This means that the difference is measurable in a blind test. Blind doesn't mean people must be blindfolded or anything like that! It means that a scientific test was done where nor the testees, nor the tester, knew what was being used, to rule out any "placebo" effects.

But there is simply zero solid proof the new formats have ANY effect at all.



You seem to have the complete misconception that a digital signal is "blocky", but in fact, it cannot be blocky, because it's bandlimited!

1) Ambient Noise will always be present - and I cant change my ears either - therefore I consider both as constant - K. Therefore they can be ignored from the factor.

2) how often do i listen to peaks above 102db? erm..... who knows or cares? would i miss them if they were gone? probably not. The statement doenst make complete sense, since that db level is contained with 0 -102db on the cd - so that peak is INSIDE the dynamic range. I'm not missing out on anything - but they have compressed the dynamic rangte to fit. The point is: can i tell and do I care?

3) Preparing a sample you'd never listen to which can be abxed is fruitless.

consider this discussion as the choice between a 4*4 off roader and an F1 car.

Sure you could race them on different tracks and get different results, one better than the other in many ways.

Now consider this discussion is only interested in a vehicle which can be used for every application, with good performance in all..........the f1 car might rip it on the race track, but give it a pot hole and you're calling the AA.

My point is you must be practical with your investigation, and not seek to denounce anything, nor prove anything - just form a conclusion from your results.

I did that - and im now convinced that it really doesnt matter between the two formats.......and i already have a cd player in almost every room in the house. Ill go for CD thanks, its fine.......even if it does feel like an old rusty 4*4 in comparison..........it'll get me there - and much cheaper.

thats what im talkin about

I agree with everyone - even though i appear to have my own viewpoint!!!

I believe i can hear the difference too - if the tracks were split out - but I dont listen to music like that, so I have to discard abxing custom samples - and stick to the usual professional mixdowns I'm so concerned sound good....

Was the 16-bit signal properly dithered+noiseshaped on playback? A properly dithered+noisehshaped 16-bit signal has an effective SNR of 111dB. This is more than your DAC's.

I don't see the sense in comparing not properly dithered 16-bit signals to the new generation formats, when one of those (SACD) has an SNR of 6dB when not dithered!

Maybe, maybe, but outside controlled blind listening test, your experience bring no more information to us than the hundreds other user opinions. Remember the Terms of Service number 8 of this forum : http://www.hydrogenaudio.org/forums/index.php?showtopic=3974
Statistically significant blind listening tests only (see FAQ or Wiki).



It suffices in order to test the claim that people notice obvious differences between two recordings of the same Fleetwood Mac track, one in cd-audio quality
and one in higher quality (with good speakers).



These can't lead to reliable statistics. The tested hypothesis must be clearly defined, and the measured parameters must be chosen before the test.

Beginning with the number of people starting to dance, you will get subjective evaluation problems. Is this guy dancing on his seat ? Is that one dancing on the way to the toilets ? I this one one the dancefloor or next the dancefloor ?
Then if you don't get the result you like, you will test the hour of people going home, and if it fails, the highest they jump, and if it doesn't show any correlation, the number of drinks they buy, their distance to the speakers, the number of couples looking at each other, etc
The number of testable things is infinite, thus the probability that you get a statistically significant result out of nothing is equal to 1.

The measured parameters must be the ones relevant to the tested hypothesis : if high definition formats are supposed to sound better, we must test if they sound better, and nothing else. And in order to do this, we must ask listeners if the sound is better or not.

The refinements that can improve the test performance are in the way the listeners are allowed to listen. It is perfectly right, and even recommended when testing small differences, to choose listeners that can easily hear these differences, and then let them train themselves with no time or protocol constraints.
The most relevant test would be the one made by the trained listener, after he has found the protocol (AB, ABX, ABA, AXY, etc), and listening conditions that allow him or her to always get statistically significant results during the training sessions.

Imposing a given musical content, a given sample duration, and demanding listeners to write down any answer after each listening session, played only once, are the most important obstacles to the success of such blind listening tests in my opinion.
I think that relaxing these restrictions, while maintaining the randomness and overall statistical requirements of the final test, should allow blind tests to show the existence of smaller audible differences than they usually do.

sorry if this was already posted, but 96k may have its benefit in reproducing (synthetic) sharp square wave (or saw-wave)- it always adds sub harmonic frequencies to it, but 96k should add less than 44k (in a very audible way) but when was the last time you heard a synthetic square wave in real music.

Might be a good idea to read the thread you are replying to next time, I am sure that it would be quite enlightening.

