It doesn't seem to follow that conversion transparency necessarily leads to a positive psycho-acoustic result. This was particularly the case for file 24 "God give me strength”, that for me was obviously more transparent. The issue was that in being so transparent it became more revealing in terms of alternative distortions being coherently connected to the source material. In other words the sound became more dynamically etched, shifting the presentation more forward and upwards in the frequency spectrum. This is generally bad from an enjoyment perspective as seemingly the reason why nearly everyone would choose file 7, or perhaps an MQA rendition.
From my perspective there are two distortion factors that contribute to enjoyment, that arguably can be described as coherent and incoherent distortion. Incoherent distortion is distortion to degree unrelated to the source material (noise artifacts for example) as such relates to transparency, whereupon coherent distortion relates to harmonic distortions. In studies on temporal coherency this relates to the manner that a subject focuses on things being heard. It can be surmised that this same mechanism of temporal coherency would also magnify the contributions of harmonic distortions being aligned with naturally occurring reproduction of sounds. This suggests that it is essential that transparency (as lacking incoherent artifacts) requires low harmonic distortions (as lacking coherent distorted artifacts) to support realism as thereupon enjoyment.
Your are constantly embedded in a sea of harmonic "fluff" 
?
alt.
your speakers don't reproduce all of the input so the input needs a "boost"...
In a reproduction system you don't want any additions or subtractions. If such mechanisms seem to improve the output, there is something fishy in there somewhere somehow....
//
?alt.
your speakers don't reproduce all of the input so the input needs a "boost"...
In a reproduction system you don't want any additions or subtractions. If such mechanisms seem to improve the output, there is something fishy in there somewhere somehow....
//
Last edited:
Transparent means no audible difference between input and output.
It's relatively easy to test for this with any audio device, the famous abx tests.
If the harmonic distortion is low, a few percent, it's inaudible.
But turn up the volume on a set, you will hear it.
I also have a tube dac with passive iv, sounds exactly the same as any comercial dac to me in a properly conducted ABX test.
Hi, Gerrit,
I've seen two definitions of subjective transparency among audiophiles. One definition is: The revealing of more information. Which, is what, I believe, you are addressing. The other definition is: Not altering the information. Yes, shades of the power amplifier distortion debate. I'm not a pure objectivist, so bear with me for a moment.
I think that, for subjective testing, the "revealing of more information" type of transparency testing can be elusive/deceptive to determine, as it requires one to qualitatively listen for which choice sounds better or worse. Doing that, involves personal preference regarding sound characteristics, such as, sounds brighter, more dynamic, more rhythmic, etc. It also opens the door for mistaking artifacts, for an improvement in those same sound characteristics, IMHO.
"Not altering the information" type of subjective transparency testing can be easier to determine differentially. Meaning, by comparison listening only for a difference in sound from a reference against it's derived alternative, and not for which sounds better or worse. Should there not be a statistically significant difference between the reference and it's derivative, then the process which created the derivative is considered subjectively transparent to the reference. Obviously, one must not know which is which, until after testing.
There is a potential complication, however, when it comes to comparing certain processes, such as digital interpolation-filters as we are here. It's possible for a digital interpolation-filter to both, not add audible undesired artifacts to a reference signal that is passed through it, while also qualitatively improving upon that reference signal. For example, by increasing the performance of the interpolation-filter utilized to create the resampled file, over that of a prior, lesser performing, filter (such as in, it's transition-band and stop-band performance) which was utilized to originally produce the reference file. In this way, the resampled signal is thereby altered. I don't consider this alteration a degradation, however, because the higher performing filter could be removing undesired artifacts left by the lower performing interpolation-filter. I don't believe, however, that the listening results tabulated by Hans reveals such an effect occurring.
Last edited:
I just received an interesting PM reaction from DDAC concerning the 88.2/24 versus the original 44.1/16.
However the reactions with 88.2/24 experiences can simply be send directly to this thread.
For NOS users the following can be said.
1) In case no differences with the original can be noticed, this will mean that removing the frequency band from 22.05 Khz to 66.15 khz does not influence sound perception and that the used up-sampling process has no noticeable impact.
2) When the sound improves at 88.2/24, this can only mean that the getting rid of of the frequency band from 22.05 Khz to 66.15 Khz influences the sound perception in a positive way.
3) And when sound looses some of it charmes, it will probably mean that the used processing has a negative impact and that removing the band from 22.05 Khz to 66.15 Khz cannot compensate this.
Hans
However the reactions with 88.2/24 experiences can simply be send directly to this thread.
For NOS users the following can be said.
1) In case no differences with the original can be noticed, this will mean that removing the frequency band from 22.05 Khz to 66.15 khz does not influence sound perception and that the used up-sampling process has no noticeable impact.
2) When the sound improves at 88.2/24, this can only mean that the getting rid of of the frequency band from 22.05 Khz to 66.15 Khz influences the sound perception in a positive way.
3) And when sound looses some of it charmes, it will probably mean that the used processing has a negative impact and that removing the band from 22.05 Khz to 66.15 Khz cannot compensate this.
Hans
I listened to the 88 tracks versus the 44 and I found it difficult if not impossible to hear a clear difference...
