The myth of high-resolution audio - busted!

[Ethan Winer]

I still see people claim that 24/96 (or higher) audio is "better" than 16/44 used on CDs. Apparently these people missed a groundbreaking article in the AES journal late last year that put this myth to rest for good. I can't post the article itself because the AES sells it on their site. But these links discuss the article and explain it in depth:

 

Audio Critic summary of the AES hi-res fallacy article:

http://theaudiocritic.com/blog/index.php?op=ViewArticle&articleId=41&blogId=1

 

More from the authors themselves:

http://www.bostonaudiosociety.org/explanation.htm

 

Paul Lehrman commenting in Mix magazine:

http://mixonline.com/recording/mixing/audio_emperors_new_sampling/index1.html

 

Nobody can hear, or perceive, or be influenced by ultrasonic content, even if they think they can. Here's my best explanation for why people sometimes report hearing differences even when none can possibly exist:

 

http://www.ethanwiner.com/believe.html

 

--Ethan

[wwwjd]

HA! I do believe it. This should be a fun "discussion"

[Rudolf von Hagenwil]

So they are testing and discussing 24-bit productions on the example of a sample rate and bitrate down converted 16-bit 44.1 Khz Compact Disc, possibly on the example of the band Bloodhound Gang and they don't hear a difference or what?

[UstadKhanAli]

I'm glad that they specify a difference during the recording process, as I could hear a noticeable difference when switching from recording numerous tracks in 16-bit to 24-bit when recording on the same converters (all other things being the same).

 

http://theaudiocritic.com/blog/index.php?op=ViewArticle&articleId=41&blogId=1

 

It should also be pointed out that more bits and a higher sampling rate in recording are still a good thing because they permit a little bit of unavoidable sloppiness, so that you can still comfortably end up with 16-bit dynamics and 20 kHz bandwidth.

 

Anyway, very interesting. The fur should surely fly in this thread...

[blue2blue]

Did everyone actually read the materials linked?

 

Because if they had, they might not have made some of their comments, I suspect.

 

The tests were of delivery formats -- not production formats. And the material selected was typically in the range of best case scenaria with regard to the relative quality of the product -- these were not crappily recorded productions, by and large, but rather top flight recordings that push the format more or less to its limits.

 

As the write-ups linked above suggest, the conclusion is that 16 bit/44.1, at or near its best, was indistinguishable from the SACD representation over a bed of test subjects and conditions.

 

And the tests appear consistent with accepted theory (specifically Shannon-Nyquist).

 

I would commend Dan Lavry's white paper on sampling to those who remain dubious about the capabilities of 16/44.1.

 

 

Also with regard to the notion that you can't have "too much resolution" -- Lavry argues forcefully that you can -- at least as it relates to contemporarly technology implementation.

 

As anyone conversant in AD technology probably already knows, the measurement process at the hard of PCM encoding is a rapid series of comparisons of the signal voltage against known voltages.

 

This process, while very fast, still takes finite amount of time for each comparison and the number of comparisons allowed by the time devoted will, over the course of a multi-sample signal, correlate with measurement accuracy. If a very high sample rate does not allow sufficient time for an accurate measure -- higher sample rates may actually decrease accuracy.

 

FWIW, Lavry, Bob Katz, and others have suggested an optimal sample rate for audio would be in the 60-70 kHz range, as that would allow plenty of frequency bandwidth for an easy-to-design/build gradual roll-off anti-alias filter as well as capture of frequencies well above the highest frequencies ever tested as being audible. (No one, has ever been tested as being able to discern 25 kHz, for instance, though some (very few) very young people have tested as high as 22kHz or so, IIRC. Maybe Ethan has good figures on that.)

 

 

Also, it's well worth looking at Ethan's paper on the broad differences that small changes of monitoring position can make in real world monitoring environments. It's a real eye-opener.

[Rudolf von Hagenwil]

Thanks for reading the stuff...

 

 

... to those who remain dubious about the capabilities of 16/44.1.

