The No.1 Website for Pro Audio
 Search This Thread  Search This Forum  Search Reviews  Search Gear Database  Search Gear for sale  Search Gearslutz Go Advanced
Launch of Pono Studio Headphones
Old 2nd April 2014
  #2131
Lives for gear
 
bogosort's Avatar
Quote:
Originally Posted by paul brown View Post
are you saying then that more samples better capture transients? an earlier 'expert' tells me the opposite (post number 2107).
Again, a transient with infinite slope (0 rise time) has infinite frequencies (infinite bandwidth). One would need infinite sample rate to capture this. As you decrease the slope the bandwidth is correspondingly decreased, and the required sampling rate decreases.

It may help to visualize the transient -- let's picture an impulse -- as its bandwidth is reduced. With infinite bandwidth the impulse looks like a perfectly straight, perfectly skinny vertical line. As we reduce the bandwidth, the slope of the rise time decreases (starts leaning toward the right) and the infinitely skinny line begins to get fatter. Keep reducing bandwidth and soon it looks like a mound or hill: smooth rise up, smooth fall down. Eventually it will become flat -- DC -- because the bandwidth is zero.

Our ears naturally reduce the bandwidth of transients. Smack two claves together and the universe produces a very pointy impulse (let's say it has 100 kHz bandwidth). Our ears, however, naturally smooth this transient into something more rounded, such that it has 20 kHz (or less) bandwidth.

So yes, we can increase sample rate and capture more bandwidth from the transients, but what for? The ear will just smooth them out anyway. Any transient you can hear is perfectly characterized by a 44.1k sample stream. This is the entire point of the sampling theorem.

And BTW, j_j is very much indeed an expert.
Old 2nd April 2014
  #2132
Lives for gear
 
paul brown's Avatar
thanks, bogosort!
Old 2nd April 2014
  #2133
A problem is, I think, that many folks seem to think of transient information in waveforms as somehow 'different' from other sound, but it's not. It just, in a sense, comprises higher frequency information. If our ears can't hear that high frequency information, they can't hear it. Same as any other sort of sound. Perceptual testing has borne this out.

The instruments we tend to gravitate toward tend to have resonances that produce somewhat steady tones, but the same principles that go into accurate capture of musical tones also go to the accurate capture of transients. Transient sounds can be analyzed and recreated up to the desired bandlimit using the Fourier Transform and Sampling Theorem because they are sound waves, too.
Old 2nd April 2014
  #2134
Quote:
Originally Posted by paul brown View Post
i still think the discussion in part, including the abx testing, is more about the performance of respective converters at different sample rates. We really should be labeling converters by their resolution, not by their sample rate to quote Bob Katz again. unless that variable is removed, where is the level playing field to hear a difference? i also trust a few ears in the industry based on my listening and what they say. their outlook and philosophy about music. it could all be bias. me, them. got to trust someone for the real world application of science though! (that is a bias as well. i know!)
I trust experienced producers for their expertise in running sessions (those I trust, that is) and experienced recording 'engineers' for their expertise in getting good sounding records made.

Neither leads me to infer that either studio role necessarily means any particular scientific understanding of the technologies involved.

In fact, in the 30+ years since I first stepped into a commercial studio to engineer and produce, I've known a number of guys and gals with great practical knowledge, studio expertise, and customer aplomb -- but those qualities -- crucial as they are to smooth running sessions -- do not necessarily confer technical, scientific knowledge.

These are two very different sets of expertise -- but there IS overlap and the qualities of discipline and logic will help a recording engineer (or producer or musician for that matter). Some try to get by on intuition and some probably do pretty well at that.

But if one doesn't have a basic understanding of how the Nyquist-Shannon Sampling Theorem works its non-magic -- I don't necessarily mean being able to work all the math, but I do mean coming to grips, intuitively, with how two+ sampling points on a sine can accurately define that sine and how complex wave forms can be broken down analytically into individual sine wave components -- then one is going to be banging his head in the dark against walls he cannot see.
Old 2nd April 2014
  #2135
Lives for gear
 
paul brown's Avatar
Quote:
Originally Posted by theblue1 View Post
A problem is, I think, that many folks seem to think of transient information in waveforms as somehow 'different' from other sound, but it's not. It just, in a sense, comprises higher frequency information. If our ears can't hear that high frequency information, they can't hear it. Same as any other sort of sound. Perceptual testing has borne this out.

