Which best represents your opinion of 96kHz high res recordings vs. redbook?

[wgscott]

Can you hear differences, and if you can, what do you think is going on?

 

I'm hoping simply to get a range of opinions, not arguments about whether or not the differences are (or at least can be) audible. If I have omitted an option, please state it.

 

I, personally, simply don't know.

 

Here's a numbered, more elaborated listing:

 

1. There is no audible difference.
   
2. I can hear above 22 kHz (conventional audiology is wrong about humans)
   
3. I can detect ultrasonics some other way (a physiological response other than conventional ear-mediated human hearing)
   
4. I can hear aliasing artifacts on redbook recordings down-sampled in the studio and/or from the initial analogue-to-digital conversion
   
5. High-res samples the audible region more accurately (over-sampling within the audible range increases the accuracy of the analog to digital conversion process, even if not theoretically necessary)
   
6. Both 4 and 5
   
7. Only if a brickwall filter has been applied incorrectly (in other words, redbook done right should be indistinguishable, but many are done wrong)
   
8. Redbook cannot capture fast transients accurately (human hearing is not a fast Fourier transform)
   
9. Options 4,5,7 and 8 may all be relevant.
   
10. Options 2 and 3, but only because I own ultrasonic super-tweeters (only systems capable of ultrasonic reproduction can reveal differences in high-res material)
    

[davidbeinct]

I'm not seeing a poll. Did you intend this to be a poll?

[wgscott]

I had to edit and shorten the responses because of the software limits. Sorry.

 

This also means you have to be generous with how you interpret the options since I was forced to abbreviate.

 

I've also posted a numbered list with a bit more elaboration just below the poll.

[cjf]

I'm going with #5 and sticking to it because, well, its true

[kumakuma]

I am not sure I hear a difference *IF* the mastering is the same and done well.

 

My preference, though, is to buy all my music in the best possible format available in case my hearing and/or equipment improve in the future.

[wgscott]

I am not sure I hear a difference *IF* the mastering is the same and done well.

 

That's Option 7

 

My preference, though, is to buy all my music in the best possible format available in case my hearing and/or equipment improve in the future.

 

I agree the safest policy is not to throw away the data.

[bottlerocket]

I've seen the discussions before about how 96KHz does not matter etc so I was glad to see the option "High-res samples the audible region more accurately". I always thought of hi res as a higher sample rate and therefore a closer approximation of an analog signal (more and smaller stair steps in the digital signal) so this option to me represents a very clear phrase that captures the benefit very succinctly. I don't understand nyquist and all that well, so if anyone can add a layman's explanation in this context, i'd appreciate it. Or is this option just a fallacy for us layman to latch onto?

[jhwalker]

I've seen the discussions before about how 96KHz does not matter etc so I was glad to see the option "High-res samples the audible region more accurately". I always thought of hi res as a higher sample rate and therefore a closer approximation of an analog signal (more and smaller stair steps in the digital signal) so this option to me represents a very clear phrase that captures the benefit very succinctly. I don't understand nyquist and all that well, so if anyone can add a layman's explanation in this context, i'd appreciate it. Or is this option just a fallacy for us layman to latch onto?

 

The "finer stair step" idea is not valid. The Nyquist theory states that a signal can be completely, accurately represented using a sampling rate just over 2x the highest frequency in the source sampled. So a sampling rate of 44100 is just as accurate as 384000 *if* the highest frequency being sampled is less than 22050.

 

The problem comes when frequencies higher than 22050 (e.g., inaudible to humans) are included in the source - if those frequencies are not filtered out, they produce artifacts in the audible frequency range due to the way digital sampling works. So the engineers have two choices: either filter those higher (inaudible) frequencies aggressively, or use a higher sampling rate so that you don't have to filter them out to avoid artifacts.

 

This is one of the reason higher sampling rates are chosen, not for "better" sampling in the audible range.

[wgscott]

Thanks John, but my hope is to keep arguments about the relative merits of the options out of the poll thread.

[jhwalker]

Thanks John, but my hope is to keep arguments about the relative merits of the options out of the poll thread.

