[Music Alchemist]

What parameters will be needed to judge the adequacy and what are the required values for those parameters. Speakers and headphones as you know are generally far worse in terns of distortion, noise and uneven frequency response than competent solid state kit kept within their limits, even good headphones can have up to or more than 1% distortion which would be regarded as unacceptable on lots of items.

 
Except electrostats, which have far lower distortion than anything else.

 

[nick_charles]

   
As you say, not all resampling is done well. Many folks also seem to agree that the resampling done by Windows itself is not amongst the best. Beyond that, resampling is a form of interpolation, which means that there are various different ways of doing it, and they do yield slightly different results. (Let's avoid another lengthy dispute about whether the differences are or are not audible; suffice it to say that there are different ways of going about it, and different opinions about which does a better job...)
 
But the better job should be quantifiable, surely  we can measure the results and look at the differences, are there not agreed on parameters for assessing the effect of X or Y ? 

 

[goodyfresh]

   
Except electrostats, which have far lower distortion than anything else.

Possibly lower than what's needed to make the distortion inaudible, though   On most decent cans, the vast majority of the higher distortion numbers are in the low bass frequencies, and most evidence points to the possibility that anything less than 3% distortion in sub-bass is next to inaudible to the vast majority of human beings, anyways ^_^  My V-Moda M-80s, for example, measure as having less than 4% distortion in the deep sub-bass when played at 90dB, and well under 1% at all frequencies higher than a couple-hundred Hz.  And lemme tell ya, honestly, I can never actually HEAR anything sounding distorted through them.  And I know I am definitely capable of hearing distortion, when it happens. . .the old cheapo panasonic earbuds, or the 20 dollar Sony MDR-ZX110's I USED to listen on, before I knew any better, have VERY clearly audible distortion in the bass frequencies, and sometimes all the way up into the mids as well.

 

[Music Alchemist]

  Possibly lower than what's needed to make the distortion inaudible, though   On most decent cans, the vast majority of the higher distortion numbers are in the low bass frequencies, and most evidence points to the possibility that anything less than 3% distortion in sub-bass is next to inaudible to the vast majority of human beings, anyways ^_^  My V-Moda M-80s, for example, measure as having less than 4% distortion in the deep sub-bass when played at 90dB, and well under 1% at all frequencies higher than a couple-hundred Hz.  And lemme tell ya, honestly, I can never actually HEAR anything sounding distorted through them.  And I know I am definitely capable of hearing distortion, when it happens. . .the old cheapo panasonic earbuds, or the 20 dollar Sony MDR-ZX110's I USED to listen on, before I knew any better, have VERY clearly audible distortion in the bass frequencies, and sometimes all the way up into the mids as well.

 
It's not that simple. Read this, if you haven't already.

 

[nick_charles]

   
Except electrostats, which have far lower distortion than anything else.

 
 
Okay, I looked up the measurements on the Stax SR007 the distortion is indeed very low (0.05%) 
 

 
...and the FR is very flat to 1K but beyond that it is business as usual would that be good enough ? If a codec caused inaccuracies in the 3k to 10K regions would they not be hidden by the peaks and valleys ? I doubt that these would be good enough for KE

 

[goodyfresh]

 

KeithEmo is right, whether or not the changes are AUDIBLE or not, there are MEASURABLE differences in the spectrogram and so-forth of a recording depending upon the resampling method used, as well as HUGELY measurable (mathematically, at least) differences in detail between, say, 16/44.1 recordings and 24/196 recordings.

HOWEVER, as he mentioned and said we shouldn't debate, whther those differences are ACTUALY AUDIBLE TO HUMAN BEINGS is a whole other question entirely, and is much harder to answer as it is subjective

Then, there's teh issue of mastering-quality, as Music Alchemist keeps mentioning over and over in many different threads where the issue of hi-res vs. standard-res recordings is debated endlessly.  And he's right, based on research I've done and actual experience I've had listening now that I have better quality cans.  The quality with which a version of a recording has been mastered, makes a MUCH bigger difference, audibly, than the resolution in which you have the file for that recording.  In fact, with how I've been checking out some of the different available Led Zeppelin masters and remasters, lately, I've found that some masters of certain songs sound much better, with more detail, in 320kbs MP3 format than other masters of the same studio-recorded song in 24/96 FLAC format O_O
 
 

And Music Alchemist, I'll read that aritcle in just a little bit, once I'm done with what I'm about to do. . .my roommate is insisting we play some Smash Brothers so that he can lord over me how much he pwns me at it, lol.