The instruments we tend to gravitate toward tend to have resonances that produce somewhat steady tones, but the same principles that go into accurate capture of musical tones also go to the accurate capture of transients. Transient sounds can be analyzed and recreated up to the desired bandlimit using the Fourier Transform and Sampling Theorem because they are sound waves, too.
indeed. it clicked...finally! i'm back to the quality of a converter at a given rate being what some people are hearing then. a great 44.1 converter sounding better than a mediocre 96 and vice-versa, etc. fascinating rabbit hole to go down! i find it more interesting now how our brains process the data, especially in how we sense the direction and position of sound sources. i was just reading about human echolocation...wow. fascinating ability.

BTW. did anyone see the video footage of the thirty-nine year old english woman who heard for the first time with new generation cochlea implants? it made me cry with joy for her. she also lost her sight in her twenties. apparently, the first song she listened to was John Lennon's Imagine.

back on topic. pono...
Old 2nd April 2014
  #2136
Lives for gear
 
bogosort's Avatar
Quote:
Originally Posted by paul brown View Post
BTW. did anyone see the video footage of the thirty-nine year old english woman who heard for the first time with new generation cochlea implants? it made me cry with joy for her. she also lost her sight in her twenties. apparently, the first song she listened to was John Lennon's Imagine.
I weep like a baby every time I see that.
Old 2nd April 2014
  #2137
Lives for gear
 
nuthinupmysleeve's Avatar
 

Quote:
Originally Posted by johnnyc View Post
Higher sample rates DO mean better transient response.
Yes for my dog.
Old 2nd April 2014
  #2138
Gear Guru
 

Quote:
Originally Posted by theblue1 View Post
These are two very different sets of expertise -- but there IS overlap and the qualities of discipline and logic will help a recording engineer (or producer or musician for that matter). Some try to get by on intuition and some probably do pretty well at that.
Yes, although motor oil manufacturers routinely present us with the face of a race car driver in the commercials, it is probably rare that the driver is as knowledgeable about lubricants as the actual Mechanic on his race car team. Scientifically speaking, that mechanic should be doing the commercial. heh

It is probably possible to win the Indy 500 on driving skill alone, if you have someone else building the car for you.

There was a post here on GS a few years ago, quoting a very successful producer on the sonic differences between different hard drives. Not a wind-up as far as I could tell, the person was apparently 100% serious.

Should we all start searching out the "best-sounding" hard drives on this person's say-so? After all, he IS very successful at making records....

When highly successful golden ears are willing to sit for double-blinded perceptual studies, they might one day provide data that forces science to re-think measurements obtained to date. Perhaps they are Outliers - sensing more than other humans. But until the time that somebody (even one somebody) can "do it with a blindfold on", it is merely Argument From Authority.
Old 2nd April 2014
  #2139
Right. Horses for courses. I might ask the magic hard drive guy a question about dealing with musicians in the studio (assuming he's got an impressive c.v.) or such, but I'm not going to ask him a serious technical question. Clearly, from his prior utterances, that is not his expertise.
Old 2nd April 2014
  #2140
Someone mentioned this earlier in the thread but it should be mentioned again, it's great and free!

https://itunes.apple.com/us/app/abxt...27554135?mt=12

I don't understand why some people have an apprehension to double blind testing. Well I assume it's because they don't want to really know if they are hearing a difference or not but that seems asinine.

For example someone claiming they can hear the difference between 24/96 and 24/192, prove it. Not only to everyone who you have made that claim to but probably more importantly, to yourself.
Old 2nd April 2014
  #2141
Gear Guru
 
Kenny Gioia's Avatar
 

Quote:
Originally Posted by jasonwagner View Post

For example someone claiming they can hear the difference between 24/96 and 24/192, prove it. Not only to everyone who you have made that claim to but probably more importantly, to yourself.
Exactly. And as a producer or mixer, you should test it with things that matter to your work.