The late Julian Dunn wrote a paper on dynamic range (bitdepth) requirements recording an distribution (playback) in 1992

http://www.nanophon.com/audio/dynrange.pdf

essentially concluding that 16 bit with noise shaping was sufficient for playback, while 20 bit was miminum for recording.

Well, let's see. Atmospheric noise at the eardrum is 6dB SPL give or take. 96dB over that is 102dB. Even without noise shaping, that's pretty close to the loudest most speakers can go.

How much does one assume they can get from noise shaping, and what shape are we proposing here?

While we're at it, what does any frequency over 30kHz have to do with human audition at all?

So let me get this straight, DVD-A and SACD will bring me no benefit in sound quality? There's no need to make the transition from CD to those formats?

They do if they feature superior mastering compared to their CD counterpart. Check out my thread from last summer comparing Porcupine Tree's Deadwing CD vs. DVD-A. http://www.hydrogenaudio.org/forums/index....topic=35572&hl=

A piano's highest frequency is 4186Hz. So, fuck it, let's sample the next-gen media down so that it only produces's frequency's up-to 4186Hz.

But wait, there's more!!



There's Life Above 20 kilohertz

So as you can see, there is a need for increased frequncy.

But hang on. My ears can only hear frequency's from 20Hz-20khz. Belive it or not, your subconscious mind can perceive sounds that your conscious mind cannot.

So while, Joe "doesn't know anybetter" Blow, who has never been to an orchestra, thinks that the piano he is listening to, derived from a frequncy limited 128kbps mp3, sounds transparent.
I think it sounds shit-house.


Piano, organ, bass drum all produce sound higher than 96db. The bass drum alone can produce up-to 115db. Mix in a few other intruments with it, and you can produce upwards of 120db.
It's all about dynamic range. And just because when your at a live orchestra, the dynamic range is say 90db due to conversation. When your listening to a recording at home, you should not be limited to the same dynamic range.

If your happy listening to music recorded in mp3 at 192kbps, Fine.
But don't bitch at companies that are helping the likes of myself, Enjoy music.
edit: or bitch at others, because "you" can't perceive the improvement.

Audionut.

No conclusive evidence found at that link to argue for increased bandwidth. The referred Ooashi paper was rejected for publication in the JAES(it never made it past submitted preprint status) and the Journal that did publish the paper(The Journal of Neurophysiology) only did so, classifying the paper as a paid for advertisement. The Ooashi paper never showed conclusive evidence of audibility(the mentioned listening test in the paper did not elaborate any of the test detail specifics, and NHK labs later did a follow up, and could not reproduce the audibility claimed results by Ooashi). Also, the Ooashi paper had very questionable results in the MRI scans. Note that no activity was detected for ultrasonic information by itself as a stimulus; only when both sonic and ultrasonic was produced. Very odd. Makes one wonder if something was up with their electronics or playback system(s). Or maybe, the entire paper is the result of poor researchers, or even fraudulent, since the only publication it could achieve is one that was an advertisement. I wonder if the MRI results(which are not proving any audibility) are reproducible.

-Chris

I think many of you here are missing the exciting aspect of the SACD/DVD-A, partly because you seem to be mainly focused on CD technology and it's accompanying MP3 format, and also, as suggested above, you're approaching it from a statistics-based view.

The fact is, it's basically an upgrade. And, of course Sony and the other companies are largely implementing it to make people buy more of their stuff, what else would be expected--but they can do all sorts of things to make people buy more stuff, which they've done. They seem to have done a piss-poor job of marketing SACD, so the idea that they've worked so hard on a new technology as a ploy to make people have to commit to a whole new technology doesn't seem to make sense. And you would think that's what they would do, but for some reason they haven't. Maybe they're waiting for some right moment.

I personally had been thinking about this for a while, before I had ever even heard of SACD and DVD-A, about the idea of higher density CDs; it just made sense to me. Then I finally learned of them and learned that for some reason they aren't being pushed at all, besides the fact that people are much more excited about MP3s than CDs that are higher than 44.1 kh and 16-bit, which is something that wouldn't mean anything to most people anyway.

The other major issue about CDs is that for many people Vinyl is still the highest fidelity, or at least the best-sounding format, and part of what people resisted about CDs in the first place is the idea listening to contiguous moments of music put together, where you're not actually hearing all of the music but a kind of simulated version. So basically, the way I would think about it, the higher the sampling rate (i.e. 196 kh), the closer you're coming to something that sounds more like analog and less "digital", while also having the ease of the digital format.

I'm not sure it's so much a matter of increasing the dynamic range as capturing more of the music, which I would assume the newer formats do.