 

 

I haven't heard much 16-bit since we record everything in 24-bit since 1994. However, we hear the difference between 16-bit and 24-bit very clearly.

 

To the folks who still record in 16-bit, the 16-bit signal is rubbish after one signal rendering in the digital domain, where in 24-bit you can render several efffect until the signal has a detectable degradation.

 

About the folks who don't hear a difference between 16-bit and 24-bit, well I guess they need a hearing aid, possibly also eyeglasses in case they don't see the difference in resolution between a 1080i and a SD video signal.

 

.

[blue2blue]

I think you either have not read the material linked -- or you substantially misunderstand it.

[Anderton]

The paper conceded that recording with 24 bits vs. 16 bits made a difference, but that it didn't make a difference in terms of playback.

[Anderton]

But check this out. I am not a fan of audio snake oil, and have always felt, for example, that the idea of identically-copied CDs sounding "different" was part of the lunatic fringe.

 

However, I recently mastered a CD. The artist wanted me to reverse the order of two of the cuts. I called up the CD Architect file I'd used originally, reversed the cuts, and burned another set of test CDs for the artist, producer, and studio owner. They thought the first CD sounded "better." I found that hard to believe, so I suggested that one of the three shuffle the CDs, and the other two write down which CD they thought was which as it played back. Then one of them shuffled for the third person.

 

Of the three, the artist got it right 7 out of 7 times. The producer and studio owner each got it right 6 out of 7 times. That's statistically very significant.

 

I took the two CDs back home and did a file comparison in Wavelab. It said the files were virtually identical, except for differences in the dithering noise distribution (as to be expected; dithering is a random process), and of course, this signal was way down in level. I then brought the cuts into Sonar and did a phase flip - perfect cancellation (except again for the low level noise).

 

I could hear no difference between the two files in my studio, BUT they were both going through the same D/A converter. There was no CD player in there to introduce a variable.

 

I emailed Bob Katz, described the situation, and asked in effect "What the hell is going on?" He said that he and others have noticed this effect as well, although not as much lately, and mentioned some possibilities involving jitter but that he was also at a loss to explain what the deal was in my particular situation.

 

Fact: Wavelab found no significant difference between the files.

Fact: Three sets of ears were able to identify which CD was which when they had no idea which CD was being played, with close to infallible results...statistically, way beyond what would be obtained by guessing.

 

All I can think of is that somehow, the two CDs somehow interacted differently with the CD player, although they tried the same thing with another CD player and could also identify the difference. The ONLY difference I can think of between the two CDs was that they were at burned at different burn rates, but why should that make a difference? And why would the CD burned at a slower speed be considered the one that didn't sound as good?

 

Color me perplexed.

[Anderton]

As to DSD vs. PCM, I do like the "character" of DSD better. There may not be a frequency response difference per se, but I think the key there might be what I'm listening to with the PCM audio: I don't have any super high end converters here. I think it's entirely possible that inexpensive conversion/filtering on PCM will simply not sound as a good as inexpensive conversion/filtering with DSD, which is far more tolerant given the ultra-high clock frequency. Perhaps if I had top of the line PCM converters and filters here, I wouldn't consider DSD as sounding better.

 

But for a given price that's not in the audiophile range, I think it makes complete sense that DSD sounds better than PCM. The technology lets you get away with a lot more.

 

As to 44.1 vs 96, again, I think it might be a question of gear quality. If there's ANY funkiness in the 44.1 filter rolloffs, that will be in the audible range whereas the ones for 96kHz won't be. But maybe there's another issue here: What about the ears of the people doing the listening? Seems to me that if you REALLY wanted accurate results, you'd need to use kids under the age of 12 whose high-frequency response hasn't been compromised. When most adults' hearing starts dropping off radically at 12kHz or so, they won't hear differences that are happening at 18-20kHz. Right?

 

Not that any of this matters. When someone asks "how can I make my music sound better," my response is often "write a catchier bridge."

[blue2blue]

Craig, there's a reason they do double blind testing in measuring subjective evaluations.

