Hi Marcel,
Good point.
To the best of my knowledge, ZB used the same upsampling/FIR filter for the 24 as for the 32 bit versions, and only quantizing, dithering and noise shaping differed between both.
So levels in that case should be absolutely the same.
But I will ask ZB for a confirmation.
Hans
OK - so either there is no sound difference for anybody as no person over 3 yo hear above 22,05 kHz (but a potential threat to the rest of the chain) or there is a difference and then it is a fault.
The logical result is that a filter should be used.
//
The logical result is that a filter should be used.
//
Hans, I meant the levels of the 176.4 kHz, 88.2 kHz and 44.1 kHz files. I could imagine that the more you oversample, the more the level has to be reduced to prevent clipping, which might explain why no-one likes the 176.4 kHz files. Fedde already reported in post #1359 that the 44.1 kHz file seems louder.
Last edited:
Yes, I have included that in my questions to ZB.
The soon as I have his answers, I will return with his feedback.
Hans
The soon as I have his answers, I will return with his feedback.
Hans
Sorry that you seem disturbed by this Hans. Lets understand that your statement is suggesting that "for all future time it is irrefutably true" that something presented can never be true. This seems aligned with a common general statement that "if something can't be heard it therefore doesn't exist". This is equally along the lines of "if you are not for us, therefore you are against us". Conclusions don't necessarily follow from premises, though the former statement seems held by those holding high positions in the audio industry, including those in the AES.
I realize that using terms such as sharpening, blurring, etc., can be very confusing and counter to current understanding, yet to stop talking about them supports the eternal continuation of misunderstanding by acting to terminate all discussion. Correlation to reality, supported or otherwise, comes from an increased understanding of what is meant by such terms, not by the termination of dialog.
Gerrit
Here is the reply that ZB gave regarding the inter overs and level differences.
Inter sample overs would be different for 88.2k vs 176.4k, if the intersample overs exceeded full scale, then they would have been scaled to be within 0dB, so yes depending on the track there could be a level difference, but this would not adversely affect the quality of conversion. If the intersample overs do not exceed full scale, no scaling is done so only the peak value within the track will be different and this will be harder/nearly impossible to perceive (in terms of loudness)
At the same time he confirmed that the 24 and 32 bit vesions were created equally, making it not very likely to hear differences between the two.
But he proposed to create new 176.4/24 samples, this time without noise shaping, which was also not used for the 88.2/24 files.
So this really makes a difference with the current 176.4/24.
I asked him in case of having to alter the level to prevent inter overs, to specify what level decrease in dB this had caused.
I can then reduce the level of the original 44.1/16 files to the same level
and will also check this with the ISO level algorithm.
So we are back in bussiness again with fresh new ideas thanks to ZB’s positive contribution.
Hans
Inter sample overs would be different for 88.2k vs 176.4k, if the intersample overs exceeded full scale, then they would have been scaled to be within 0dB, so yes depending on the track there could be a level difference, but this would not adversely affect the quality of conversion. If the intersample overs do not exceed full scale, no scaling is done so only the peak value within the track will be different and this will be harder/nearly impossible to perceive (in terms of loudness)
At the same time he confirmed that the 24 and 32 bit vesions were created equally, making it not very likely to hear differences between the two.
But he proposed to create new 176.4/24 samples, this time without noise shaping, which was also not used for the 88.2/24 files.
So this really makes a difference with the current 176.4/24.
I asked him in case of having to alter the level to prevent inter overs, to specify what level decrease in dB this had caused.
I can then reduce the level of the original 44.1/16 files to the same level
and will also check this with the ISO level algorithm.
So we are back in bussiness again with fresh new ideas thanks to ZB’s positive contribution.
Hans
In looking at the PCM1794 data sheet page 10, figure 7, this shows (total harmonic distortion +noise) vs Supply voltage. In looking at this graph for a 5Volt supply voltage it shows dramatic increased in THD/noise at 192KHz. Can this being considered as a possible justification for not using sampling frequencies at 176KHz?
