Computer Music Audio Quality

bramankp -- Tue, 12/27/2011 - 08:51

 Published in TAS #219, I'm having some real problems with this article co-authored by Charles Zellig and Jay Clawson.
 
While most of the article is actually quite good I have a very hard time with the following sentence, "Although JRMC reported an accurate rip for all reading speeds, and are bit-for-bit identicle at all reading speeds, we are still able to detect sonic differences in the resulting files."
 
Really? Maybe that's not exactly what the authors meant to say but it's the equivalent of saying one can make a copy of a ripped file and somehow the exact copy sounds different. I want to trust what's being presented in the article but this is a sticking point for me. The same bits (*the same bits*) played back from a different position on the hard drive sounds different?
 
Someone please clear the air on this.

rossop -- Wed, 12/28/2011 - 18:27

Beats me every time. Its kinda like flac v wav v aif. I use aif but am hard pressed to tell the difference between them. I read this article and must confess that it was rather confusing to me.

Steven Stone -- Thu, 12/29/2011 - 10:27

 File space is so cheap why would anyone bother with compression of any kind?

If you need compressed files for your iPod you can make them yourself.

As for different ripping programs yielding different results, I will need to see more robust testing results to prove this is a sonic factor.
And since these findings are PC-only, the have absolutely NO relevance to Mac-based audio systems.

Steven Stone
Contributor to The Absolute Sound, EnjoytheMusic.com, Vintage Guitar Magazine, and other fine publications

bramankp -- Thu, 12/29/2011 - 14:06

 I don't think it's necessarily just a matter of disk space. FLAC (and other lossless formats) support metatags. WAV has no standard metatag support. Ergo, if you want your rips to carry metatags (artist, album, year, cover art, etc.) then something like FLAC is the way to go.

Steven Stone -- Thu, 12/29/2011 - 14:58

In the Mac world AIFF is the best option. It does support meta-info changes, including additional artwork.

Steven Stone
Contributor to The Absolute Sound, EnjoytheMusic.com, Vintage Guitar Magazine, and other fine publications

Robert Bertrando FB -- Sat, 12/31/2011 - 10:30

AIF or AIFF play on Windows as well, as of course WAV does on Mac. I've never been able to hear a difference (has anyone except Dr. Zeilig?), but there are certainly differences in the way each format handles metadata (most people would agree AIF/AIFF has the edge here).

HIFIDR@AOL.COM -- Thu, 12/29/2011 - 15:33

To bramankp, rossop, and Steven:
As one of the authors of this article, let me make a few comments. First we have taken some pains to describe our methods in detail in order to encourage readers to repeat our tests for themselves. Only then can they be in a position to dispute or confirm our findings. Opinions don't really count for much. Experimental results are more compelling.
In the quote cited by bramankp, we agree that one can make copies that are truly "identicle" but one cannot conclude that they therefore must sound the same. There are many factors that can change the sound between the information storage source and the DAC. For example, compare identical music files played back from an optical drive, a defragmented hard drive, and a memory stick: they will all sound different. Bits are bits all right, but time- and amplitude-based jitter mechanisms can account for many of the sonic differences we report.
To rossop: See Part 3 of our article in TAS #220 on FLAC. There are many sources and reasons why FLAC and WAV sound different, sometimes to a disturbingly large degree.
Hi Steven: We do think our results have some relevance to MAC systems. The electronic principles of the two types of computer systems are quite similar; it is the OS's which differ. It is highly likely that different software programs will also sound different on a MAC depending on how they are written and what computer resources are used. Why should jitter mechanisms not interfere with sound quality in a MAC computer? Why should lossless file compression not be audible on a MAC? If ripping programs vary in sound quality on a PC, why shouldn't they also vary in a MAC environment? The real question that audiophiles would like answered is whether there is a sonic advantage of a MAC vs. a PC. We do know of one case in which JPlay was used in a Windows 7 environment on a MAC using Bootcamp. The result strongly favored the Windows playback. Even still, this isn't a properly controlled experiment. Using iTunes as the common software would be better, but then you would still have the extra overhead of Bootcamp and Windows. I suspect that if one could match up the hardware of a MAC and PC as closely as possible, JPlay would likely outperform the best-sounding MAC playback software to a degree proportional to computing resources used. Lastly, we couldn't agree more with the argument against using any kind of file compression. While we individual users might not care about storage costs, commercial download suppliers might care quite a bit about storage and bandwidth requirements for a large operation. So while we argue this point in Part 4 of our article, unfortunately, we end users are not the deciders in how music is supplied. So finding the least harmful software for converting FLAC (or other lossless formats) to WAV is a big deal for both PC and MAC.

Chuck Zeilig, The HiFi Doctor

jcb -- Thu, 01/19/2012 - 20:55

I would like to thank you for the very useful information in your series of articles on computer audio in TAS. Prior to reading your article I was using EAC for CD ripping and then converting to MP3 at 320kb/s using Lame for my portable music player.

Based on your recommendation I tried dbpoweramp for ripping and found that indeed the wav files from dbpa sound better than the wav files from EAC. I then tried using dbpa to convert the waves to flac and found that the flac files are better sounding than MP3.