 

Fair enough - I was answering bottlerocket's question and forgot your original instructions. Sorry about that.

[Paul R]

The "finer stair step" idea is not valid. The Nyquist theory states that a signal can be completely, accurately represented using a sampling rate just over 2x the highest frequency in the source sampled. So a sampling rate of 44100 is just as accurate as 384000 *if* the highest frequency being sampled is less than 22050.

 

The problem comes when frequencies higher than 22050 (e.g., inaudible to humans) are included in the source - if those frequencies are not filtered out, they produce artifacts in the audible frequency range due to the way digital sampling works. So the engineers have two choices: either filter those higher (inaudible) frequencies aggressively, or use a higher sampling rate so that you don't have to filter them out to avoid artifacts.

 

This is one of the reason higher sampling rates are chosen, not for "better" sampling in the audible range.

 

I disagree in practice John - just because something can does not mean that more information does not help produce a better, more accurate, reproduction. Filters and all that come into play.

 

And actually, that seems to be somewhat the case no? When you reach DSD frequencies, you are pretty much emulating analog behavior I think. In any case, I am not saying you are wrong, just that there may be more to the story.

 

I honestly cannot think which option to go with on the poll though.

 

I have heard a 24/96 version sound the same as, or better than, or even worse than the corresponding 16/44.1 version. I think in general 24/96 has the potential to sound better, but it is far from guaranteed. I would have like to seen the poll extend up to 24/192K or beyond, and into DSD.

 

 

-Paul

[wgscott]

If you can't detect a difference with 96kHz, why bother with 192kHz or DSD?

[Skeptic]

I've seen the discussions before about how 96KHz does not matter etc so I was glad to see the option "High-res samples the audible region more accurately". I always thought of hi res as a higher sample rate and therefore a closer approximation of an analog signal (more and smaller stair steps in the digital signal) so this option to me represents a very clear phrase that captures the benefit very succinctly. I don't understand nyquist and all that well, so if anyone can add a layman's explanation in this context, i'd appreciate it. Or is this option just a fallacy for us layman to latch onto?

Digital signals are not stair-steps though. They are samples.

The stair-stepped images you see are not a true representation of the actual signal; it's done because it requires very little computation to display, and is easier to read than simply displaying a point for each sample.

 

When you represent a signal this way, it does indeed appear that higher sample rates have a positive effect in the audible region.

 

But when you actually look at the data correctly, or the analog waveform that comes out the other side of your DAC, you see that signals <22.05kHz are represented perfectly with 44.1kHz sampling.

 

Here is a demonstration of this:

As you can see, the digital signal looks awful when you get to 15kHz and beyond if you are using a stair-step representation of the signal.

But when you look at the analog output, it's a perfect sine wave.

 

I recommend watching at least a couple of minutes of the video from that start point (4m46s) but the whole thing is worthwhile if you have a spare 24 minutes.

 

The problem comes when frequencies higher than 22050 (e.g., inaudible to humans) are included in the source - if those frequencies are not filtered out, they produce artifacts in the audible frequency range due to the way digital sampling works. So the engineers have two choices: either filter those higher (inaudible) frequencies aggressively, or use a higher sampling rate so that you don't have to filter them out to avoid artifacts.

 

This is one of the reason higher sampling rates are chosen, not for "better" sampling in the audible range.

I would say that this is somewhat fallacious though.

In a huge number of recordings, there's a lot of spurious tones/noise in the 25-35kHz range - to the point that I would actually be concerned about your tweeters if it is not filtered out of some recordings.

 

So even if the higher sampling rates would in theory allow you to keep the filtering well outside of the audible range, there's still a lot of noise there which needs to be filtered out regardless.

 

You can maybe push the filter up a couple of kHz, but not by much.

 

The fact that I haven't seen a single person on CA complain of continuous piercing tones when playing back affected high-res recordings leads me to believe that no-one is hearing or otherwise affected by any of the ultrasonic content.

 

I agree the safest policy is not to throw away the data.

This is how I feel about it really. If the editing format was 96kHz, then it's a 96kHz file I want, whether that 44.1kHz file should sound the same or not.

 

It avoids the problem of say a 44.1kHz file being produced without proper filtering/downsampling applied, and if it turns out that higher sample rates do actually matter, well then I already have it.

 

 

I don't really have many albums where I have both the high res and 44.1kHz files which were produced from the same master. Most of the time the high res files come from a remastering project.

 

I'd say that getting the files in the original format used for mastering is nice to have, but sample rate is low down on the list of things that matter to me.

I'll take a better master at 16-bit 44.1kHz over a 24/96 remaster with a compressed dynamic range any day.

[cjf]

if for just a moment you ignore whether or not you can hear anything above 22Khz or not and only think about the recording aspect of capturing a performance I can't see how it can be argued that having more samples taken per second of a sound wave is not better then less samples being taken of that same performance. Nyquist be damned!

 

The more samples that are taken per second can only help to insure a less "digital" sounding recording. In doing so, you end up with a more seamless representation of the original performance which is why the DSD format sounds so appealing to many people, its got that continuous, uninterrupted "Analog" sound.

[Skeptic]

if for just a moment you ignore whether or not you can hear anything above 22Khz or not and only think about the recording aspect of capturing a performance I can't see how it can be argued that having more samples taken per second of a sound wave is not better then less samples being taken of that same performance. Nyquist be damned!

 

The more samples that are taken per second can only help to insure a less "digital" sounding recording. In doing so, you end up with a more seamless representation of the original performance which is why the DSD format sounds so appealing to many people, its got that continuous, uninterrupted "Analog" sound.

You get a smooth continuous waveform whether you're playing back 44.1kHz or 5.6MHz files unless you are using a broken-by-design NOS DAC, which will output a stair-stepped signal at lower sample rates.

 

Higher sample rates do not produce "smoother" waveforms.

They only appear to when they are improperly represented.

 

The signal coming out of your DAC will not be stair-stepped.

 

Timing precision is not affected by the sample rate either.

Here is a demonstration of representing a signal "between the samples" of a 44.1kHz signal:

[Paul R]

If you can't detect a difference with 96kHz, why bother with 192kHz or DSD?

 

Because you might be able to tell the difference at 192K, and DSD is almost always going to sound different, though it is arguable exactly why.

 

I can usually tell the difference between a 16/44.1 file played back at 16/44.1 and the same file converted to 96K on the fly, but that may have more to do with the DAC, Software, or some other component than the sample rate.

 

-Paul

[Boris75]

 

It avoids the problem of say a 44.1kHz file being produced without proper filtering/downsampling applied, and if it turns out that higher sample rates do actually matter, well then I already have it.

 

 

I don't really have many albums where I have both the high res and 44.1kHz files which were produced from the same master. Most of the time the high res files come from a remastering project.

 

+1; however, I don't know which box to tick so as to express this opinion.

 

I'd say that getting the files in the original format used for mastering is nice to have, but sample rate is low down on the list of things that matter to me.

I'll take a better master at 16-bit 44.1kHz over a 24/96 remaster with a compressed dynamic range any day.

 

+1. Example: Talking Heads '77.

[new_media]

You must have studied under my virology professor. He was the master of the multiple multiple exam question.

[wgscott]

You must have studied under my virology professor. He was the master of the multiple multiple exam question.

 

My wife is a virologist.

 

I, myself, refuse to give multiple choice exams.

 

Isn't multiple-multiple choice what you find on the medical boards?

[new_media]

My wife is a virologist.

 

I, myself, refuse to give multiple choice exams.

 

Isn't multiple-multiple choice what you find on the medical boards?

 

I don't recall any on the USMLE, but that was many years ago. Neither of my recent board recert exams had multiple multiples.

[sandyk]

I have heard a 24/96 version sound the same as, or better than, or even worse than the corresponding 16/44.1 version. I think in general 24/96 has the potential to sound better, but it is far from guaranteed. I would have like to seen the poll extend up to 24/192K or beyond, and into DSD.