This is basically a repeat of an earlier discussion...
http://www.hydrogenaudio.org/forums/index....topic=41162&hl=

And here's a very nice discussion about the higher sampling rates from James Johnston (co-inventor of MP3) not being necessary. http://www.skepticforum.com/viewtopic.php?...der=asc&start=0

You should note the title of this forum[Scientifid/R&D Discussion]. Your post does not seem to fit the discussion.
Did you read this thread?

-Chris

I read the first couple of pages or so. Maybe I missed something. Edit: Actually I read a lot of the first page and then read the last page. I think my post fits the discussion because I was addressing directly why some people would want higher bitrates/kHz, but maybe he was just asking about it whether it was technically a necessity whereas I just said why people would want it. I'm quite tired.

http://cm.bell-labs.com/cm/ms/what/shannon...shannon1948.pdf

Theorem 13, page 34.

"Do we need more bits/samplerate?" Episode #543.... location: hydrogenaudio.org....... pagenumber #5...... still trying to find even a subtle difference.......

Those "obvious difference" must be hiding themselves really good......... good enough that almost no one can notice it..... mean, we're on THE forum for knowledgeable people regarding psychoacoustics including quite a few golden ears..... and even they after 5 pages still cannot tell the difference...... if its so fucking hard to notice at all, then why the hell should we need it, even if the difference actually exists? How useful is something in practice which makes almost no or no difference at all?

Or is the reason for the talk in this thread by any chance not about if there actually is a difference, but more about that some people want to BELIEVE that there is a difference? If yes, then why the heck are you discussing this on ha.org? You will get much more support to believe somewhere else.

- Lyx

But seriously. I sometimes think of stopping to argument with people who claim that SACD and DVD-A really sounds much better and more natural and blabla, without performing proper blind testing. It's a never-ending story.

Agreed, and that's why I'd like to investigate the possibility of a "professional user listening test" in a (top-quality) recording studio. It will be very difficult to set up a "proper" listening test so I'm hoping to get some useful information here, even though HA seems mainly targeted at endusers. Any insights are more than welcome. The test is still in an early (alpha) stage, so most options are still open.

A serious test would take quite some time, with quite some listeners and quite some equipment. A period of 3 days seems to be technically and financially feasible.
The studio can provide:
-high quality musical instruments as a source
-several recording rooms with very low noise levels
-several (large) mixing consoles, both analog and digital, as used in normal production
-a large choice of top quality microphones, pre-amps, AD/DA converters (both PCM and DSD) and monitoring equipment (stereo and surround). Any equipment that isn't already available can probably be arranged for the test.
-several experienced, professional recording/mixing/mastering engineers. Total group should probably be limited to about 20 people.

My assumption is that in order to compare (subtle) differences in audio devices, it is important to have a high quality source. In my view that has to be a microphone signal, fed into a high-quality pre-amp. I doubt if you can get higher quality than this (assuming acoustical music reproduction).

The test will be double blind with a reasonably long learning period (within the 3-days of the test).
Some of the potential contributors don't like ABX tests so another (double blind) system has to be agreed upon. I still have to collect opinions about the preferred method.

I'm basically trying to find out if a test like this can be done at all. One of the problems I foresee is that audio professionals might not even be interested in finding out what's just good enough, but want to use the best equipment they can afford, even if that means overkill.

In the eyes of HA, how should this listening test ideally be set up ?

Thanks for your input.

Fiedler sets the minimum at 4.




Did you read the paper? Or do you imagine I'm actually advocating greater-than-redbook distribution parameters?

I'm not. 20 bits for *recording and production* has a rational basis, though.

Of course you are familiar with the arguments of Brother ... Oh, excuse, I though this was the angels and pin heads discussion.

ABX seems the only option if you expect a clear cut answer to "is there an audible difference ?".

There's ABC/HR, but you would have to make it ABX style (several trials for each test) to tell if they could hear an impairment or if it was mere luck.

They don't like ABX because they know damn well that if they do it, they won't be able to hear any difference at all, and because they can't hear a difference, they claim that ABXing somehow screws up your hearing/perception and makes you miss those "subtle differences".

That is why ABXing is meant for the experts! During MPEG standardization processes, a team of audio experts can hear the differences. You need to "train" your hearing in order to tell the differences- plain practical skills.

Even if you do train your hearing, you still can't magically hear frequencies higher than what physical limitations let you hear (not including insane tests with SPLs that are probably harmful to you)...

I'm not sure if you're being sarcastic or anything. If you're not, do you have any actual proof that higher sampling rates (for the SOURCE material, not oversampling anti-aliasing filters) actually make an audible difference? Especially from these so-called audio experts you speak of. I sure can't find any.

Congratulations. You are now qualified to write editorials for Stereophile magazine.

I don't really get it, because I don't know anything about that magazine.