 

Because doing single blind testing was found to introduce signficant bias error. This is well supported by available science and there is nothing to contradict it. This is why anything less than double blind testing of subjective material is considered inherently compromised.

 

 

If a phase inversion sum produced no significant (ie, audible) differential, I'd say that strongly suggests that one of your testing methodologies was compromised.

 

 

Let's apply Occam's razor: you did single a bed of 7 single blind subjective tests with what can only be assumed to be a highly informal rigor. Your subjects were able to identify one element of the test bed from another with high regularity.

 

But an objective test -- which all available science as well as common sense suggests should be accurate -- indicated there should be no audible difference...

 

This is not hard to weigh out.

 

 

I am not suggesting science has all the final answers here. I am merely suggesting that rigorously arrived at science has the best answer supported by the best evidence we can collect at any given time

[Dean Roddey]

 

well I guess they need a hearing aid, possibly also eyeglasses in case they don't see the difference in resolution between a 1080i and a SD video signal.

 

 

That's not really a valid comparison. Our visual system is enormously higher bandwidth than our hearing. Our ears aren't that great. Our eyes aren't that great relative to some other animal's but they are far more discriminating than our ears.

 

44Khz was chosen because it's known to be able to represent the highest frequencies we can hear. High def video's resolution was chosen because it was a practical compromise, not because it was believed to be at the limits our of resolution capabilities. It's still about half of what would be required to be the visual equivalent of CD audio, at the recommended subtended angle of the screen and viewing distance. So SD and HD video are probably the equivalent of an 92K MP3 and a 240K MP3, not anything in the realm of a CD to SACD type of comparison.

[Anderton]

>

 

Well, if you shuffle CDs randomly and ask people to pick out which is which, and they do so with almost 100% accuracy, that's hard to argue with. They didn't know which CD they were listening to yet they were still able to identify them.

 

All that Wavelab told me is that the digital data in the files was close to identical. It didn't tell me anything about reflectivity in the CD, possible differences in media, whether burning at a higher speed caused the error correction to react differently, etc. In other words, there are a ton of variables between the time the file gets burned on to the CD and the time it reaches people's ears. I simply can't discount the ability of three people to hear a difference when they had no idea which CD they were listening to. To me, that's as significant as what Wavelab reported. I think it more likely they were hearing something that we simply haven't quantified yet, rather than assuming they were all delusional, or perhaps communicating the identity of the CD telepathically...although I suppose that's an option, too

[Dean Roddey]

Testing has been done on error rates in commercial CDs and they are actually quite low. There's error correction on CDs and the number that cannot be corrected are very low as long as the CD isn't physically defective. Don't know about home burned CR-Rs, but I don't think that error rates on commercial CDs would have anything to do with any perceived differences.

[blue2blue]

> Well, if you shuffle CDs randomly and ask people to pick out which is which, and they do so with almost 100% accuracy, that's hard to argue with. They didn't know which CD they were listening to yet they were still able to identify them. All that Wavelab told me is that the digital data in the files was close to identical. It didn't tell me anything about reflectivity in the CD, possible differences in media, whether burning at a higher speed caused the error correction to react differently, etc. In other words, there are a ton of variables between the time the file gets burned on to the CD and the time it reaches people's ears. I simply can't discount the ability of three people to hear a difference when they had no idea which CD they were listening to. To me, that's as significant as what Wavelab reported. I think it more likely they were hearing something that we simply haven't quantified yet, rather than assuming they were all delusional, or perhaps communicating the identity of the CD telepathically...although I suppose that's an option, too

 

I think I missed the shuffle bit. Sorry. And, of course, Occam's razor would cut against the telepathy explanation, for the most part.

 

 

CD playback error is something one would have to consider, given the givens. And one would have to give a thorough review of the other details of your test, as well, before one rushed to revise the body of accepted science on this, it seems to me.

 

As persuasive as you make these 7 tests by three individuals on these two CDs seem, I think there's a large body of other work (that does have rigorous methodology behind it) that must be considered and reconciled before we do any serious paradigm shifting.