 

 

-Paul

 

Likewise.

The fact remains that a large number of C.A. members markedly prefer DSD over RB CD.

[sandyk]

if for just a moment you ignore whether or not you can hear anything above 22Khz or not and only think about the recording aspect of capturing a performance I can't see how it can be argued that having more samples taken per second of a sound wave is not better then less samples being taken of that same performance. Nyquist be damned! - cjf

 

+1

[Jud]

I voted for 4, 5, 7 and 8 (i.e., option 9), which was closest to what I feel is accurate.

 

7 will always be true because a "correct" brickwall filter requires the signal to be changed faster than light, a physical impossibility.

 

I'm surprised to see no votes for 3, since there's some support for bone conduction in the academic literature (more than just Oohashi).

 

I'd like to know who's letting their 3 year old daughter loose on this site (the vote for "I can hear above 22kHz").

 

Re #4, I doubt many people know a specific sonic signature for aliasing artifacts, so folks could be getting those in their music unawares. The better the job that is done removing aliases, the worse the job that is done reproducing transients. (This issue is mathematically impossible to avoid with the filters in use today. But to put it mildly, opinions vary with regard to audibility of the transient problem.)

 

Re #5, depends what theory you are talking about. If it's the Nyquist theorem, then that's of course proven correct within its idealizing assumptions (instantaneous filter effect, infinite time). If it's digital filter theory, then since no filter is perfect, you can't really say *any* sample rate is "unnecessary." You can always make things a little easier for the filter, theoretically, with a higher sample rate. (No, this is *not* because the shape of the curve will be more accurately reproduced. It is to avoid filtering artifacts.)

 

Re #8, something like that appears to be true for most folks (that they can pass discrimination tests at better than Fourier limits). Here's the relevant figure from a recent academic paper (the red dots to the lower left of the red line, and the blue dots to the lower left of the blue line, represent human time vs. frequency discrimination better than the Fourier uncertainty limits for the test signal):

 

[Jud]

Dang, I forgot to include the requisite personal anecdote: I have a couple of recordings (same masters) in both 16/44.1 and 24/96, and I like the 24/96 better in non-blinded listening. I also liked player software oversampling better than in-DAC oversampling for each. It was quite a while ago I compared these recordings, but if I remember correctly the order of preference ran like this, from worst to best:

 

- 16/44.1 oversampled in the DAC

 

- 16/44.1 oversampled in player software

 

- 24/96 oversampled in the DAC

 

- 24/96 oversampled in player software

[gmgraves]

IMHO, none of the choices in the poll are in the least bit relevant. If one hears any differences between files at sampling rates higher that 48KHz, it has little or nothing to do with ultrasonics, per se. I'll grant that higher sampling rates are sonically superior for moving the sampling frequency further away from the audio passband (assuming 20 KHz as the upper limit of human hearing). In recordings I've made, I have captured performances at 44.1 KHz, 48 KHz, 88.2 KHz, 96 KHz, and 192 KHz as well as DSD, and frankly there is no difference in the presentation of high-frequency content (with the same mikes, mixer and playback chain). I don't pretend to be able to hear those frequencies above 14 KHz, but my 16-year-old "nephew" Brian (a buddy of mine's son) can and as a budding audiophile, I trust his ears for that region of the spectrum. However, the highs at 88 KHz and above all exhibit a similar high-frequency cleanliness which I can hear, that is missing at either 44.1 or 48 KHz. I put this down to moving the sampling frequency far enough away from the passband to not cause any harmonic interference with those frequencies thereby lowering perceived distortion.

 

On the other hand, 24-bit (or higher) is very easily and audibly superior to 16-bit in every way by anybody willing to listen. The increased headroom and greater dynamic range decreases noise and distortion, increases low-level detail, and gives music greater impact, and being able to capture virtually the entire loudness spectrum of human hearing from very near the threshold of audibility to pretty close to the threshold of pain imparts a realism to recorded sound that is otherwise unobtainable by any previous LPCM or analog methodology.