Consider yourself lucky that you don't know. Hint: Imagine the opposite of ha.org.

Hello Kees de Visser,

If you want to run a very serious test about high definition digital formats, here are some things that I thought about.

Documentation.

You first have to learn about the similar tests that have already been done. Here is all that I have got in my links :

http://www.hydrogenaudio.org/forums/index....ndpost&p=372649

http://www.hydrogenaudio.org/forums/index....40&#entry374740

The first thing to notice is that this kind of test have already been done, and that it is very difficult to get any success.

Basic training.

Start with very low definition material. You can get some samples of low definition audio here : http://ff123.net/samples.html
Usually, it is hard to distinguish between 15 bits and 16 bits of resolution, and between a lowpass of 15 khz and a lowpass of 16 kHz !

Choice of sonic material

However, a sample was found where a 16 kHz lowpass was easily audible. I didn't keep the link but I'm sure that someone here have got it. Thus the samples used for the test are very important. The listener can fail to distinguish between the formats just because the presented material is not critical.

For the bit depth, you need very dynamic material. The "rach" sample in the above page contains no silence. I think that a critical sample should. I'm not used to classical music, but in the few CD that I've got, there is the recording of Grieg's Peer Gynt directed by Neeme Jarvi, (Deutsche Gramophon). The RMS level of the first instrument of the track "I Dovregubbens Hall" is -60 dB, while the end of the track is clipping ! This recording have a resolution of 14 bits, and the quantisation noise is very audible at the beginning of the track.

For the sample rate, since only Oohashi claims to have got significant results, why not include similar instruments ? The "gamelan" that was used is a set of metallic percussions (metallophons). James Johnston ( http://www.skepticforum.com/viewtopic.php?...der=asc&start=0 ) cites this class of instruments (bells, glockenspiel...) as very sensitive to phase shift, and capable of producing hypersonic intermodulation directly in air, regardless of recording.

Scope

What is the question that this test is to answer ? Are you going to compare sample rates ? Bit depths ? DSD ? Analog vs digital ? Online or recorded on CD ? On tape ? Commerical formats ?...

Result analysis

The classical statistic evaluation (ABX) doesn't seem to fit this kind of tests. The NHK test, the Detmold tests, Julian Dunn et al's tests about pressed CD all make the same mistake. They evaluate the individual statistical significance, then say "this result is significant, this one is not". This is mathematical nonsense ! Statistical significance can only be applied to one hypothesis, not to individual results.
There are other methods (Anova, Tukey...) more suited to listening tests with many people, but I don't think that they work well for this kind of test, because you don't want to know if the group can hear a difference, but if one people at least in the group can hear a difference.

I've got a solution for this kind of hypothesis, but it might require quite a lot of trials for everyone. I'll got it together in the next days. I set it up for my interconnect blind listening test, but the only written description of it is in a french forum, scattered in several posts, with some mistakes and corrections.

Basically, it just consists in letting people perform ABX tests (or similar tests), each one with its individual p value, and then compute the P value of the following event : "one people at least get an individual result equal or inferior to p by chance".
It has the advantage that listeners can communicate, and help each other, without affecting the significance of the final result. And it let the possibility for anyone to get a significant result, while usual statistical evaluations dismiss individual successes as non representative.

Protocol

The ABX protocol is not required in itself. You can choose A/B, AXY, XY, or anything you want, as long as it is randomized. If the randomization does not lead to a binary choice with equal probabilities, the individual p values will have to be recalculated from the right formula, since the one that is used for ABX won't be correct anymore. It is however correct for any protocol that leads to choose between two options of equal probabilities. For example "X=A or X=B", or "A was first or B was first", or also "X was the same as Y,or X was not the same as Y". Just be careful that the randomisation assign an equal probability for both answers.
The most secure way to get random numbers, in my opinion, is to use dices, with dice cups (like these : http://www.bgshop.com/ )

If listeners pass the test individually, each one can even choose the protocol that he prefers, as long as you get his individual p.

Hardware

Are you going to use active bi amplification, like Oohashi and the NHK ? If not, how to evaluate the role of intermodulation ? According to David Griesinger ( http://world.std.com/~griesngr/intermod.ppt ) hypersonic intermodulation mostly occurs in amplifiers, not in tweeters, while James Jhonston (link above) says that tweeters are very prone to hypersonic non-linear distortion. If you get hypersonic intermodulation, it will mean that the high definition format is inferior to the low definition one, since it adds distortion !
If you get active biamplification, how to evaluate the atmospheric hypersonic intermodulation ? If it is significant, then the test result will completely depend on the distance between the microphones and the instruments. Far from the instruments, no high definition required, since hypersonic intermodulation occurs in the performance room. Close to the instruments, and high definition is required in order to let hypersonic intermodulation occur in the listening room.
And can the difference between Griesingers' conclusion and Johnston's one come from the fact that the former tested harmonic intermodulation while the later speaks about transient non linear distortion ?



