http://www.sfgate.com/cgi-bin/article.cgi?...L&type=tech

A few highlights:

"It could be that MP3s actually reach the receptors in our brains in entirely different ways than analog phonograph records. The difference could be as fundamental as which brain hemisphere the music engages."

"Most of today's pop records are already compressed before they leave the studio in the first place..."

"How much the audio quality is affected by the MP3 process depends on the compression strategy, the encoder used, the playback equipment, computer speed and many other steps along the way. Experts agree, however, that the audio quality of most MP3s is somewhere around FM radio. The best digital audio, even with increased sampling rates and higher bit rates, still falls short of the natural quality of now-obsolete analog tape recording."

How do these people get/keep their jobs?

This quote (aside from making me laugh at how little sense it makes) instantly reminded me of this Koss KSC75 mod:

Yet another misunderstanding of the difference between compression and encoding.

ughh, and we're supposed to be so techy here in Silicon Valley! :|

Joel Selvin may turn out to be an audiophile!

Because their employers haven't a clue either. No different really than big news papers that print fabricated stories. "Feed them b.s., and they'll gladly and unknowingly eat it everyday"

Uggh, I saw this article on slashdot and I couldn't even finish half of it before I closed it in disgust. I bet this "Phil Ramone" couldn't ABX a 128kbps AAC if his life depended on it.

Against what? If it was against the CD, most of the time I (& I suspect many others) wouldn't be able to tell the difference either. Or did you mean he's never heard of ABXing, let alone done one?

He cannot be able find the difference between a CD and LameMP3 -cbr 320 -js -q2 !!! If he does, I will pay 1000$ for him :-)))

i doubt we'd have to reach -b 320. I'm pretty sure he'll trip somewhere near -V2.

I can't believe this trash made the front page of Slashdot. The comments there are quite disgusting, too.

Uh, /. is trash. Quite enjoyable trash, yes. But still trash. Men read FHM, women read tabloids, nerds read slashdot.

I love how analog-worshipers conveniently neglect the noise and distortion fundamental to the analog recording process. They'd have you believe that the signal on an LP record or tape is closer to the "real" sound waves from a performance simply because they're not sampled.

Just one problem with that - vinyl, for example, isn't a perfectly smooth, linear medium - it's actually rather grainy. Audiophiles make it sound like the grooves on an LP or tape are exact, linear replicas of sound waves. If in reality, the stylus is tracing over thousands of microscopic, randomly-placed grains and pits per cm and not a perfectly smooth, continuous groove (ignoring dust, of course), that seems like a primitive (if unintended) version of sampling to me. So much for a "pure" signal......and that's to say nothing of the other butchering a signal must endure to be stored on an LP or tape.

My brain will stick with digital, thanks.

Not only do nerds read Slashdot, they continue to do so a full five years after they claim it sold out and replaced competent editing with monkeys. It's trashed competitors, though; kuro5hin seems completely abandoned. I can't believe the site is still operating with the content on its front page. The only general geek site with potential for competent audio coverage I can think of is Ars Technica, and even that's gone the way of "PC Magazine".

The aspect vinyl lovers gloss over at all costs is the medium's ~60 dB signal-to-noise ratio, while, hypocritically, criticizing compact disc's ~100 dB SNR. It's such a weird argument, especially when even 60 dB is excessive, unless you like bloody ears or live in a sealed chamber.

They also neglect to mention that those wonderful 90 kHz harmonics that make vinyl such a great medium are either obliterated the second the needle first touches the platter or denature into random noise far outside our perception.

The whole concept is weird to me. While I enjoy the charms of old equipment, I'd be out of my mind to suggest a Commodore 64 is even within a light year of a modern PC.

^ think of what that FLAC could do with compression eh?

See it positive. More and more I come, when it's about audio signaling, lossy compression, mastering etc... to the conclusion: "I am an expert. I am a super expert. Only a few developers are above me. 0.0000001% of earths population. 99.99999999% is FAR below my knowledge"

@work, IT department, the other day:

me: "I have a notebook set up, which plays files... lossy, aac and mp3.. and wavpack lossless..."
him[1]: "Umm..." (leaves soon)

--
[1] he's supposed to be the depts. audio expert, because he trades with audio gear sometimes...

I've dealt with an IT person at work who had to look up .flac, and didn't want me to install WinAmp, because "it comes with spyware". Good thing I didn't tell him I was going to install Foobar2000, that would have broken his brain.

I've seen this misconception in various other articles. Naturally, one would expect a certain degree of confusion between dynamics compression and data compression, but that's something that can be easily resolved with a bit of explanation. Such a flawed statement in a published article seems unacceptable, and this article has many such flawed statements.

I've done more research for forum posts than this guy seems to have done for his e-news article. It's somewhat upsetting, but I don't imagine it really hurts anything. People probably don't pick up the SF Chronicle for the latest bleeding-edge advice on "MP3 music".

And hello, by the way.

That one is screamingly funny, indeed.

Welcome to world of pseudo-science. Make up anything you please and announce that it "could be" so. Digital encoding meet theosophy; astrology meet the LAME software; physics meet "psychotherapy". All shake hands, please.

Maybe Richard Dawkins could look into this one when he's finished debunking mediums for Britain's Channel 4:

But for heaven's sake don't let anyone undertake a blind ABX test.

Not to start a forum war or anything (that would be terrible), but I always find it interesting when two different groups of people talk about the same thing. Like when you go see a movie with one group of friends and then talk about it with another, sometimes the opinions differ.

It's interesting to see what other audiophiles are saying about this.

I should show this to my non-audiophile friends and see what they think. I want them to evaluate it normally though, I hope they don't get misconceptions...

I'll never cease to be amazed at the blind way people follow an idea without the need for any kind of proof simply because that idea fits there agenda / pre-conceived bias. Such is the way of world I guess.

I've always considered calling audio formats, such as MP3, "compressors" to be a misuse of the word. Compression means reduction of volume without loss of mass. MP3 et al might be more appropriately referred to as "reduced data" formats.

Calling MP3 a "compressor" might well have been a marketing decision...

That's an interesting point, actually. I suppose "reduced data encoder" (or RDE, if you fancy) would be technically preferable to "compressor", but perhaps the meaning has evolved beyond its traditionally pure form. Lossy image formats and lossy video formats have been using the term "compression" for ages, even if it may seem technically inappropriate given the currently accepted definition of "data compression".

Again, though: interesting point.

I suggest we all start using the word "encoding" (and variants) instead of "compressing" (and variants.) So that will help avoid confusion among the "light-minded" people out there.

And yet another nonsensical article about music and compression algorithms, backed by known experts as Phil Ramone, who, if the Wikipedia article is right, was one the first producers to release music in CD, and has nothing about producing modern artists, and great scientists. One said that you need a great mind to fight against the placebo effect of MP3s, another one throws in there old and new music, as if it's related to the discussion in some way. Finally, the article contradicts itself saying that MP3 is the same quality of FM radio and a bit later, that MP3 has great quality.

To finish everything, we discover that as FLAC users we devote ourselves to Grateful Dead and live the hippie way of life, and Josh Coalson is stealing from us lots of MB

Throw this in the recycle bin along that famous bit of the Wikipedia article about the MP3.

I never thought of it before this thread, but I've always called it "encoding", to my knowledge. Or plain "Make it into MP3".

I think some people here are hastily dismissing elements of the argument presented in the article - especially the argument that the brain has to sometimes work harder on the products of some lossy codecs. The article, for all its weaknesses, does legitimately discuss the opinions of researchers who are interested in this problem as trained experimental psychologists. But my impression here is that any attempt to reconcile audiophile debates with what we know about how auditory perception works is met with accusations of pseudoscience and/or a defense of a crazy analog agenda. Before you know it, the experimental psychologists will have you buying $2000 speaker cables.