Maybe this that the special filter software with millions of taps is doing a good job? With the earlier processed ones it was more clear
So I wanted to know and get myself an indication and did the following:
I up-sampled the original 44.1 tracks with Roon to 88.2 while playing/listening and compared them to your 88.2 tracks and there I can hear, even if it is little, the same feeling/difference, that there is less air and micro detail in the Roon ones...
it is marginal. Now, as the differences heard between the org 44 and your 88.2 versions is for me not significant - that WAS the case with Roon up-sampling.
I am assuming that the Roon up-sampler isn't bad at all as it can use the high power of a windows PC, but that it isn't as good as that special software? Also Up-sampling in a digital chip is certainly limited compared to a strong Windows PC....
Ken, you are every time pretty good to connect some dots amongst these experiences, so what is your take on this?
Maybe this that the special filter software with millions of taps is doing a good job? With the earlier processed ones it was more clear
So I wanted to know and get myself an indication and did the following:
I up-sampled the original 44.1 tracks with Roon to 88.2 while playing/listening and compared them to your 88.2 tracks and there I can hear, even if it is little, the same feeling/difference, that there is less air and micro detail in the Roon ones...
it is marginal. Now, as the differences heard between the org 44 and your 88.2 versions is for me not significant - that WAS the case with Roon up-sampling.
I am assuming that the Roon up-sampler isn't bad at all as it can use the high power of a windows PC, but that it isn't as good as that special software? Also Up-sampling in a digital chip is certainly limited compared to a strong Windows PC....
Ken, you are every time pretty good to connect some dots amongst these experiences, so what is your take on this?
Slowly an image is developing that music seems extremely sensitive to processing and that the supersonic images above 20Khz, as generated by a NOS Dac, are not or hardly influencing the perceived sound (unless some part of the reproduction chain is driven into IMD distortion by those supersonics).
Audacity was clearly inferior in doing the job of up- and down-sampling, the same might be the case for Roon and according to my own experience for JRiver.
After all, up-sampling cannot add any information and therefore cannot improve sound. It can only distort the original signal to some degree.
The up-sampling job should be to just leave the signal content intact and produce interpolated samples to transform this content to a higher Fs.
That's what PGGB seems to have achieved in a very good way, proving the importance of a very long Sinc filter.
It is almost jumping to conclusions, but I think we are on the way of answering the issue of this thread, which starts to point strongly in the direction of a not good enough up-sample process being the cause for the perceived difference in sound between NOS and OS.
Hans
Audacity was clearly inferior in doing the job of up- and down-sampling, the same might be the case for Roon and according to my own experience for JRiver.
After all, up-sampling cannot add any information and therefore cannot improve sound. It can only distort the original signal to some degree.
The up-sampling job should be to just leave the signal content intact and produce interpolated samples to transform this content to a higher Fs.
That's what PGGB seems to have achieved in a very good way, proving the importance of a very long Sinc filter.
It is almost jumping to conclusions, but I think we are on the way of answering the issue of this thread, which starts to point strongly in the direction of a not good enough up-sample process being the cause for the perceived difference in sound between NOS and OS.
Hans
Just to be sure that I understand correctly: upsampling a 44.1/16 to 88.2/16 without reconstruction filter means every sample is repeated, i.e. duplicated. So the new 88.2 intermediate sample is identical to the previous 44.1 sample. And with a recontruction filter the new 88.2 intermediate sample is calculated by a clever algorithm. In the test we compare the two 88.2 versions, right? Sorry for the sloppy interpretation.
Thanks, Doede. However, I think that I'll wait a bit longer for listening results to trickle in, and to think about what it all may indicate, before offering my assessment.
yep, agree on jumping .... but my hunch was always that (poor oversampling and filtering) and is now strengthend by the last several tests we did
Doede, this result is 88.2 versus the original 44.1, or the up/down resampled 44.1 files?
There is no doubt that exceptional results can be achieved with what you are using. Notwithstanding the contribution of other electronics, a pair of mono-block Cary 300b amplifiers sounded extraordinary with a pair of high efficiency 15" Tannoy's. The problem with such a setup is that harmonic distortions, as can be contributing to high quality reproduction, are not linear with gain. This means that quality variances can drastically change as a function of listening level preference, and perhaps most particularly the efficiencies of loudspeakers being used. This doesn't exclude that this can yield superior results, just that variances can be time consuming and expensive in being worked out and/or changed.
Obscurity can be considered an antonym to transparency as something I want to address later.
Can you confirm something that is implied in your results? Which is that the PGGB 88.2 upsampled files, essentially, produce the same sound as the original files played 44.1 NOS? Meaning, the 88.2 up-sampled files produce the positive subjective characteristics typical of NOS?
Except, that there is no clever algorithm preforming the interpolation. No guessing. It's simply a SINC function based low-pass filter. Guessing isn't required, because the signal need only be sampled at a rate of at least twice the highest signal frequency (in band-limited sampling systems) to perfectly reproduce it later. If you don't accept that truth, then you will never accept that samples are not 'missing', and, therefore, guessing them is not required to perfectly reconstruct the original analog signal.
For the 88.2 experiment, there is only one version of 88.2 sample rate file for each song, so there are only four in total. In the test, you need to simply choose for each of those four 88.2 files, whether it sounds the same as, or worse than, or better than it's matching 44.1 sample rate file from the 'transparency' test we just finished.
Last edited: