Hi-res downloads versus upsampling

that explains the footsteps I hear on “All Blues” :crazy_face::astonished:

2 Likes

After a year of experimenting with upsampling vs. bit perfect and CD quality vs 24/x quality I have had to conclude that to my ears there is no difference. Hence, my music library has shifted to very good quality CD rips at 16/44 rather than ‘full fat’ hi-res stuff in .wav format which is just crazy IMHO. The only time I can see any value in going with 24 bit is that it gives you plenty of headroom for DSP stuff that can reduce the bit depth (but won’t get anywhere near 16 bits).

2 Likes

I’ve come to the same conclusion. I use flac (instead of wav) files to save storage space. I don’t hear any difference between the two formats.

2 Likes

It really shouldn’t surprise you given that CD represents a perfectly reproduced waveform between 20-22.5kHz and 96dB of undithered dynamic range (more with dithering).

Mark. We’ll have to disagree. Frequency response is only part of the game.

1 Like

The other being bandwidth.

Hi-res gives you more higher frequency response (sampling rate) and more bandwidth (bit depth).

What else do you think it gives you?

The sad truth of the matter is you simply don’t know what you are getting until you buy it and put it in your system - that applies to both CDs and Hi-res downloads. Sometimes you have a CD source which was mastered well and the Hi-res version doesn’t add anything, on the other hand you may also end up buying a Hi-res download that was mastered properly and completely different to the CD and thus sounds miles better than the CD master which was highly compressed etc. It’s a lottery. Thankfully thus far the Hi-res downloads I’ve purchased are mostly better mastered than the CD version of the same song/album (usually in the form of less hot, not as prone to clipping and also much less dynamically compressed), which makes them worth the purchase.

3 Likes

@HWZ : That was an excellent post ! Understood and (mostly) agreed to. Although I am in a different financial position :slight_smile: .

Regarding the following posts : you may not see it, but this sequence of posts is actually going somewhere. And that point supports the main doubt I have with the previous paper :

Indeed, high sample rates don’t give you any other advantage, than an extended frequency response. No, really nothing at all.
Before you ask : no, not even ‘more precise timing’. This is already perfectly encoded within the 44.1kHz file. Yes, I wrote ‘perfectly’: there is nothing to improve upon.
Very basic sampling theory, proven in practice and mathematics for decades.

But this does NOT mean, that there are no differences. The extra >20kHz content will be fed into the rest of your playback chain.
Since there is no such thing as a ‘flawless’ playback chain, this content will be subject to distortions. Just like what happpens to your normal redbooks.

The problem is : these distortions can shift down to (what I’ll now call) the audible part of the frequency spectrum. The part below 20kHz. Especially the case with intermodulation; you’ll find them mostly in the amplification and loudspeaker stage.
This is not so hard to prove by measurement.

Now suddenly something in the audible spectrum is happening. Barely audible, most people would not notice, but it’s there. And it’s not a change for the better either…

This is what I wonder about, in the paper. It seems that people can be trained on detecting a redbook/hires difference. But what is being detected there ? The actual benefits of the hires file ? Or the additional problems that this file created ? The studies tell me nothing about it…(unless I have been a lazy reader).

1 Like

Time domain accuracy.

1 Like

Hi-Res can be 10-15% better than 44.1 tracks of the same recording, assuming that they are mastered from the original tape in hi-res. The difference between 24/96 and 24/192 is small, but obvious in highly resolving systems.

Generally, the vocals will be less hissy and more smooth and natural and cymbols will sound more like bells and less like running water. Better decay.

Upsampling must be done with really good DSP software. I don’t know what Roon uses. The best way to upsample is to convert the file using a really good DSP like Wave Editor or R8Brain and then play the new track. Don’t ever upsample on the fly IME.

Steve N.
Empirical Audio

Please explain to me (I’m an idiot) how a 16/44.1 file is less accurate in the time domain than a 24/192 file, given that at the Nyquist frequency an audio waveform is perfectly and reversibly encoded.

1 Like

Adding more samples makes it closer to the original analog event, even if they are not perfectly positioned or perfect amplitude. Still closer to the amalog than the abrupt changes of a lower sample rate.

Steve N.
Empirical Audio

No, it doesn’t.

1 Like

I feel the Roon team should strive to become gurus on this subject so they can enlighten and educate the uninitiated and mostly ignorant hobbyists like me! Apologies to those in the know. As often the case with forums like this there are many people with differing opinions and points of view (flat earthers and agnostics) mixing fact and fiction. We’d like, “Justthe facts M’aam.”

Yes-No-Yes-No :). A discusion is not really going anywhere like that :).

To me it’s pretty clear that @anon55914447 is the one of the few here, with at least some understanding on sampling. “Mr. Idiot” is probably asking for a layman’s explation. And I guess he already knows, that it’s not really possible to explain this in layman terms…

A signal in a 44.1kHz file can have any start or end time, or any phase you wish. Really any. Not limited by steps or anything. There is simply no room for improvement there.

I’m sorry already, but : I am also not capable of explaining that, without you having background knowledge.
But I do have two suggestions :

  1. Do all Roon guys work in the same office ? If so : I suggest @joel talks to @brian. I’m pretty sure that Brian understands the point being made.
  2. If not, or for others interested : I highly recommend to watch the following video. Start to (most importanly) ending.
    The best ‘explanation’ for laymen, I’ve ever come across. The nice thing : hardly anything is explained. It is shown (visually). You won’t need to understand, to see why the truth is very diffent to what you thought.
    Movie here : https://www.youtube.com/watch?v=cIQ9IXSUzuM

Alternatively, you can also decide to stick to your own opinion. That’s perfectly fine with me. But in that case, please refrain from stating facts in this regard.

