192kHz/24bit vs. 96kHz/24bit "debate"- Interesting revelation

Yep, I go that. We needed some humor.

 

I make it a point to never assume audiophiles are kidding.

 

Nick,

I found a paper that is also helpful for us. It is by Milind Kuncher in the physics dept. at the University of South Carolina. He did several experiments in a controlled blind listening test:

http://www.physics.sc.edu/~kunchur/papers/Temporal-resolution-by-bandwidth-restriction--Kunchur.pdf

 

Yet another paper by Milind Kuncher on topic.

http://www.physics.sc.edu/~kunchur/...isalignment-of-acoustic-signals---Kunchur.pdf

 

Interesting papers, Lee. I'm still reading and digesting. Without further testing, it's hard to say what effect it would have on reproduction of music. A 7 kHz square wave contains an unusual amount of ultrasonics -- far more than would be found with actual music (and with far higher amplitude). These results can't be generalized to make the case that ultrasonic frequencies must be reproduced in order to make digital sampling transparent. To me it's a bit like the Meyer/Moran test where the higher noise floor of 16 bit audio was audible compared to 24 bit when the signal was turned up 20 dB during quiet periods...that's not unexpected, but it doesn't represent a real-world scenario, since when the music returned to normal level it would be hitting the threshold of pain. At normal levels, listeners could not distinguish between 24 bit and 16 bit noise floors. It seems to me that more testing would be needed to see if Kunchur's tests on temporal resolution would apply to anything found in real music.

I also find interesting the section on "Implications for Sound Reproduction," which you partially quoted above. The other part:

I'm curious as to why he makes this reference because I don't believe Stuart actually carried out any listening tests in his paper, and it seems to be assumed here that "it is known that the bandwidth requirement for sonically transparent audio reproduction is higher than 20 kHz." It is most certainly not known for sure...that's why there's so much debate about it! The next sentence is also very suspect, starting with "It is sometimes thought that this may be due..." (!)

The tests are very interesting, but I am unsure of their actual implications to reproduction of music. I don't think it's time quite yet to throw out all the other testing and accumulated knowledge that has failed to demonstrate what seems to be "known" according to Kunchur. It is easier and cheaper today to demonstrate this with real music. The Meyer/Moran test for me is much more representative of real-world experiences, and listeners were stunningly unable to distinguish hi-res PCM and DSD signals redbook. It should be very easy for those with good listening skills to be able to distinguish between those two scenarios, since, not only was it redbook vs. hi-res, but there was an extra A/D/A loop in the signal path to create the redbook signal! If some believe that test was flawed, why have they not tried to replicate it? That would be the easiest way to debunk it. That particular test is unusually structured to give hi-res audio the advantage -- it would be the easiest test I've seen for listeners to discriminate between the two sources if audible differences exist.

I will be reading up on these experiments more to try to understand them. I am not an engineer and am only an enthused hobbyist struggling to grasp some of this technology. Thanks again for posting these very interesting papers.

 

Nick,

You are most welcome. I believe these papers are relevant to music because they establish thresholds of hearing and they prove that the hirez content can be heard. You don't always have to use music to test the value of things if you deploy a test that gets to the fundamentals of how sensitive the ear is. I think the value of Kuncher is that it shows how incredibly sensitive the ear is to timing information. And if you have proven that timing information (this is much like Mike Story's paper) is not robust at 16/44 then 24/96 or possibly more must be the answer.

I'm not an engineer either so the Kuncher papers get technical. I do analytics for a living so I can follow the math pretty well. I will try to help both of us by looking for any articles in plain English by Kuncher.

 

P.S. Many instruments like trumpet, xylophone, etc. routinely hit 7khz and well into the ultrasonic arena as well.

 

I agree you don't always have to use music to test things, but I think that would be the next logical step.

I don't think the data support such a conclusion...yet. It is certainly possible that what is demonstrated in this test would not apply to real-world situations. As I mentioned, a 7 kHz square wave was very carefully chosen because of the unusual level of the ultrasonics within 1 kHz of the normally accepted limit of human hearing. Someone would have to take this interesting result and apply it to testing different types of real music. Again, our goal is transparency, so unless we are listening to square waves, we can't directly apply this testing to indicate that 44.1 kHz sampling will not transparently reproduce real-world signals (music). As I mentioned before, it seems exceedingly difficult for listening tests to show that 44.1 kHz is not transparent, and I don't think this experiment refutes all of those other tests. It does certainly warrant further investigation, though. For instance, if my goal is to find a medium that will store square waves perfectly, PCM is a bad choice due to the requirement that signals be bandlimited. However, I don't think anyone would say that PCM is a bad choice for storing all digital audio because of this.

As I mentioned before, we can use completely improbable scenarios to prove a lot of things. Say for the sake of argument we could use fade-outs and very quiet passages amplified 40 dB to prove that even 24-bit quantization noise is audible...does that mean that listeners would ever hear that noise while actually listening to music? No. To prove that, we would have to conduct such a listening test with real-world examples.