Hi Gerrit,
Yes, I reacted a bit irritated because I had the feeling that this valuable thread alienated from it’s goal, sorry for that.
And believe me, I’m the last who thinks that what can’t be measured cannot exist or anything in a direction that is blocking new ideas.
But there are limits.
The first sound image is mirrored in direction frequency wise.
Can you imagine sharpening a picture with it’s upside down and left right reversed image and that blurring would bring you back to it’s original state?
Well, I can only see this as creating a useless unrecognisable picture.
Hans
This is to confirm I have no “guts”
(my intestines are in a worse state than my stomach which in turn is in a worst state than the stomach of most of you)
My previous post was meant to say that I could not detect any difference in sound btn those two files. Not only the absence of detecting the “images” due to severe limited freq response of my ears but no difference below the 5.5kHz content btn those files.
I wrote it being intrigued by the “sounds terrible” in Hans post
Except Hans comment above, I haven’t found anyone’s else listening impressions on Marcel’s files:
George
(my intestines are in a worse state than my stomach which in turn is in a worst state than the stomach of most of you)
My previous post was meant to say that I could not detect any difference in sound btn those two files. Not only the absence of detecting the “images” due to severe limited freq response of my ears but no difference below the 5.5kHz content btn those files.
I wrote it being intrigued by the “sounds terrible” in Hans post
Except Hans comment above, I haven’t found anyone’s else listening impressions on Marcel’s files:
George
I think TNT's remarks about oversharpening were also related to these files. To me, the filtered file without images sounds very dull. The one with images doesn't sound dull, but it sounds grossly distorted.
In any case, these files were meant to get an impression of what 44.1 kHz sample rate audio sounds like to a cat, with and without a NOS DAC with no filters other than the zero-order hold. The only thing you can conclude from them for a human listener, if anything at all, is that NOS DACs would sound a lot worse if we could hear the images.
In any case, these files were meant to get an impression of what 44.1 kHz sample rate audio sounds like to a cat, with and without a NOS DAC with no filters other than the zero-order hold. The only thing you can conclude from them for a human listener, if anything at all, is that NOS DACs would sound a lot worse if we could hear the images.
I think we should be a little cautious in using digital image processing analogies toward digital audio processing. Still images are geometrically spatial constructs, while digital audio is a time construct. As such, capturing higher resolution (higher information level) digital images require digitizing at a higher sample-rate and/or at a greater data dynamic range. At some increased resolution, the limits of the human eye are exceeded. Capturing a higher resolution (higher information level) in digital audio also requires digitizing at a higher sample-rate and/or at a greater data dynamic range. At 44.1kHz, however, digital audio has already exceeded the frequency information level of the human ear. At 16-bits, it also has exceeded the sample dynamic range resolution level of most human ears, particularly within a domestic listening room, and certainly so at 20-bits.
If I correctly understand Hans' point, and please correct me should I be mistaken Hans, is that all the information needed to exceed the limits of the human hearing system is already present, even from CD. Perfect sound forever? 😱 What it appears that we've identified is that audible problems with digital playback have mostly to do with implementation related errors and artifacts in the DAC chain, one of which is typical OS filters. Becasue even the CD standard objectively exceeds the bandwidth of human hearing, the consequent image-bands are all necessarily rendered ultrasonic, and therefore are not directly audible even without electronic filtering. The image-bands are present in an electrical signal sense, but are subsequently acoustically filtered by the ear, and to a lesser extent, by the speakers.
One answered question, however, seems to remain regarding acoustic filtering. How much is necessary for subjectively perfect playback? From a sampling theory persepctive, the lowest frequency of the first image-band should be suppressed to below the dynamic range of the sample quantization. Which requires OS SINC interpolation-filtering to do so with 44.1 sampling. Acoustic filtering alone doesn't provide the objective level of suppression needed. An unanswered question, however, is whether or not NOS playback subjectively benefits from more than acoustic levels of image-band suppression.