1 Like

Ah, so that would be high res at 48.51-50.715 khz sample rates.

AJ

1 Like

That’s a great video.

Nyquist-Shannon’s sampling theorem states:

If a function x(t) contains no frequencies higher than B hertz, it is completely determined by giving its ordinates at a series of points spaced 1/(2B) seconds apart.

There are two important points to make that are not totally coming across in your argument:

  1. Shannon’s points are real numbers, not finite-resolution integer values like we find in digital audio systems.

  2. The theorem only holds for band-limited signals. Sound in the air is not inherently band-limited.

Nyquist-Shannon as applied in the digital audio domain is an approximation of reality, not the whole thing. The engineering choice of Fs and sample resolution cannot be ignored. Nyquist-Shannon only guarantees that after the signal has been band-limited, we can sample and reconstruct the signal without loss of information.

Which brings me to this…

Please explain to me (I’m an idiot) how a 16/44.1 file is less accurate in the time domain than a 24/192 file, given that at the Nyquist frequency an audio waveform is perfectly and reversibly encoded.

If the input signals are the same, then there is no difference, so long as we are just talking about the sampling step. In practice–the 44.1kHz file was probably produced with a signal that was more strictly band-limited than the 192kHz file, and band-limiting is not information-preserving.

I don’t think that anyone would argue that the same information is preserved in these two situations:

  1. Analog audio signal, band limited to 96kHz, sampled at 192kHz
  2. Analog audio signal, band limited to 22.05kHz, sampled at 44.1kHz

#1 clearly has more information preserved. Whether or not that information is important or “worth it” is a very different question.

There is some apparent confusion in this thread about “frequency domain” and “time domain”, and arguments flying around that treat them as if separate. The best way to think about them is as a representational choice, and not as different information. The same information is represented in both the frequency and time domain. In the frequency domain view, a higher sampling frequency allows you to represent higher frequency components to the sound. In the time domain view, a higher sampling frequency allows you to represent a steeper transient.

Marketing justifications for keeping more around may be justified by one or both views. Because it’s scientifically more established that humans don’t hear tones very far above 20kHz, anyone with a business interest in pushing higher resolution content will probably focus on the time domain since there is more “wiggle room” for making claims in that area.

How high to go with those parameters, and what effects it has for real world listening are where the trouble is. Is 44.1kHz enough? Can higher frequency content produce perceptible differences even when humans cannot prove that they hear tones of those frequencies? What is the effect size of preserving more information in source material? These questions are, unfortunately, not very well explored scientifically.

I’m going to shift topic a bit to the original question in this thread.

The #1 reason to use a hi-res download is because sometimes a human paid more attention, or more careful attention when producing the recording with an audiophile customer in mind. This is not always true, but since high-res is a premium product, it sometimes is. The effect size of revisiting the production is far greater than anything else we are discussing here.

The #2 reason is because the hi-res download probably has not been so stringently band-limited, and thus more information is present. This is not necessarily the case, of course–just because you have a 192kHz recording does not mean that 80kHz frequency content is intact. But this does create an increase in the information available to downstream processes. Storage is cheap, if it’s easy to take advantage, may as well in case there are benefits to realize, now or in the future.

Upsampling is a rather different topic. It’s not a replacement for higher resolution source material, and is much more about the playback side–it is mostly used as a tweak to the analog reconstruction process.

7 Likes

Time domain in my own opinion is more important than frequency domain. Frequency domain can be addressed by over-sampling techniques but digital filters also create time domain distortion in the form of pre and after ‘ringings’. Our ears are more sensitive to time domain distortion. Unfortunately not much research is gone into this area.

A 44.1kHz is more than enough to cover the whole audio range up to 22.05kHz, however the very steep digital filter have to happen before that causing ‘ringing’ in the audio range, this has an effect of sounding ‘harsh’, ‘hard’ and not natural sounding. Moving to 88.2kHz, a lesser steep digital filters can come gradually around 40kHz with less ‘ringing’ since 40kHz is far from 20kHz, our ear is less sensitive to the effect of ‘ringing’ occurred at 40kHz. Hi-Res is just not about capturing more information, it also shift the digital filter above the 20kHz audio range, thus improve time domain distortion.

The design of digital filters have a great effect on the SQ. Some digital filters are less steep thus creates less ‘ringing’ effect however, it also allows aliasing effect to propagate into the audio range. There’s must be ‘balance’ between frequency and time domain correction.

Hi @Brian!

That’s a very important point!!! In this thread, we should perhaps differentiate more clearly between music production and music reproduction. I don’t think anyone would seriously claim that the difference in sound quality between Hi-Res playback formats and non-Hi-Res playback formats can vary anywhere near as dramatically as the overall quality of the whole recording, mixing and mastering process that precedes the release of an album. As an experienced session musician who’s worked in recording studios almost all over the world, I know that even many sound engineers who record at 24/96 (or higher) and always work extremely hard to achieve the best SQ possible aren’t really sure if spending extra money on Hi Res files is worth it. That’s why I think it’s a great thing that music lovers who think they don’t hear any difference between the two kinds of playback format can still benefit from such audiophile productions even if they download (or stream) the CD-quality version of the album they’re interested in.

2 Likes

That was a very good post @brian, thanks a lot !
I somehow doubt however, that the actual messages within you post will be correctly received/understood by the average reader…
(Which would definately not be your ‘fault’; you did way better than I could have done).

I do wish to add, that the audibility of this whole subject has been explored scientifically pretty deeply, and for quite some time already.
The simple observation that it’s so hard to make strict conclusions on them, can (or should?) be an indication for the significance of the whole story…

No further important objections, really. Excellent writing, thanks again.