This is the scientific approach. It have become very hard because of the tests already set up. If you want to learn more, you will have to run better tests than what was already done !

However, you can also choose the debunking approach : find people who claim they can hear the night-and-day difference immediately without possible mistake. Let them use their equipment and usual listening conditions, and just add randomization. Be sure to allow the listener to give null answers, otherwise, in case of failure, he will argue that the stress confused his hearing. If he can answer nothing when he hears nothing, answering something will mean that he heard something, and not that stress prevented him to do so.

Erm, not entirely.

SPL Car installs regularly reach in excess of 150db of bass, and i am sure that my Missions could blow my eardrums if i whacked them right up - which is at least 119db

so your only 2/3 of the way towards 150db of dynamic range, and 15% short on the human ear at 119db.

Surely with all that optical storage available it would be good sense to overkill the digital capabilities of the format to dispell this argument completely?

Sorry for being anal, I kind of agree with you as well, but I'm playing devils advocate.

Or.....have I misunderstood something?

Dear Pio2001,

thanks for your elaborate post. This is the kind of constructive advice I was hoping to find here.
I'll need some time to digest all your info and do more reading about the subject. Work is very busy so please be patient

"this kind of test have already been done"
I know, and the results indicate that high(er) bitrates have no (or very little) benefits. This can mean that audible differences are negligible (or non-existant), or the test was flawed. If it turns out that we can't design a better test, it's useless to start one.

"Start with very low definition material."
That's exactly what we had in mind. Don't discourage people with impossible tasks.

"Choice of sonic material"
We're in the luxury position of having a large library of high-quality recordings and at the same time being able to record "live" music or sounds in the studio. Basically any instrument can be used. A choice of sources should be made and agreed upon before the actual listening test takes place.

"For the bit depth, you need very dynamic material."
That's available, ranging from the microphone/room self-noise to (just) acceptable clipping levels.

"only Oohashi claims to have got significant results"
I've been thinking about adding EEG (or even MEG) measurements but in practice this will probably be too complicated (MEG even impossible to install in a studio). Have EEG results, besides Oohashi, ever been used to prove audibility of stimuli ?

"What is the question that this test is to answer ?"
Good question! Actually I'm not sure wether the test should "prove" or "investigate" something. As a recording engineer I'm interested in capturing a microphone's output with as little loss as possible. Enduser formats are a different matter, though related, so this test should focus on the recording and mixing formats.

"The classical statistic evaluation (ABX) doesn't seem to fit this kind of tests."
I'm glad to find this open-minded attitude on this forum and I'm very interested to hear more about your proposed evaluation method.

"Are you going to use active bi amplification ?"
The studio is equipped with a large selection of active and passive monitors. What isn't available can probably be arranged. The large active Genelec 1035 monitors are tri-amp models, designed for high spl (136dB peak) but without "super" tweeters. Of course there are other, more subtle monitors as well. We would have to examine which monitors perform best with respect to hypersonic non-linear distortion.
I'm in doubt wether it's better to test under laboratory conditions or to use a real-life music studio setup with equipment that's actually available.
Thanks for the links to the Griesinger and Johnston theories. I'll read them and see what consequences they have for the proposed test.

"This is the scientific approach. It have become very hard because of the tests already set up. If you want to learn more, you will have to run better tests than what was already done !"
Again: if we think this test won't add anything meaningful to the existing ones, we simply won't even start.

"the debunking approach : find people who claim they can hear the night-and-day difference immediately without possible mistake."
This might be a good warming-up test to prepare for the real test. Is it safe to say that if the warming-up test fails (no significant results) it's useless to continue ?

Kees

EEG might be able to distinguish between hearing a sound and not hearing a sound, but there's no way it could tell you anything about hearing or not hearing particular frequency components of sounds*. Even MEG would not be any help unless by some chance the proposed "high frequency response" area of the brain is in a different physical location of the brain than the rest of our hearing circuits. These devices actually provide a surprisingly low amount of real information, despite the pretty graphs and pictures, and are as easy to misinterpret by a non-expert as the subtle details of the sounds you are studying.

I would be very suspicious of studies using these instruments as "proof" that we can hear or not hear particular sounds. Most of all I would want to know the credentials of the person analyzing the results, and whether this was a qualified neuroscientist or not.

*Edit for clatity: that is, broad frequency spectrum sounds. Does it really matter, from a audio reproduction standpoint, is there is some sort of physiological response in the brain to isolated high-frequency sounds? Until it can be proved to have an effect on our conscious hearing of the music I would say it does not.