My only hesitation to ABX testing has been to use other's material. Without having an intimate relationship with the material, it's hard to care as much about the differences.
Old 2nd April 2014
  #2142
Lives for gear
 
matyas's Avatar
 

Quote:
Originally Posted by theblue1 View Post

But if one doesn't have a basic understanding of how the Nyquist-Shannon Sampling Theorem works its non-magic -- I don't necessarily mean being able to work all the math, but I do mean coming to grips, intuitively, with how two+ sampling points on a sine can accurately define that sine and how complex wave forms can be broken down analytically into individual sine wave components -- then one is going to be banging his head in the dark against walls he cannot see.
The issue isn't the Nyquist-Shannon Theorem. We all know how that works. (Or should.) The issues are two:

1. Real-world converters do not always behave like mathematical models.

2. Hearing and comprehension of music is an intensely subjective process. We know much more about the physics of musical reproduction that we do about the psychoacoustics of musical cognition. Some people think that 22.05 kHz is insufficient with regards to bandwidth for the reproduction of at least some musical signals under some conditions.

The conflation of the mathematics of the Nyquist-Shannon Theorem (which, I readily agree, is pretty much irrefutable) with the psychoacoustics of musical perception (which is much more difficult to quantify) is fast becoming a pet peeve of mine.
Old 2nd April 2014
  #2143
Lives for gear
 
matyas's Avatar
 

Quote:
Originally Posted by Kenny Gioia View Post
My only hesitation to ABX testing has been to use other's material. Without having an intimate relationship with the material, it's hard to care as much about the differences.
This is the big elephant in the room with regard to ABX testing. The variables which would differentiate two otherwise identical recordings at different sampling rates are rarely the first things anyone - even trained audio engineers - notice. We are accustomed to listen first for content (melody, harmony, chord progressions, rhythm, etc.), then for things like tone color, then for details of interpretation, and only last for high-frequency extension, spatial cues, lack of transient smear, or any of the other qualities ascribed to high-sampling rate recordings. Such qualities are usually only apparent with highly familiar material. By that point, one may readily concede that confirmation bias becomes an issue. But that does not necessarily disprove the existence of such qualities - merely that they are elusive and difficult to quantify. Then again, most of what we love about music is elusive and difficult to quantify.
Old 2nd April 2014
  #2144
Lives for gear
 
bogosort's Avatar
Quote:
Originally Posted by jasonwagner View Post
I don't understand why some people have an apprehension to double blind testing. Well I assume it's because they don't want to really know if they are hearing a difference or not but that seems asinine.
One narrative goes like this:

It seems obvious that the more samples you take, the more accurate the resulting picture will be. Intuition firmly takes root. Listening to 192k playback you hear a night and day difference, like someone turned on the lights. It's completely obvious how much more realistic everything sounds compared to 44.1k.

Then you read a post on GS by some curmudgeonly so-called expert that claims there is no audible difference. They post some files at 192k and 44.1k with the challenge to guess which is which by listening. You laugh and think to yourself, this will be easy. You playback each of the files, switching back and forth, and to your dismay you can't hear a difference.

So do you conclude that there is no audible difference? No, you make an angry post calling out ABX testing as pseudoscience. You indignantly point out that an ABX test cannot measure the subtle, almost subconscious differences that us real engineers deal with on a daily basis.

But what happened here? What was at first a night-and-day difference between 192k and 44.1k is now a subtle difference that manifests itself only after significant time listening, and only on certain sources. You of course continue to use 192k because deep down you know it's better, "evidence" and "theory" be damned. And you never subject yourself to another ABX test because that's just pseudoscience. Obviously.
Old 2nd April 2014
  #2145
Lives for gear
 
paul brown's Avatar
Quote:
Originally Posted by jasonwagner View Post

For example someone claiming they can hear the difference between 24/96 and 24/192, prove it. Not only to everyone who you have made that claim to but probably more importantly, to yourself.
what is the best way to create the files for comparison? start with a file recorded at 192/24 and copy it at a lower sample rate? my DAC automatically up-samples everything to 192. the designer says it is to do with less phase rotation. what are the implications, if any, in doing a sample rate abx test knowing this?
Old 2nd April 2014
  #2146
Lives for gear
 
nuthinupmysleeve's Avatar
 

Quote:
Originally Posted by paul brown View Post
what is the best way to create the files for comparison? start with a file recorded at 192/24 and copy it at a lower sample rate? my DAC automatically up-samples everything to 192. the designer says it is to do with less phase rotation. what are the implications, if any, in doing a sample rate abx test knowing this?
This is NOT a perfect test, but will tell you if the frequencies above 20k will have a meaningful difference.