I'm also curious what role IM distortion could possibly play in this or other tests involving ultrasonic audibility. I'm sure one would have to go to great lengths to be sure listeners aren't hearing distortion of some piece of equipment (very much in the audible range) when they're supposed to only be hearing ultrasonics.

 

What evidence suggests that 16-bit/44.1 kHz isn't as "robust" in terms of timing information as digital audio with higher sampling rates and/or bit rates? Because this Xiph video clearly demonstrates that timing isn't affected by these parameters.

 

Check the Mike Story paper I posted on energy smear exhibited by various hirez formats. It occurs from filtering problems in lower rez PCM. Mike shows the energy smear (as he calls it) in a chart that plots energy versus time. I believe the Kuncher paper provides evidence that humans respond to microsecond differences in such timing.

 

Nick,

I found a great "plain English" article on the Kuncher experiments: http://www.physics.sc.edu/~kunchur/papers/HIFI-Critic-article-by-George-Foster.pdf

 

This is great as it is all making sense and corroborating the Story paper. Kuncher explains:

That's your answer Nick. 24/192 solves the time smearing that Mike Story showed. There is your technical proof of why 24/192 has value. It is, based on Kuncher's experiments, the minimum sampling needed to match the "temporal resolution" of the ear/brain.

And this is the scientific basis of why I have heard improvement on the 24/192 layer of HDAD discs versus the 24/96 side.

And the experiments all presented in peer reviewed scientific journals.

 

Ya' know LeeS, you use terms like "plain english" I also remember you using "slow on the up-take". It makes me think of terms like "condescending".

 

Did you see the previous discussion where Nick and I both agreed that some of the Kuncher papers were a bit deeply technical? I simply found an easier to understand overview.

 

Again, I'm curious to see this listening test repeated with actual music. It's interesting that 44.1 kHz can't preserve a square wave transparently, but I wasn't aware that PCM could do that at all. These timing differences may or may not apply to actual music. I can't stand to listen to a square wave long enough to duplicate the experiment.

The other thing of note is that Kunchur's experiment is not double-blind. I thought that was the standard for scientific experiments, but apparently not. Single-blind tests are useful but there are many well-studied cues that can be given by test conductors unintentionally.

It's a very, very interesting study, but as I said, I think it's far from "proving" that 16/44 is not transparent for actual music (which until now has been supported by many controlled listening tests). Since good scientific results should be reproducible, I'm sure this will happen. Surely if 16/44 is easily trumped by hi-res as often claimed, it should be very simple to demonstrate in even the most rigorous tests.

 

Well we have an honest disagreement Nick. Kunchur is an accomplished scientist and has devised several ways to measure the temporal resolution of humans and that level is down to very low levels. And based on those levels, he can mathematically proved that 16/44 is not transparent enough. It's a very, very compelling argument for 24/192.

It's a shame you won't buy an HDAD and listen to the two different layers of audio on a good system. That is the easiest way to experiment and hear any differences in the two resolutions you are focused on.

It's also very frustrating that you cannot accept that a 7khz signal can be used to represent music. The HiFi Critic article goes into some detail on the scientific reasons why Kunchur chose that frequency.

That is the problem with these threads, no amount of quality published scientific experiments ever convinces people to change their minds.

And that's a shame because this hobby should be about experimentation and having an open mind. That leads to discoveries that improve the sound...much like Chesky and Classic Records' early hirez efforts led to great sounding, classic albums on disc.

It's also a shame you yet again fail to mention the listening tests from Ohashi to Pras/Gaustavino to Kunchur that do show that 16/44 is inadequate.

But your mind is really made up here. Enjoy your CDs.

 

How many times do I need to say that I *have listened* and do not hear the difference? Whether it was an HDAD or a 24/192 file vs a 16/44 one doesn't matter. I do not hear any difference with hi-res, and I don't expect to. I do admit that I am biased in that direction and I do not trust these results to be "fact."

 

If you're talking about something like I've used as an example before - The Staples Singers - Be Attitude (HDtracks @ 88.2 vs CD) - identical remasters from tape - I get it. I would never try to convince anyone one sounds different than the other.

If you are saying that you cannot hear the difference between hmmm let's pick an example - Soundkeepers sample tracks - then that is a different story. Whether you judge them good or bad, the formats sound different.

 

I made that point earlier. I was a little more subtle though - it's good to be the king!

 

And the arguments are so polarizing. Without spending too much energy can you think of a similar topic that you might have been or are invested in (besides mastering techniques i suppose)?

 

I've listened to 16/44 versions of songs I've previously downloaded at 24/192 and found the differences immediate and startling – but that's just me.
Listen to whatever you love and enjoy it, whatever the reasons for 24/192 sounding better (lack of time smears, etc.), to me it's a no-brainer, if I can hear the difference, why wouldn't I want the one that sounds better to me?

 

Any specific examples to recommend?