[Rudolf von Hagenwil]

 

I think you either have not read the material linked -- or you substantially misunderstand it.

 

 

I read what Ethan said: "I still see people claim that 24/96 (or higher) audio is "better" than 16/44 used on CDs. Apparently these people missed a groundbreaking article in the AES journal late last year that put this myth to rest for good."

 

Of course this statement is false.

[blue2blue]

OK... that statement would be rash and unsupportable -- except that he gives himself some wiggle room with the qualifier apparently -- which implies a subjective point of view.

 

One cannot reasonably argue with how someone says something appears to them.

 

Right?

 

 

But I would allow that that would be an injudicious observation, to be sure.

[dekkersj]

 

As to DSD vs. PCM, I do like the "character" of DSD better. There may not be a frequency response difference per se, but I think the key there might be what I'm listening to with the PCM audio: I don't have any super high end converters here. I think it's entirely possible that inexpensive conversion/filtering on PCM will simply not sound as a good as inexpensive conversion/filtering with DSD, which is far more tolerant given the ultra-high clock frequency. Perhaps if I had top of the line PCM converters and filters here, I wouldn't consider DSD as sounding better. But for a given price that's not in the audiophile range, I think it makes complete sense that DSD sounds better than PCM. The technology lets you get away with a lot more.

 

It is clear that you don't have the slightest idea what DSD is about. It is worse than PCM in a technical sense. PCM is nowadays so cheap to make that there is no need for DSD anymore.

 

Besides that, for a delivery format there is no need for a format that has more resolution than 44k1/16bits. Those 16 bits are always correctly dithered from 24 bits sources so there is no loss to be expected >90dBFS or so.

 

(anyone who claims that he can hear the difference between 24 and 16 bits is fooling himself)

 

Regards,

Jacco

[Rudolf von Hagenwil]

 

OK... that statement would be rash and unsupportable -- except that he gives himself some wiggle room with the qualifier apparently -- which implies a subjective point of view. One cannot reasonably argue with how someone says something appears to them. Right? But I would allow that that would be an injudicious observation, to be sure.

 

 

That's Right! I actually don't know what Ethan meant but took his statement as is

[Rudolf von Hagenwil]

 

) Regards, Jacco

 

 

Here we have it again. I have no idea why dekkersj says that, could be that he works with playback devices on which he simply can't hear the diffrence.

 

I deliver 24-bit tracks for mixing, for example to Bruce and he tells me which tracks are 24-bit recordings without knowing which tracks are real 24-bit recordings and which one are upsampled from 16-bit.

[Lee Flier]

It is clear that you don't have the slightest idea what DSD is about. It is worse than PCM in a technical sense. PCM is nowadays so cheap to make that there is no need for DSD anymore. Besides that, for a delivery format there is no need for a format that has more resolution than 44k1/16bits. Those 16 bits are always correctly dithered from 24 bits sources so there is no loss to be expected >90dBFS or so. (anyone who claims that he can hear the difference between 24 and 16 bits is fooling himself)

 

Well all righty then. Because you say so, it must be true, eh?

 

Uhh... wait, who are you again? And how exactly can you tell what other people are hearing? You must be God or somethin'.

[blue2blue]

I was just going to say that rash and intemperate statements seem to be going around.

 

Again, if anyone cares to read the materials Ethan linked to, it's clear they are talking about delivery formats of finished recordings. They are not talking about AD; they are not talking about production process files or formats. They are strictly comparing presumed best case, optimally formatted and packaged program material in competing formats and, over a run of tests.

[dekkersj]

 

 

Nonsense. If you would understand how properly dithered material relates to it's original counterpart, you wouldn't make this kind of statements. In the audible spectrum there is no difference to be found.

 

Regards,

Jacco

[blue2blue]

[Rudolf von Hagenwil]

Okay, I make a dithering of a Berliner Philharmoniker 24-bit recording and will analyze the recording and the dithered 16-bit with a Rhode & Schwartz and post the result