 

[Music Alchemist]

  Okay, I looked up the measurements on the Stax SR007 the distortion is indeed very low (0.05%) 
 
...and the FR is very flat to 1K but beyond that it is business as usual would that be good enough ? If a codec caused inaccuracies in the 3k to 10K regions would they not be hidden by the peaks and valleys ? I doubt that these would be good enough for KE

 
The SR-009 is more neutral and has better measurements. My SR-207 is very neutral as well.
 
Here is a quote from that article I linked to:
 

By comparison, the Stax SR-Lambda Professional of 1982 was rated at 0.007% distortion at 400 Hz at a live-concert level 100 dB SPL. Looking at its actual distortion curve, we can see that its distortion is about -67 dB (0.05%) from 30 Hz on up, at 100 dB SPL. Not only is this about one-thirtieth the level of the world's best conventional loudspeaker, this is measured at 100 dB at the ear, or at ten times the apparent acoustic power, and down to 30 Hz, not 45 Hz! Even at 100 dB SPL, these electrostatic headphones have only about 0.1% distortion down to 20 Hz!

 

And Music Alchemist, I'll read that aritcle in just a little bit, once I'm done with what I'm about to do. . .my roommate is insisting we play some Smash Brothers so that he can lord over me how much he pwns me at it, lol.

 
You should play as Ness. lol
 
About distortion in general...it's not always going to be extreme audible distortion vs super clear. Once you hear enough headphones, you'll see that it's more subtle than that.

 

[nick_charles]

  KeithEmo is right, whether or not the changes are AUDIBLE or not, there are MEASURABLE differences in the spectrogram and so-forth of a recording depending upon the resampling method used, as well as HUGELY measurable (mathematically, at least) differences in detail between, say, 16/44.1 recordings and 24/196 recordings.

HOWEVER, as he mentioned and said we shouldn't debate, whther those differences are ACTUALY AUDIBLE TO HUMAN BEINGS is a whole other question entirely, and is much harder to answer as it is subjective

 
 
I've never opined that there are no measurable differences, but I do not accept your use of the word detail here that is subjective unless you can give me a better definition, I'll accept the word data instead. I've done dozens of comparative measurements between the effect of different codecs, CD players and even cables (the measurable differences between cables were small however) . Actually whether the effect is audible is one any of us who have access to Foobar and and the appropriate (re)sampling mechanisms can do for ourselves. If one thinks there are audible differences one can test it empirically, if your kit is good enough for KeithEmo 

 

[arnyk]

   
As you say, not all resampling is done well. Many folks also seem to agree that the resampling done by Windows itself is not amongst the best. Beyond that, resampling is a form of interpolation, which means that there are various different ways of doing it, and they do yield slightly different results. (Let's avoid another lengthy dispute about whether the differences are or are not audible; suffice it to say that there are different ways of going about it, and different opinions about which does a better job...)
 
However, since most DACs can handle all the standard sample rates perfectly well, isn't it simpler to avoid the question the easy way - and NOT resample the audio?
(Assuming that your DAC can play the file at its native sample rate, why go out of the way to alter it, when doing so might affect the quality negatively, and serves no positive purpose at all.)
 

 
The above seems pretty weird to me with its apparent attempt to escape from the crux of the matter, which is whether resampling is or can be audible or not.
 
There seems to be plenty of evidence that while resampling has been on occasion done badly in the past,  Similarly the usual kinds of resampling can be done today in a blameless way.
 
In short, its like just about every other kind of technology - a few dropped balls, and after a while a long history of doing it right.

 

[arnyk]

HOWEVER, as he mentioned and said we shouldn't debate, whther those differences are ACTUALY AUDIBLE TO HUMAN BEINGS is a whole other question entirely, and is much harder to answer as it is subjective
 

 
Whether or not certain technical differences are audible to human beings is not subjective, no more than the ability to run a 1 minute mile is subjective. It can be done or not, and its pretty easy to settle.
 
Most audiophile claims related to audibility are very similar to claims of being able to run a 1 minute mile, except I'm understating the true impossible nature of their claims.
 
Hearing certain sonic differences is a simple athletic feat, and we've been able to judge claims based on athletic events for thousands of years.

 

[goodyfresh]

   
Whether or not certain technical differences are audible to human beings is not subjective, no more than the ability to run a 1 minute mile is subjective. It can be done or not, and its pretty easy to settle.
 