But ABX tests might not be the ultimate paradigm to resolve the debate - there are other related types of response-based perceptual tests that tap into unconscious information processing - such as tasks requiring a motor output (response time tests). These sorts of tests often reveal processing differences that ABX verbal-report tests do not show. Of course there is a long thread about this here, and is worth checking out. On the second page I launch into a long series of posts trying to address this. I am not an audiophile; however, I do recognize that this long-standing debate is likely similar to many debates in the history of science -- both sides are partially right. The issue is an empirical one, and it will be settled one day.

Someone asked how the people described in the article "keep their jobs" -- one of the researchers is Daniel Levitan, and he's unquestionably one of the world's experts in the cognitive neuroscience of music. This man's opinion should not, in any way, be dismissed as pseudo scientific wackiness.

If mp3 works (sometimes it does) and only make changes to the sound that are below the level which can actually be transduced (from air-borne vibration to neural impulse) within a human ear, then the brain cannot be working any more or less hard than it would be already, because it's receiving the exact same neural impulses.

If the changes that encoding makes are above the level which is transduced within a human ear, then the brain is receiving the information. You can be damn sure (unless it's throwing it away in subsequent processing, or has no processing to recover that particular details) that this positive detection can be proven in some properly designed test. Whether it's ABX, reaction time, comprehension, annoyance etc is irrelevant - an appropriate double blind test can be designed.


It's not quite so clear cut. Ears vary, and the detection process is stochastic, rather than deterministic. The brain is great at pulling information out of noisy data from (on some levels) crappy sensors (e.g. eyes and ears). However, there is a level at which changes simply can't make it to the brain because of the limits of our ears themselves.

Of course, if you want to believe in people detecting undetectable differences which can't be proven, fine. That's religion, not science. (Not to bash religion, but in audio we deal with real measurable signals - what goes down the wire is what makes the speaker cone move, not magic!)

Cheers,
David.

Wasn't JPEG compression long in use before then?

Hi David



Agreed. Just don't equate transduction with perception - in both audition and vision, there is information being processed that individuals are not consciously aware of, but could potentially affect their perceptions, and will not be measurable by just asking them. For example, effort and fatigue vary as a function of flicker rate in reading tasks on CRT monitors, but at the same time subjects do not report a subjective awareness of any change in the stimuli, or notice, for example, that they are changing the number eye movements they are making (Bridgeman and Montegut, 1993). Is the flicker rate "above threshold"? Not by your definition, but it's affecting their processing, and in long-term preference judgments, I bet they would prefer monitors with a flicker rate that afforded the easiest reading condition (but not know why).



I also agree with this generally - however, many people's beliefs on this matter (e.g., perception of lossy compressed versus uncompressed sounds) are shaped by the results of ABX tests with verbal report. With a different methodology, one might well get different results, so I think which test one uses is quite relevant. For example, if subjects were unable to discriminate between uncompressed and compressed samples in an ABX paradigm, but a reaction time experiment showed that one took longer to process, and/or subjects systematically preferred* the one that was uncompressed, this would reveal an inadequacy of the ABX paradigm for measuring discrimination, wouldn't it? This could quite easily be the case - in fact results like this have already been shown in vision and audition.



Right - but what exactly gets through the initial stages of the auditory system is an empirical question - one that neuroscientists like Levitin are attempting to answer. The consensus here seems to be this is understood, but it's not. It simply is not known how lossy compression schemes change the way a sound is processed relative to its uncompressed counterpart.



This is what I was addressing in my post above. It is a mistake to attribute some magical thinking to these researchers, or myself (this is not to say that many analog purists DON'T engage in delusion and magical thinking, of course they do). The whole point I'm making is that there are alternative ways to measure people's perception that address counter-intuitive organizational principles in auditory and visual perception. These tools, developed by cognitive scientists decades ago, are relevant to current debates in audio. How is this religious? Again, the prevailing assumption is that ABX with verbal report is the test par excellence, but that might be false. Even a simple reaction time experiment could easily show that samples deemed perceptually equal by ABX tests are in fact processed differently. That's all I'm saying.