I think that people here are not necessarily married to the ABX test in particular. But we do want to see a list of things common to all good science including: detailed explanation of the set-up and procedure, logical and well-thought-out methods, controlled tests, and not least repeatability. Plenty of scientific studies have been done double blind, but then later found to be invalid because of some overlooked flaw elsewhere in the experiment. It's easy to sink yourself in this game.
As Fenyman said: "The first principle is that you must not fool yourself -- and you are the easiest person to fool."

...


Heh, more off topic: I saw that and was totally struck by how much the ufo looks like a 1970's speaker cone. I wonder if that's what it actually is, suspended by a bit of fishing line? Anyways, it inspired me:

(link to slightly bigger version)

That's what I guessed, but I didn't want to dismiss any possibility to make the listening test as scientific as possible. My understanding of the brain and neuroscience is very limited but from what I understood MEG has a very high time-resolution (in the one digit ms region). Now imagine a stimulus containing sound (music?) with intermittent hypersonic content, on and off. If the hypersonic content has an influence on human perception, wouldn't there be a possibility that this would show on an MEG readout ? (I'm not even sure if loudspeakers/magnets are allowed in the MEG room) Ah well, you're probably right that it will turn into a sort of SETI project that hardly serves the purpose of the test.


Exactly, that's why it's a good idea to study the flaws from other tests and try to avoid them by carefully preparing the test with help of experts and open communities like HA.

The user was talking about typical high quality full range home speakers. Only extremely large line arrays or very large horns(for example, Avente Garde) can reach SPLs of 120+ dB at the listening position. CD should be good for about 120dB SPL in real situations[assuming the recording noisefloor is sufficient]; read my text further down in this post.



No need for this range, short of wanting to reproduce a live band in a small room at close range in a very quiet room [requiring very special high output speakers] to full SPL range, which I can not see a use for in a practical sense. Such levels would be painful/undesirable to almost anyone. And the reproduction of a live band in a small room would probably require a 130dB SPL ability at listening position for some peaks.

Another consideration is the masking noise floor of the room itself. An average room, considering HVAC noise, traffic noise, fans, etc., 40dB is doing fairly good. If you live in a secluded area like in a rural area, away from a busy road, turn off the HVAC, the neighbor is not cutting the grass, and any/all other noise making devices in the house are disabled, you might get into the 20dB range in this very rare scenario. Then there is the noise floor of the recording to consider. I have not yet come across a commercial recording [made from microphones, as opposed to purely synthetic] that had a noise floor hovering around the limit of the CD format itself[though I could see it happening if a digital noise reduction filter was used aggressively -- which would probably result in other problems], the noise is always substantially higher. Such may exist, but it must be very rare and limited to a special demo recording. The range would be such as to require the aforementioned high output speakers. In reality, the 96dB range seems to allow in the neighborhood of 120 dB of practical SPL, if considering the noise floor of real rooms. Beyond what almost any high quality speaker can provide. In a car, the noise floor will be far worse than the examples above, unless the car is stopped, in a quiet area[very little exterior noise] and with engine turned off.

-Chris

figured id missed something - thanks

No insult intended. But if you were to write an editorial there based on the idea that ABX testing
*itself* is the problem...you'd probably get a bagful of subscriber letters praising your insight,
(along with anecdotes how even their *wives* can hear the differences), and possibly a promotion
to a corner office.


Can you expand on this? Are you saying results from a test must always be analysed in aggregate of all subjects? How does this account for an 'exceptional' listener hearing a real difference , among a group of 'average' who fail to perform above chance?

You can't do that, because there is no way to tell a priori who the exceptional listeners are. They could just be the ones who by random chance got the "right" answers. You can't do that kind of selection on your data after the fact, and if you do you can "proove" just about anything.

I do a test of 100 people for psychic powers by having them guess the number/suit of a randomly drawn card from a standard deck, and three get the right card. I then say that these three are psychic and the rest are normals, and I have proven the existance of psi. I am wrong, because three of 100 is not much outside the normal statistical chance of 1/52 for this experiment. The three people were just lucky. This is a simplified example, but your idea is wrong for the same reasons.

Why do we need 24bit/48kHz/96kHz/192kHz if 16bit/44.1kHz is good enough? Are there any situations that 16bit/44.1kHz simply cannot satisfy? In other words, is there any real need for the higher bit depth and sampling rate?


There are such situations if you record (or produce) your own music. Any new mix decreases the S/N ratio as well as it increases the group delay of filters. Even if you
stay in the digital domain, you can experience these effects with a wave editor by using the "mix paste" function. The final product, however, is o.k. with the cdda standard.