Most audiophile claims related to audibility are very similar to claims of being able to run a 1 minute mile, except I'm understating the true impossible nature of their claims.
 
Hearing certain sonic differences is a simple athletic feat, and we've been able to judge claims based on athletic events for thousands of years.

Good point

 

[KeithEmo]

   
The above seems pretty weird to me with its apparent attempt to escape from the crux of the matter, which is whether resampling is or can be audible or not.
 
There seems to be plenty of evidence that while resampling has been on occasion done badly in the past,  Similarly the usual kinds of resampling can be done today in a blameless way.
 
In short, its like just about every other kind of technology - a few dropped balls, and after a while a long history of doing it right.

 
There's only one "problem" with your assertion. Most of the industry standard programs that do sample rate conversions do in fact offer various options. For example, when I choose to resample a file in Izotope RX 3.0, I am asked to choose settings for "filter steepness", "cutoff shift" and "pre-ringing". The documentation describes how each affects the results; and, while there are default settings, there is no explicit statement as to which setting is "right" or "most neutral". Since the more extreme settings do in fact produce easily audible differences, and there is no claim as to which settings will be "actually correct", which setting should we be using - if we want a totally neutral conversion? So, even if I were to agree (for the sake of discussion) that a totally neutral and inaudible conversion process is in fact available today, which program - and which setting - should I use to get it? I've also seen comparisons where several files were converted - from and to the same sample rates - using different conversion programs, and, again, the results were audibly different. Now, arguably, there might be some choice of settings on each that would all produce the exact same result, but I don't know what it is. So, obviously, at least some "reputable" programs produce conversions at their default settings that are audibly different. If you want to assert that a totally neutral conversion is in fact possible, then you'll need to tell me which program does it, and... er... I guess we need to do a study to confirm that you're right. To put that in the context of my previous statement about test requirements: Before we can test whether differences in sample rates are themselves audible, first we need to determine what program - and what settings - we should use to create test samples that we can confirm are OTHERWISE audibly identical.  
 
However, to respond to one of your previous posts asking for my idea for a good test protocol...... here goes.....
 
HERE IS MY SUGGESTION FOR A TEST PROTOCOL - TO TEST THE AUDIBILITY OF DIFFERENCES IN SAMPLE RATE
 
First, we need to state our intent, and some of the problems we expect to have to overcome:
 
1) We're trying to prove whether differences in sample rate ALONE are audible - specifically when we're talking about sample rates above 44k.
(So, for the sake of simplicity, lets limit it to one case: Are there audible differences between "the same content" when recorded at 16/44k and 24/96k?)
 
2) We aren't asking what percentage of people can hear such differences, or which equipment they're audible on, or with which sample content. The assertion is that the difference is INAUDIBLE, so, if a single test subject can RELIABLY AND CONSISTENTLY detect and note the difference on ANY equipment, with ANY test file, under ANY listening condition, then we will have proven that there are audible differences. (And we're not asking whether our test subject can tell which is which, or which he or she prefers - simply whether they can detect a difference or not.)
 
3) We've acknowledged that at least some programs will NOT give us a "perfect" conversion. Therefore, the fact that a certain observer can detect a difference with certain files will not be conclusive - because the audible difference may simply be an artifact of an imperfect (or "non-transparent") conversion.
 
4) We must acknowledge that we can't possibly test every combination of speaker, headphone, amplifier, test subject, and source material. Therefore, we need to find a way to minimize the possibility that we might get a null result simply because we chose equipment that isn't able to allow our test subjects to notice a difference that really exists.
 
5) Since some DACs "handle" different sample rates differently (different filters, different DSP processing, different oversampling ratios), we would like to rule that out as a factor. (What if someone hears a difference because the particular DAC he used applies a different filter to 44k files and 96k files, and the difference in those filters is audible.)
 
Since we don't have an unlimited budget to purchase every DAC, headphone, amp, and speaker ever made, but we still need to make a fair attempt to avoid "missing" the one combination that might prove the case, I suggest we leverage the idea of "motivated self-selected test subjects".
 
So here's my protocol.
 
1) First we select one or more conversion programs that we "agree" are "neutral". Since you are the one claiming that several modern programs meet this requirement, I'll defer to your judgment on this (or we can enlist some "industry experts" and ask them).
 
2) Then we secure a reasonable selection of "high quality 24/96k content. We can again ask some industry experts to tell us which content they consider to be "critical" or "difficult" (which content they expect a difference to be most likely with).
 