* ABX methodologies don't generally use preference as a proxy for discrimination, but that is often a good way to do it. Personally, I do it when I test people's Vodka discrimination abilities.

If you can discriminate, you can point to differences, so you can ABX. can't you? ABC-HR is what lets you rank. ABX is just to find out if a difference is discernable.

I wasn't referring to the researchers, but rather Joel Selvin and other "journalists" like him who write such poorly-constructed articles, drawing their own flawed conclusions, e.g. connected with the slimmest of threads to a statement made by a researcher.

I am saddened to see someone like John Meyer make such unsubstantiated claims as Meyer makes some very good loudspeakers - I've mixed through them myself on a few occasions - and they're rightfully highly-regarded in the sound-reinforcement industry. It's unfortunate that someone with such a solid engineering background would blindly buy into the belief that there just *has* to be something wrong with lossy audio...just because.

I know that back in my early days of encoding with LAME (about three years ago ), I used to think there was "something not quite right" with the sound. This feeling persisted for a month or so until I finally broke down and did an ABX test - lo and behold, I was guessing every single time. I've since listened to hundreds of LAME V2-encoded tracks, many of them through Sony MDR-7506 headphones, and not once have I felt that "something is missing" or "I have to work harder to enjoy/connect with the music."

This sounds like the cyber-age religion from Christion Oliver Windler, who, at the age of 19, "had got the cosmic mission to build a completely new kind of high tech multimedia superlearning school to teach this mankind in sovereignous holistical thinking to prevent it from self- destruction " : http://weltenschule.de/e_index.html
Scroll down to health information / Ear damage by MP3, DVD and digital television?



There is a difference between "being the test", and "being processed differently".
ABX tests are used in order to tell if a difference is audible, not if information is processed differently. They are appropriate for the fist kind of question.
Maybe results of different tests can show that information is processed differently between undistinguishable audio samples, but this is not relevant for the evaluation of sound quality, since they sound the same.

This said, are these samples processed differently or not ?
Can you set up an experiment ? Let's sort it out. I can take part in a test if you tell me how to do.

The man apparantly isn't a fool. Anybody interested in emailing him to (politely) ask for clarification?

Indeed not. He's, apparently, an audiologist.

But there's not much grist there for Joel Selvin's mill, anyway--since "Encoded with an MP3 encoder" is not necessarily equivalent to "poorer fidelity". Selvin also appears to be trying to hang everything on some simple-minded fixed notion about "right" and "wrong" halves of the brain and excusing himself with a "could be".

Let's analyze the statement :



1 Poorer-fidelity music stimulates the brain in different ways

This part implies that fidelity is an absolute characteristic of an audio stimulus. This is not true, it is relative to an original audio stimulus and measures the likeness between them. Fidelity has no meaning when no reference stimulus is given, this it can't fundamentally stimulate the brain in a different way than any other stimulus.

2 With different neurons, perhaps lesser neurons, stimulated, there are fewer cortical neurons connected back to the limbic system

Something is missing is this statement. Let A be a set of stimulated neurons, and B a different set of stimulated neurons.
If there is an anount Na of neurons connected back to the limbic system with the set A and Nb with the set B, it means that if Na is inferior to Nb, then Nb is superior to Na, while if Nb is inferior to Na, then Na is superior to Nb.
So, with different stimulated neurons, there can be more or less cortical neurons connected back to the limbic system, not always less.

3 fewer cortical neurons connected back to the limbic system, where the emotions are stored.

This sentence invites the reader to think that less emotions are triggered. But is the emotional impact of a stimulus directly proportional to the number of cortical neurons connected back to the limbic system ?

I also think that the emotional impact of an audio stimulus cannot be fully determinated by its audio "fidelity".
The sentence "my father doesn't like apples", however faithfully reproduced by the telephone's speaker won't have as much emotional impact as "your wife gave birth to a boy", even if the later is heard through a bad quality telephone.

Not necessarily. ABX relies on verbal report - but like I've described, sometimes only motor responses will differ in reaction to stimulus pairs that differ on a particular dimension. The flicker rate experiment described above is a good example.