An undisputed benefit of the industry's efforts is that the hardware codecs now finally live up to the standard of "good old" cd audio.

Purely listening purposes, you would be about right. Obviously there are purists who would like to listen to as high-resolution a recording as possible, whether or not the difference can be heard.

The strongest point of using 24bit/48kHz/96kHz/192kHz is for mixing and recording. Heavy editng of a 16-bit/44.1kHz recording will introduce audible artifacts. Not so with a higher bit-depth/sampling-rate.

Yes, but on several trials ABX testing (say over 16), getting all the right answers is unlikely. So I'd safely assume that one who did it, could have heard a difference. I wouldn't jump on conclusions yet, I'd rather make him/her take another set of tests. If the subject succeeds is because there's an audible difference, which could lead to the conclusion that at least 96/24 sounds different (and in theory better).

One verified pair of "golden ears" is all it takes for this silly debate to end.

Exactly one pair, more like. One out of twenty is meaningless.

In that case I'd just call 19 people and make them do the same tests as guruboolez and if they ever fail I should say that guruboolez' results are meaningless.

The probability of someone guessing 16 trials is 1/65536. If you have 20 subjects you increase this probability merely to 20/65536 (0.000305). If one succeeded there's no reason to believe that just because he was on a group his result should be discarded.

You can retest the putative 'exceptional' person. If they keep scoring well above chance, don't you think that means something?



Again, retesting should return these people to 'normal', if they were just lucky during *that test*.

And what if you are testing one or more persons who *already claim to have psychic powers*? This is analogous to the normal situation in the audio hobby (if not in codec testing). The 'sighted' portion of the test usually involves fiding if out the subject think they hear a difference in the first place. If not, there's no reason to continue.

I think that using dynamic recordings in dedicated listening rooms, you may be able to show the benefit of bit depths superior to 16 bits.
You may also try to generate artificial signals that would show the theoretical audibility of a given parameter, even if you fail to find a musical recording that suffers from this parameter. For example I recently tried to ABX a phase shift at 30 Hz. I chose a recording with 30 Hz notes with sharp attacks. I failed. But I was told that the ideal signal for this test was a low frequency "saw-teeth" signal.
The main problem is that strong high frequencies can damage tweeters.



I don't know, but I was not talking about the EEG results, that are also questionable because the EEG excitation started one minute after the stimulus was presented, and also ceased quite a lot of time after the stimulus have been removed
I was talking about the subjective appreciation of the sound by the listeners. Table 2 gives an impressive set of significant p values associated with direct listening test, not through EEG. However, not a word about the way they were computed.

If you want to try the same experiment, that is asking people if what they hear sound "harsh, dynamic etc" instead of asking them to identify X, then we would have to setup a mathematical model in order to get the statistical significance of the answers.



For this kind of test, where we want to see if a difference can be heard by some people under some conditions at least, I suggest a protocol divided into three parts :

In part 1, the listeners, that are supposed to be familiar with the kind of difference tested, are allowed to play with the system. They must find the hardawre and the musical samples on which the difference is the easiest to spot. This phase goes on until they think that the difference is obvious enough for a blind test to easily succeed.
In part 2, some fake blind tests are done. This is the training. Listeners try to recognize the difference under the real test conditions. They can compare ABX with other methods. They can choose what seems to be the best delay between the trials. This part ends when the listeners, or at least some of them, consistently get statistically good results. Remember that this is only training. These results won't be taken into account in the final conclusion, no matter what happens.
In part three, the real test is done, according to the protocol chosen in part 2. If the number of trials was decided in advance, listeners are told their score after each trial. If they begin to make some mistakes, they can interrupt part 3 in order to undergo some more training, or stop for a while. In part 3, they are allowed to give null answers when they are not sure. In ABX, it would be a three choice test : "X is A", or "X is B", or "I'm not completely sure".
Only the X is A or X is B answers are recorded. The part 3 goes on until the right amount of these kind of answers is collected.

The advantage of dividing the test in three parts is to dismiss the usual arguments opposed to blind tests :
The listeners are deaf : dismissed by part 1
The system is not good enough for the difference to be heard : dismissed by part 1
Listening in ABX doesn't allow to spot these kind of differences : dismissed by part 2
A decision process cannot account for the unconcious influences at work : dismissed by part 2

If one listener decides to do an ABX test in 8 trials, here is an example of phase 3 :

Trial 1 : X is A : right
Trial 2 : X is A : right
Trial 3 : X is A : right
Trial 4 : I'm not completely sure
Trial 5 : I'm not completely sure

Pause

Trial 6 : X is A : wrong

Training

Trial 7 : X is B : right
Trial 8 : X is A : right
Trial 9 : I'm not completely sure

Pause

Trial 10 : I'm not completely sure
Trial 11 : X is A : wrong

This is the second error, the test has failed. Otherwise, it would have gone on until one more "X is A" or "X is B" answer would have been got, which would have totalized 8 answers of this kind.