Take your 192/24 file and make a copy. Then in a standalone editor do a very steep filter above 20khz. Save that.

Now do an a/b/x test between the new file and the original one. Take as long as you want. There are free testing tools on both mac and windows platforms.

Be sure to ensure you are listening at 192khz.. be sure that is what your audio interface is set for.

Report back! Let us know your success rate.
Old 2nd April 2014
  #2147
Lives for gear
 
bogosort's Avatar
Quote:
Originally Posted by matyas View Post
The issue isn't the Nyquist-Shannon Theorem. We all know how that works. (Or should.)
Except that most don't. And many, it would seem, simply don't believe the theorem. In other words, on a gut level they are highly suspicious that 2.1 samples per signal is enough to completely characterize it. Further, many have no idea how quantization really works. They somehow imagine that 24-bit word lengths provides more "air" and other euphonic qualities than 16-bit word lengths, even though the difference is indisputably a matter of noise floor.

Quote:
The issues are two:

1. Real-world converters do not always behave like mathematical models.
Indeed, but every single factor that deviates from the pure mathematical model also has a mathematical model. In other words, the real-world behavior of converter systems is perfectly quantifiable.

Quote:
2. Hearing and comprehension of music is an intensely subjective process. We know much more about the physics of musical reproduction that we do about the psychoacoustics of musical cognition. Some people think that 22.05 kHz is insufficient with regards to bandwidth for the reproduction of at least some musical signals under some conditions.
You're correct that we know more about the physics than the psychophysics, but we've learned a lot about the latter over the past few decades. And in that time many studies have tried to demonstrate extra-22.05k perception and not one has been deemed definitive. I'm not calling for the abandonment of further research in these areas -- quite the contrary -- but it is certainly clear by now that if there are any ultrasonic mechanisms involved in music perception, they are extremely subtle at best. I don't see how anyone can reasonably deny this.

Quote:
The conflation of the mathematics of the Nyquist-Shannon Theorem (which, I readily agree, is pretty much irrefutable) with the psychoacoustics of musical perception (which is much more difficult to quantify) is fast becoming a pet peeve of mine.
Not "pretty much irrefutable"; just irrefutable. Full stop. (Unless you want to revise 3000 years of mathematics.) I get why you're peeved by those who would (in your eyes) trump the math over reality, but mathematics is our best guide. We have no reason to believe that the mathematics of information theory does not apply to the ear-brain system; on the contrary, with the profound success of information theory in so many other facets of the real world, it would be frankly shocking if human hearing were outside its domain.

Nonetheless, I fully grant that there is no last word on this deeply interesting subject of psychophysics. The more intelligent discussion and debate on this, the better!
Old 2nd April 2014
  #2148
Quote:
Originally Posted by paul brown View Post
what is the best way to create the files for comparison? start with a file recorded at 192/24 and copy it at a lower sample rate? my DAC automatically up-samples everything to 192. the designer says it is to do with less phase rotation. what are the implications, if any, in doing a sample rate abx test knowing this?
I am the wrong person to answer this, I am more of a tool user than tool creator or even understanding what the tool is doing beyond what I hear.

Setting up the files for blind testing is the hard part since there are so many variables to take into account. If I were going to test 24/192 vs 24/96 I would import the 24/192 file into Logic and bounce it down to 24/96 then take the original and bounced version into the ABX tester and have a go. But I would check here or some other board to make sure that is the best way to prepare the files. But like someone mentioned early, even if I screwed up the conversion to 24/96 then in that particular test it's only going to be a benefit to the 24/192 sounding better.