So the point is: some of the complaints of people who do not like the sound of lossy files, or digital in general, might be rooted in processing differences that alternative tests would reveal. So while they fail ABX tests, there is a testable correlate to the fatigue and extra effort associated with the greater difficulty of processing lossy encoded data.



Fair enough.



Who knows? I agree this is weak - I'm not arguing for this part of Sweetow's argument - and like Nick E says, he seems to have a highly oversimplified view of hemispheric lateralization.



This is also a reasonable point - however, as the researchers do try to explain (however poorly through Selvin's weak interpretation), when you throw out 90% or so of audio information that is often consciously indiscernible, it is still quite possibly being processed differently. The psychoacoustic models that inform codec development are selected on this dimension - that is, the parts of the signal that can be removed are determined by what parts can be effectively inferred by later stages of auditory processing. The codec is tricking the system, but the (proposed) extra processing effort might manifest itself in other ways, such as discomfort with prolonged exposure.

It just might be the case that (immediate) conscious discernibility is not the only measure of subtle long term effects of impoverished stimuli.



This is currently in the works - thanks for the offer - we (Bryant and I) might take you up on that!

ABX means side by side comparison. If you have two monitors with different rates side by side (ok, not too close, so that they don't interfere with each other), with the same image on them (prefferably white), i'm sure that you'd spot it most of the times. I remember using to recognize if a monitor was at 60Hz or higher ( non-interlazed. Interlazed video was much more visible)




I want to point out that saying that 90% of audio is thrown out is also an overstatement (that i've seen other times in other threads). Would you say that a lossless coded throws out half of the audio?
A lossy coded converts the sampled data to a different data representation, then uses psychoacoustics to determine what of that data can be completely discarded transparently, then appies its own tools to reduce that data representation ( SBR, TNS, scalefactors/quantization) , and finally, apply a lossless compression to this remaining data.
Of course lossy codecs discard audio data. But not 90% of it.

Re-wind a second though. "people who do not like the sound of lossy files". If they "don't like the sound" but can't distinguish in an ABX test, what's gone wrong?

Relax the time constraints - present each of A, B and X for a day, or week or whatever. At the end of the time period, ask they if they liked what they'd been listening to. Presumably if A=original and B=mp3, then, unless they've been imagining it all these years, they should report that they liked the first week's listening, and didn't like the second. Then they have X for the third week - did they like that?


I buy your argument about the reaction time experiment, but this is different. If the only chance was reaction time, and the only way of measuring it was in a reaction time experiment - but otherwise the subject was completely unaware of the differences - no conscious effect, no fatigue, no long term effect etc, then arguably it doesn't matter for the subject.

Likewise, if the subject is unaware, on any level, that the audio signal is different, then that audio signal is actually the same for them! If someone says (over whatever time period) they don't like the sound, or the emotion, or the increase in fatigue due to some method of reproducing audio - then that should be ABX-able - or preference testable, which can be just the same thing.



btw, I think trying to pick through statements from scientists (or quacks) when they've been through a journalist is a fairly hopeless pursuit!

Cheers,
David.

wupps... I wrote a "rant" letter and it got posted!



Julian Wiegmann "

even got my name heheh now I am a ranter crap... maybe I should have written a bit more subjectively.

This all seems quite feasible, however nowhere does Sweetow actually say that the “Poorer-fidelity music” that he’s referring to is actually mp3’s. Personally I doubt that he's even referring specifically to mp3's here. I wouldn't be the least bit surprised if the author of the news article just took this quote out of context to help support his bogus arguement.

Here is a copy of the study I'm referring to. They didn't do an ABX test, but subjects never suspected the flicker rate manipulation. The point here is that subjective awareness was independent from the underlying motor processes. The authors write, "The sparser sampling implies that more cognitive processing of the text had to take place, however, and this may be one of the sources of the complaints about fatigue following prolonged use of low-frequency CRT."

Keep in mind that ABX tests are different for audio than for visual - you can view two stimuli simultaneously which could make comparison easier (or less reliant on memory). To make it comparable to an audio test, you should do a serial presentation.