If more than one listener is taking part, the required number of right answers must be mathematically decided. We must compute the probability for one listener to fail its own ABX test by chance. Then put it to the power N, when N is the number of listeners. It gives the probability that everyone fails. The complementary event is that one listener at least have succeeded.
This is our final statistical result : the probability that among all the listeners, one of them at least gets by chance the same or more than the highest individual score recorded.

All the listeners can pass the test together, if they want. Uncontrolled influences between them can only decrease the probability of this event, thus increase the statistical significance of the result.

Advantages of this kind of statistical evaluation over a classical one :
-Listeners who cannot hear the difference don't prevent listeners who can hear it from demonstrating that the difference is audible
-Listeners can communicate and help each other during the test. They don't need to pass it one by one.

Drawback :
-More trials are needed in order to reach an acceptable level of confidence.

It is very probable, in case of a difference that cannot be heard at all, that the test doesn't get past part 2. The listeners must then explain why the differences heard in part 1 have vanished in part 2, and possibly get back in part 1 in order to find a better way to pass part 2. It's up to them. They are the one hearing a difference, they are the one who can tell how the test must be done.

This protocol was discussed here, in french, during the setup of the interconnect blind listening test : http://www.homecinema-fr.com/forum/viewtop...r=asc&start=195

I resample to 24/96, because if I just played files at 16/44,1, Audigy would just make its own crappy resampling. So I do the resampling with software and add some extra just because I have a reasonably fast cpu.

CD's are encoded in 44.1khz, and most soundcards upsample them to 48khz (because they cant play at 44.1), and this actually degrades quality... not improving.

It's entirely possible I'm posting on the far end of a thread like this because I'm new, but having read through all six pages, there's something that strikes me :

The prevailing opinion seems to be that an arbitrary format (16/44.1) is in itself surplus to requirements. So the next format (24/96) is even more surplus to requirements, except in highly specialised environments like studios. Whether it's better or not is questionable, but is it actually worse? If not, then what's the problem? Now, the tangent while I try and structure my thoughts :

I'm a PC modder, and I'll do all sorts of things to get my PC to run faster and cooler and with more lights in it. I have benchmarks and tests that will show just how much faster my PC is than other people's. Numbers! Never mind that it's physically impossible to tell, from the user's point of view, whether you're going at 29.4 fps or 31.2 fps in a game. But if I was to sell you a graphics card I'd gloss over that, give you the numbers, and you would decide for yourself that the one that goes faster is the better one. Apart from cost, what real difference is it going to make to your life? Only one : it will make you feel better. Your experience of playing a game will be better. It's subjective, but you'll have more fun knowing that you're not losing out to the limitations of your own optical nerves, which can only detect rates up to 25fps or thereabouts. So : should progress of graphics technology be stopped, because it's "good enough"? Of course not.

Everything goes faster than it did, especially electronics. Once you reach a certain point, you're not going to notice much difference, but it will be there. You wouldn't notice much difference between a car trip in this year's Rolls Royce and last year's. Given the choice, would you go for the old one? Especially - and here's a fun bit - if it cost the same? Technological advancements are very rarely driven by what the consumer wants, but what he can be told and convinced that he wants. What he can be told to buy. You convince enough people to buy, it becomes the standard and then boom, there's no difference to argue about any more and it all costs the same anyway.

We can, so we do. I certainly couldn't tell the difference between 24/96 and 16/44.1, but that doesn't mean I'm going to be feeding my 24/96-capable amp with half of what it can chew on. I'm also not going to run out and get DVD-A replacements for all my CDs. Is the difference there? Yes, mathematically and by the oscillations of the crystals, there is a difference. Is it perceptible? No. Is there a difference in the experience of owning and operating one of these things? Yes. And that's what matters. Limited edition CDs sound better for just this reason. That, for me, is an integral part of the listening experience, and if I wasn't reasonably scientifically-minded and sceptical already, I'd hate for someone to take that part of the experience away from me.

I'm sorry, an empty bank-account caused by all the surplus on equipment and media I have to purchase for something I can't hear won't make me feel better. And I mean this seriously, I don't want to spend money on illusions just to make me feel better. I don't want to spend money just because company X invented a format which brings me nothing more than things there already are.

Give the money to charity, it will make you feel better too and it at least might end up somewhere where it is needed.