With this tester having only 5 comparisons per test I would want to be able to consistently get at least 4/5 before I consider something better. If someone is claiming night and day difference they should always get 5/5. It's pretty crazy to do a blind test, I didn't really have the ability before finding this app and it's eye/ear opening for sure. It's actually down right amazing to think you are hearing a difference and then when doing a blind test to hear those differences completely disappear.
Old 2nd April 2014
  #2149
Lives for gear
 
paul brown's Avatar
Quote:
Originally Posted by nuthinupmysleeve View Post
This is NOT a perfect test, but will tell you if the frequencies above 20k will have a meaningful difference.

Take your 192/24 file and make a copy. Then in a standalone editor do a very steep filter above 20khz. Save that.

Now do an a/b/x test between the new file and the original one. Take as long as you want. There are free testing tools on both mac and windows platforms.

Be sure to ensure you are listening at 192khz.. be sure that is what your audio interface is set for.

Report back! Let us know your success rate.

thanks! i'll report back with a screenshot of the results when i am done.
Old 2nd April 2014
  #2150
Lives for gear
 
bogosort's Avatar
Quote:
Originally Posted by nuthinupmysleeve View Post
This is NOT a perfect test, but will tell you if the frequencies above 20k will have a meaningful difference.

Take your 192/24 file and make a copy. Then in a standalone editor do a very steep filter above 20khz. Save that.
This is the worst possible way to test! One of the most important aspects in converter design is the filtering; throwing a random editor's 48 dB/octave LPF at 20k will make mince meat of the upper band. If you don't hear a difference, then it might be time to go to the audiologist.

The proper way to do this is to use a high quality sample-rate conversion (SRC) program. The DSP coder will have carefully created a filter appropriate to the task. The freeware sox lib seems to have good press.
Old 2nd April 2014
  #2151
Quote:
Originally Posted by matyas View Post
The issue isn't the Nyquist-Shannon Theorem. We all know how that works. (Or should.) The issues are two:

1. Real-world converters do not always behave like mathematical models.

2. Hearing and comprehension of music is an intensely subjective process. We know much more about the physics of musical reproduction that we do about the psychoacoustics of musical cognition. Some people think that 22.05 kHz is insufficient with regards to bandwidth for the reproduction of at least some musical signals under some conditions.

The conflation of the mathematics of the Nyquist-Shannon Theorem (which, I readily agree, is pretty much irrefutable) with the psychoacoustics of musical perception (which is much more difficult to quantify) is fast becoming a pet peeve of mine.
Understandably, if someone attempts to conflate them. They are, indeed, completely separate issues and separate (if overlapping) fields of study.

That said, I know of no credible evidence proving that audio content above the nominal upper limits of human audibility is perceivable when it is 'missing' from normal listening material.

Some people think being abducted by space aliens is a significant problem, too.

Some people, eh? heh
Old 2nd April 2014
  #2152
Lives for gear
 
nuthinupmysleeve's Avatar
 

Quote:
Originally Posted by bogosort View Post
This is the worst possible way to test! One of the most important aspects in converter design is the filtering; throwing a random editor's 48 dB/octave LPF at 20k will make mince meat of the upper band. If you don't hear a difference, then it might be time to go to the audiologist.

The proper way to do this is to use a high quality sample-rate conversion (SRC) program. The DSP coder will have carefully created a filter appropriate to the task. The freeware sox lib seems to have good press.
We're testing two different things.

I'm suggesting only a test to hear whether audio above 20k makes a difference.

I think your test would involve testing the same file at different sample rates, which would make it a test of the converter's performance at a specific sample rate AS WELL AS testing higher frequencies... i.e., you are adding additional variables.

I don't hear a difference, and my audiologist thinks my hearing is pretty good, thank you.
Old 2nd April 2014
  #2153
j_j
Lives for gear
Quote:
Originally Posted by paul brown View Post
thanks! i'll report back with a screenshot of the results when i am done.
Make sure the in-band gain is the same, too.

You just read a completely horrid ABX protocol that depends on the actual filter implementation more than anything else.

Given prior experience I have great faith in the ability of people to make bad filter designs.
Old 2nd April 2014
  #2154
Lives for gear
 
bogosort's Avatar
Quote:
Originally Posted by nuthinupmysleeve View Post
We're testing two different things.

I'm suggesting only a test to hear whether audio above 20k makes a difference.

I think your test would involve testing the same file at different sample rates, which would make it a test of the converter's performance at a specific sample rate AS WELL AS testing higher frequencies... i.e., you are adding additional variables.

