said:
Maybe I should clarify my statistics position. Suppose you do many (this is important) trials of 20 with three different groups:

Group A consistently gets 12 of 20
Group B consistently gets 10 of 20
Group C consistently gets 8 of 20

So the total is 50% correct, but is this random? Actually no.  You have actually proved that both Group A and Group C can hear a difference, even if it's a subtle not easily obvious one.

It seems your test was an A/B, "pick the 192k" correct? So even a consistent 8 of 20 shows an audible difference too,  it's just that the preconceived notion of 44.1k vs 192k was wrong.  I'm not necessarily concluding this is the case,  as I don't know all the details,  just a point to consider.

I really feel you need a control.  A/B/C, live tape vs 44.1k loop through vs 192k loop through,  "pick which is closest to the source". This eliminates preconceptions of what should be better.

I hear ya.

john12ax7 said:
Side question: how transparent is SRC (I use samplitude / sequoia) ? Would a 96k file converted to 44.1k then back to 96k, then compared to the original, is this a valid comparison?

I think the ole' phase cancellation test might be a good start to answer that one. If after processing  the file, they (SRC processed file and original 96K file) cancel perfectly, I don't know maybe dig deeper in, but if you hear some high frequency hash when phase flipped and played together then the SRC has audibly changed the sound.

john12ax7 said:
I appreciate that things on here tend towards a civil discourse  :)

Yeah as long as tands doesn't get involved :p
 
john12ax7 said:
I use samplitude / sequoia
So do I. I check regularly this site
http://src.infinitewave.ca/
for updates, and I see that Samp/Seq is never in the winning position. I have tried once to convert a file with Audacity (supposed to be one of the best); I couldn't justify the extra work.
 
80hinhiding said:
I don't think anyone is being gearslutty here Bluebird, are they??

Nope.

80hinhiding said:
I would like to point out not all converters and accompanying circuitry are equal.. so one may produce differences that are audible/able to be felt while another may not.  Some may run better at lower rate than another at a higher rate, etc etc etc.

Welp, like I said we DID test different converters (Digidesign, Apogee, Lynx, Prism) whatever the current multi channel models were at that time, 6 or 7 years ago I think? And again no one could reliably pick one converters transfer over another when level matched and guesses tallied. The pick in the end was the blue/grey Digidesign 192's. The thinking was, if no one could tell the difference, we might as well use what the studio had the most of...192's.


80hinhiding said:
I would not be the least bit surprised if the audible frequency range was affected by running a converter at a rate of 192.  I would be surprised if it doesn't..

Well if someone told me what I told you, I would think the exact opposite. That was the whole point of the story! I was trying to convince you that 192K is a waste of your time and energy, not to mention hard drive space.

Ultimately I was giving this thread my time for YOU. And at the end of it all, you want to believe exactly the opposite of what I'm trying to prove to you, Lol.

Again not trying to be an official know it all here, Just sharing an experience that I think should have some value to you considering the time and effort your going to spend on your mixer. I'm telling you don't do it. Spend your time and money on things that will actually have an impact on your recordings.

80hinhiding said:
Do you have sample files on hand we could compare?

No I don't sorry.

I thought Dan Lavry's white papers on sampling rate were  pretty common knowledge around here. This guy is WAY more intelligent and qualified to explain the ridiculousness of 192K sampling rate.
I'll just let him take it from here...
http://lavryengineering.com/pdfs/lavry-sampling-theory.pdf


 
80hinhiding said:
I said I wanted to find a way of trying 192 rate (without much hassle), and asked if anyone knew how to do it.  This led to pages upon pages of debate on whether or not it's worth it, or if it's even audible....
You wrote: "not all converters and accompanying circuitry are equal.. so one may produce differences that are audible/able to be felt while another may not.  Some may run better at lower rate than another at a higher rate, etc etc etc."
The logical conclusion is that, if you tried a converter at 192, you would not evaluate 192k SR per se, but you would be first of all evaluating a particular converter.
Indeed, i would think there is a difference between a good Single Speed converter think Benchmark) and a weird 192kconverter (e.g. Icon Cube).

Then "I would not be the least bit surprised if the audible frequency range was affected by running a converter at a rate of 192.  I would be surprised if it doesn't.. have some impact on the converter performance for the audible range (varies for everyone)."
Well, I've measured a number of 192k converters (trying - and failing - to find one suitable for measurements up to 80-90kHz), but, apart from the Icon Cube, that has frequency response anomalies at all SR, the others have no differences whatsoever in the 20-20kHz, whatever the SR.
 
abbey road d enfer said:
So do I. I check regularly this site
http://src.infinitewave.ca/
for updates, and I see that Samp/Seq is never in the winning position. I have tried once to convert a file with Audacity (supposed to be one of the best); I couldn't justify the extra work.

I've looked at that too and been surprised.  I've always figured (perhaps incorrectly) that since sequoia is somewhat of a staple in mastering that it's SRC would be top notch. 
 
bluebird said:
Welp, like I said we DID test different converters (Digidesign, Apogee, Lynx, Prism) whatever the current multi channel models were at that time, 6 or 7 years ago I think? And again no one could reliably pick one converters transfer over another when level matched and guesses tallied. The pick in the end was the blue/grey Digidesign 192's. The thinking was, if no one could tell the difference, we might as well use what the studio had the most of...192's

The different makes of converters was not mentioned before,  at least not explicitly.  I find it rather surprising that no differences were heard between converters.  In my (albeit subjective)  experience something like lynx aurora and apogee ad/da16x have a very different sound.
 
Back
Top