3) We will then, of course, generate a selection of file pairs - each consisting of the same content at 16/44k and 24/96k. 
 
4) We will now offer ten of these pairs for public download, so that people can try them for themselves, along with an offer/challenge. (This will serve both to solve our "equipment problem", to self-select interested individuals, and to provide motivation both for individuals to participate, and to self-screen themselves to avoid wasting our time.)
 
The challenge will be that, if an individual is convinced that they can in fact hear a difference, using their own equipment, and using the files we provided, they will be invited to participate in the study. They will be expected to pay their own expenses to transport themselves to the study location, and will NOT receive any payment simply for participating. However, if they can successfully identify the 24/96k version of the test files we provide 90% of the time, that will be considered "a significant result" - and, if they can demonstrate such a result, they will be reimbursed for their travel expenses, and will be given a $1000 cash prize. (We have allowed the "claimants" to pick - and provide - their own equipment; we have given them an incentive both to participate and to try their best to succeed, and we have provided them with an incentive to self-screen, and so to not waste our time with vague and specious claims.) We could add the option for them to provide their own 24/96k file, which we would then convert using OUR "known neutral" conversion software, and include in their test run.
 
Note that, since the "actual test" will be run by us, there will be no opportunity for the participants to "cheat", yet we have also provided them with sufficient motivation that we can expect their "best effort". And, since we have invited participants to provide their own equipment, nobody can claim that the equipment we provided "isn't revealing enough", or that we have otherwise biased the test to fail. (We can't claim to have tried all equipment, or all test conditions, but we have at least made a "fair try" to cover the equipment that the folks who claim to hear a difference themselves claim is "most revealing".)
 
* Incidentally, I see one thing that I've FAILED to account for.... #5 in my "intents and problems" section - the possibility that the particular DAC chosen by one test subject handles files of different sample rates differently, and so introduces audible differences between them when they're played that aren't due to the content itself or to the sample rate itself. This is actually a common problem with NOS DACs, many of which have a frequency response that is -3 dB at the Nyquist frequency (so they're -3 dB at 20 kHz with 44k files, but -3 dB at 48k with 96k files, which difference is clearly audible as a high-frequency roll off with 44k files, even though not strictly due to the difference in sample rate itself). There may also be other DACs out there which introduce similar, but less obvious, differences.  [Ooops.... I may have just demonstrated that even files that really are "information identical", but at different sample rates, will in fact sound audibly very different - with some DACs. Is it really reasonable to "arbitrarily" exclude those DACs from our study? This may make what we're trying to prove somewhat more complicated... or we may need to restate our intent to sidestep it. Now you see why designing GOOD test protocols can be complicated... and people make whole careers out of it ]

 

[arnyk]

   
There's only one "problem" with your assertion. Most of the industry standard programs that do sample rate conversions do in fact offer various optiions

 
That's quite a stretch - basing an argument that all resampling has audible defects on the fact that a few specialist tools have optional parameters. 
 
Your proposed protocol test fails on the grounds that it substitutes very awkward and inherently insensitive comparison procedures that only a few people can use for the ones we already have (e.g. Foobar2k ABX)  that are far more democratic, convenient and sensitive.
 
Why not do a little homework with a ready-made solution?
 
http://www.hydrogenaud.io/forums/index.php?showtopic=107570&view=findpost&p=904306

 

[mindbomb]

For asio, the windows settings don't affect it usually. For bit depth settings, it is easy for me to suggest 24 bit, since it can do 16 bit values as well. Sampling rate is kinda tricky. Some dacs prefer 48khz to 44.1khz, so you are better off resampling to 48khz. Sometimes there are firmware issues and you are better off going the full 192khz. Also, when you upsample in software (rather than oversample in hardware), you get to control the characteristics of it, so that might appeal to you. Assuming no weird dac issues, and you are listening to cd quality audio, you can say 44.1khz is probably the setting you want.

 

[goodyfresh]

Interestingly enough, guys, I've been experimenting in Foobar2000 while using my Fiio X3ii as a DAC, and have found that quite consisently, I get the least lag/latency in starting playback, pausing, and starting again when using Kernel Streaming output from Foobar2000 to the X3ii, than when using either the TUSB ASIO driver that comes with the X3ii OR WASAPI.  Of course, the sound-quality is the same with all three, but I seem to be (who am I kidding, I definitely am, it's actually quite noticable) getting the least lag between pressing play or pause, and teh music actually playing or pausing, when using Kernel Streaming. So I think I'm going to stick with that.