I don't hear a difference, and my audiologist thinks my hearing is pretty good, thank you.
Fair enough, but a bit of thought should be given to the filter, otherwise you're only testing whether you can hear the phase shift of an 8th-order LPF at 20k. I'd recommend moving the cutoff to at least 22k and experiment whether using a linear phase filter improves things.

BTW, wasn't trying to take a jab at your ears; I actually have no idea if I would hear it either (I should test it). Was just trying to point out that a steep filter at 20k will definitely have in-band artifacts.
Old 2nd April 2014
  #2155
j_j
Lives for gear
Quote:
Originally Posted by bogosort View Post
The proper way to do this is to use a high quality sample-rate conversion (SRC) program. The DSP coder will have carefully created a filter appropriate to the task.
Um, I wouldn't count on that without measurement, myself. I've encountered very different tolerances hither and yon.

Hmm, is there some way to post a .wav here?

And you only want to filter, you do not want to change the sampling rate, that's just going to introduce more confusion.

octave (freeware) with the audio and signal processing packages offers one way to do this, and to have complete control over your filter. It does, however, require some DSP understanding.
Old 2nd April 2014
  #2156
Lives for gear
 
paul brown's Avatar
Quote:
Originally Posted by j_j View Post
Make sure the in-band gain is the same, too.

You just read a completely horrid ABX protocol that depends on the actual filter implementation more than anything else.

Given prior experience I have great faith in the ability of people to make bad filter designs.
what about if i go out of my DAC back into my ADC and then record the output using the ADC to change the sample rate? create one file at 192 and another at 48 from a 192 original.
Old 2nd April 2014
  #2157
Quote:
Indeed, but every single factor that deviates from the pure mathematical model also has a mathematical model. In other words, the real-world behavior of converter systems is perfectly quantifiable.
INSIDE the digital domain where math rules.

However, if we include the analog circuit components necessary for ADC and DAC, all of a sudden, we are back in the land of potentially idiosyncratic performance subject to variations in component/build quality, heat and other performance condition issues, etc.

Just sayin'.
Old 2nd April 2014
  #2158
Lives for gear
 
bogosort's Avatar
Quote:
Originally Posted by j_j View Post
Um, I wouldn't count on that without measurement, myself. I've encountered very different tolerances hither and yon.

Hmm, is there some way to post a .wav here?

And you only want to filter, you do not want to change the sampling rate, that's just going to introduce more confusion.

octave (freeware) with the audio and signal processing packages offers one way to do this, and to have complete control over your filter. It does, however, require some DSP understanding.
You can post wav files under some maximum size (I think 9 MB?) by clicking the reply button and clicking on the paper clip icon in the toolbar. An upload file dialogue will open. (If you don't see the toolbar in the reply window, click Advanced.)
Old 2nd April 2014
  #2159
Lives for gear
 
bogosort's Avatar
Quote:
Originally Posted by theblue1 View Post
INSIDE the digital domain where math rules.

However, if we include the analog circuit components necessary for ADC and DAC, all of a sudden, we are back in the land of potentially idiosyncratic performance subject to variations in component/build quality, heat and other performance condition issues, etc.

Just sayin'.
As long as the circuit is behaving linearly (and if it's not, you've got bigger problems), every single one of those analog performance variations is quantifiable and perfectly understood.

Just sayin'.
Old 2nd April 2014
  #2160
If all analog systems behaved with reliable linearity, we wouldn't need digital. heh


Topic:
Post Reply

Welcome to the Gearslutz Pro Audio Community!

Registration benefits include:
  • The ability to reply to and create new discussions
  • Access to members-only giveaways & competitions
  • Interact with VIP industry experts in our guest Q&As
  • Access to members-only sub forum discussions
  • Access to members-only Chat Room
  • Get INSTANT ACCESS to the world's best private pro audio Classifieds for only USD $20/year
  • Promote your eBay auctions and Reverb.com listings for free
  • Remove this message!
You need an account to post a reply. Create a username and password below and an account will be created and your post entered.


 
 
Slide to join now Processing…
Thread Tools
Search this Thread
Search this Thread:

Advanced Search
Forum Jump
Forum Jump