Right - because probably half of the data being removed is lossless, so the figure is more like 80%. Thanks.



I agree that long-term ABX tests could conceivably address aspects of the argument - problem is, there are many other potentially confounding factors when you stretch the tests out - makes for data that could be difficult to interpret. But you're right - and I suspect that there are effects like people's preferences for certain songs, and their vague impressions are affected by low-level components.

I still think there are issues with introducing the memory/linguistic element into the equation - because perception and decision processes are distinct (psychologically and neurologically), the paradigm cannot tap into all of the effects that purely data-driven (bottom up) factors have on subsequent psychological effects. ABX is a decision task.

Again, I want to stress that I value the paradigm a lot, it just isn't the final word on whether lossy codecs can have a negative impact on listeners' experiences. People tend to over-privilege conscious judgments and vastly under appreciate the underlying processes that get you there.



Right - but what I'm saying is that while there might not be some immediate conscious detection, there could easily be long-term fatigue effects, and long-term preferences could conceivably be affected as well. Again, it is a way to test for processing correlates of some people's complaints about lossy audio. They are not all completely crazy! Just kinda.



Well, no. Most of what goes on in our brains is beyond our awareness. Again, I must go back to the flicker rate study, because it so perfectly illustrates the phenomenon. Their subjective percepts were the same, but subjects were doing many different things as a function of the flicker rates.



Of course - it's just that Selvin was interpreting researchers who share my viewpoint - so I'm arguing what I know they would argue independent of what Selvin wrote.



Could be, but then what poorer-fidelity audio is he referring to you think?

ABX tests are (more than) enough to state somebody's ability to distinguish between the quality of two sound samples.

The more number of times you ABX (the same samples.) The more reliable the results will be.

Reason, logic and statistics are on ABX "side."



And what a "strange" point of view!!!

Connecting the two contradictory statements, ABX only reveals if there is a difference in sources. The beauty of the methodology is that the tester can define his own environment and parameters; all he needs to do is somehow distinguish between two samples--it doesn't really matter when or how he does it. I assume "compounding factors" refers to changes in the environment. If so, any changes will affect both samples equally. If it doesn't, that means the samples are not identical and thus can be distinguished. If the tester believes that an "off week" somehow skewed his results, then he should resume testing until he is confident that he can/can't tell the difference; one "bad week" in eight would translate to 7/8, which is still high confidence. If p is low, then he's heard a difference; if p is high, then he's guessing.

Any way you slice it, ABX is the "winner".

Hi Light-Fire,

It's hard for me to address this since you only assert your view, but provide no argument against the many things I've written. In the other thread I link to, as well in this thread, I have explained in detail why ABX has limitations, and I've described what other tests one could do. Reason, logic, and statistics are on my side actually, haha. This isn't about ABX against non-ABX - this about measuring other responses to stimuli that reveal processing that ABX tests can miss.

If someone can't diferentiate between A and B (without previously knowing which one is the best.) It is obvious and logical that this person is not able to distinguish between the two samples. There is no argument against ABX.




How?! That doesn't make any sense!



ABX is about measuring responses to stimuli and it doesn't "miss" anything. But it erases the placebo effect out of peoples "imaginative" brains.

Contesting the validity of ABX tests is absurd!

ABX relies on conscious processing, but there is more to perception than what is consciously processed. That is a basic fact of cognitive science. Yes, ABX does eliminate placebo effects (of course), but it *can* miss processing effects that other methods measure. This is a fact actually - the question is whether this fact applies to the domain of lossy codecs.

Please explain the results of the flicker rate experiment presented above. It is a perfect demonstration that subjective perception can be oblivious to relatively dramatic processing differences. Just stating that contesting ABX tests is absurd doesn't make it so. I urge you to get beyond the orthodoxy and try to understand the argument I'm making. Would you change your view if I could show that two stimuli could not be distinguished in an ABX test, but the lossy version took longer to respond to in a reaction time task?

That experiment is coming...