If I correctly understand Hans' point, and please correct me should I be mistaken Hans, is that all the information needed to exceed the limits of the human hearing system is already present, even from CD. Perfect sound forever? 😱 What it appears that we've identified is that audible problems with digital playback have mostly to do with implementation related errors and artifacts in the DAC chain, one of which is typical OS filters. Becasue even the CD standard objectively exceeds the bandwidth of human hearing, the consequent image-bands are all necessarily rendered ultrasonic, and therefore are not directly audible even without electronic filtering. The image-bands are present in an electrical signal sense, but are subsequently acoustically filtered by the ear, and to a lesser extent, by the speakers.
One answered question, however, seems to remain regarding acoustic filtering. How much is necessary for subjectively perfect playback? From a sampling theory persepctive, the lowest frequency of the first image-band should be suppressed to below the dynamic range of the sample quantization. Which requires OS SINC interpolation-filtering to do so with 44.1 sampling. Acoustic filtering alone doesn't provide the objective level of suppression needed. An unanswered question, however, is whether or not NOS playback subjectively benefits from more than acoustic levels of image-band suppression.
Thank you George, your reaction confirms my opinion 😀 😀
In the meantime I received new files from ZB without noise shaping.
Two files have no level change, one is -0.18dB and the last -0.34dB.
When everybody agrees, this doesn’t have to be corrected.
I need some time to prepare everything, like the questionaire, and will distribute the files tomorrow morning.
Hans
In the meantime I received new files from ZB without noise shaping.
Two files have no level change, one is -0.18dB and the last -0.34dB.
When everybody agrees, this doesn’t have to be corrected.
I need some time to prepare everything, like the questionaire, and will distribute the files tomorrow morning.
Hans
That's too much. Usually levels are matched within 0.1 dB, as some can hear 0.2 dB differences. (I'm pretty sure I can't, but Lipshitz and Vanderkooy have found people who could hear that under double-blind conditions.)
Very interesting find. At least on paper these differences should be far greater than differences between the ofiginal files and the resampled files in the latest listening test. Everbody can draw his/her own conclusions...
I think that we are going to have to live with this very small level difference. No doubt, the AES would not approve, but this looks like it will be just another practical reality we will have to accept. My primary concern at this time is whether the new 176.4 files still sound decidedly worse than the 88.2 files. If they do, there's no point in worrying about a very small level differences in two of the files.
Last edited:
I'm not ready to accept this statement due to the fact that the rest of the chain that is needed in order to "listen to a DAC" (which is of course impossible and this fact is part of the essence of this sentence) contains so many worse "faults" than what the produces so the analysis is impossible as I see it.
We don't listen to DACs, we listen to a reproduction system (chain) that contains a DAC. Want to argue that we listen to the differences - in some sense but still the rest of the chain is there in both cases - so, no, not really.
If one could safely dismiss compensation mechanisms in play - I could.
//
Hi, TNT,
The home playback chain of each individual conducting the experiments is exactly the same for them when switching between test file sets. We do know with certainty that there is a difference in how the OS interpolation-filter performs because we are purposely introducing that difference via the files. It is the subject of the test. What we don't know with certainty is the degree to which a given playback chain reacts in it's own unique way to the shortened sample-period inherently caused by OS.
No doubt, some DACs are better designed with regards to the potential issues, such as jitter, slew-rate distortions of the I/V circuit, supply regulation, etc. to minimize the errors which might be undesirably induced by higher input sample-rates. However, to uniformly control such factors for an experiment would require everyone to utilize exactly the same DAC box, as well as streaming file source and other components. That just is not going to happen under the limitations of our circumstances. It falls under the heading of this NOT being a scientifically controlled series of experiments, and contributes to the findings as therefore being questionable. As has already been acknowledged, repeatedly. However, this is not the same as the findings necessarily being wrong.