I would like to offer a tip based on my own experience. The current version of dbpa has an uncompressed flac option that sounds better than any of the compressed versions. It is not as good as wav, but my portable music player does not accept wav files. Also I found that converting directly from CD to flac does not sound as good as CD to wav and then wav to flac. It's a little puzzling how this can make a difference, but I trust my ears and those of TAS authors to find the software settings that yield the best sound.

proben -- Fri, 01/20/2012 - 06:03

It's more than "a little puzzling" -- it's quite impossible. At least, it's impossible that the files sound different from one another themselves, since there is no difference in them bit-for-bit. It's quite possible, as others have noted, that the different spaces they inhabit on a hard drive and the different ways different programs deal with fragmentation, etc. can create different listening experiences, but you can't attribute those differences to the files themselves if they are exact copies of one another. A wav file copied from one place to another or a file converted CD-wav or wav-flac-wav -- these files should all be bit-for-bit identical if they are properly copied/converted. If they are not identical, there is another problem in the chain. Many of the claims in the article series seem to betray a poor understanding of the basics of digital information.

HIFIDR@AOL.COM -- Thu, 12/29/2011 - 15:35

bramankp:
J River Media Center and dBPowerAmp both support metafile tagging of WAV files while ripping.
 

Chuck Zeilig, The HiFi Doctor

bramankp -- Thu, 12/29/2011 - 19:11

 To Chuck:
 
What you said in your reply, about the same bits sounding different when coming from different media, could be perfectly correct. However, they may or may not be the same bits. Especially concerning an optical drive with error correction.
 
However, that's not what the article claims. The article states that the same bits in two different files from the same media can sound different. Noticeably different. That's equivalent to saying that the same exact file could sound different from playback to playback depending on cacheing and other things going on in the mechanism. That is certainly not proving that different ripping speeds that produce the exact same bits are going to make different-sounding sound files. I'm sorry, but there's no way to back up that claim.
 
As for certain software supporting WAV tagging? Will that work when I play back that file from a USB HD over my Oppo player? I get metadata from the FLACs I play. Hence the inclination towards formats that support metadata tagging.
 
 
Paul Braman

HIFIDR@AOL.COM -- Fri, 12/30/2011 - 19:21

Hi Paul,
As you no doubt noticed, in the immediately following 2 sentences, we stated: "We know these results drive engineers crazy. We would love it if someone could come up with a definitive explanation that could provide input to software developers."; and I would add, that could explain these particular results. There are three responses I can make to your objection. Either 1) we are delusional, 2) there is a rational explanation, perhaps jitter, that someone may figure out how to measure in the future, and 3) how about trying to replicate our experiment using our specific equipment (you are welcome to visit and I will repeat the experiment with you). For the record, we do not believe choice 1) is correct. 
As far as whether WAV tagging being recognized on your Oppo, we haven't tested this so I can't give you an answer. Perhaps you can try it and let us know your results.
 

Chuck Zeilig, The HiFi Doctor

software_engineer -- Mon, 01/16/2012 - 19:26

@hifidr:
There is indeed a rational explanation, and it is unfortunately #1 -- you are delusional.
Jitter, by definition, does not occur at any stage of the digital process *except* at the conversion between the digital and the analog domains. It is an artifact of timing the bits; at no other point in the chain does timing matter. Your statement to the contrary is simply incorrect.

You may, indeed, be hearing differences when you play two identical files. These differences are not because of the bits in the files, as they are identical -- and will always be identical, regardless of (lossless) conversions from one file format to another. If you think this is not the case, I submit to you a counter-argument: The Internet. It just wouldn't work any other way.

If your argument is that identical bits can, by themselves, cause a difference in what you are hearing (and not something in the rest of the chain), then you are, in fact, delusional. You have heard of Claude Shannon? Perhaps your qualifications are better, I think we'd all be interested.

Robert Bertrando FB -- Sat, 12/31/2011 - 09:22

 I know I've already put this in a letter to the editor, but I would suggest that only choices 1) and 3) are likely.  I don't have your specific playback hardware, but can say that on a couple of different computer audio systems, using the software and settings you describe, I and others have been unable to replicate your findings.  I would propose that if the differences you hear are that hardware specific they are relatively meaningless and describing them does little to add to the field of computer audio or audio in general.
Think about it a little more:  two identical audio files (I assume you verified that the files are identical), read from the same hard drive by JPlay (meaning that the files first go through the hard drive's buffer, then to computer memory before being read and sent to the DAC), sound different??  IMHO, You would have to present much better listening data than you have to convince anyone that this contention might be valid.

HIFIDR@AOL.COM -- Sat, 12/31/2011 - 12:05

Hi Robert,
Lets make sure we are talking about the same experiment first. Using JRMC, under ripping options, there is a sub-menu for read speed: 1x, 2x, 4x, 8x, 16x (version 15 of the software). This is what we were testing. As an aside, it is an Interesting question to ask why the manufacturer includes this option in their software.  Second, our hearing sensitivity and acuity was highly conditioned after listening to the same test tracks over and over again for 8 months. Third, I have over 30 years of professional industry listening training. Fourth, our auditioning system was optimized with respect to power cabling, digital cabling, AC  grounding, vibration control, and speaker resolution. If your situation is not the same, then you have not replicated our experimental conditions exactly.  I contend that if the differences are hearable on any system, they are not meaningless, but rather indicate a real effect regardless of whether they can be heard in all systems or just select systems.   And the history of science and engineering includes many examples of disputed concepts being eventually proven correct. If you think the results in Part 2 are outrageously wrong, wait until you read what we have to say about FLAC files in Part 3.
On the other hand, there are criticisms that can be made of our procedures. For example, testing was usually conducted under single, not double blind conditions. Our results lack proper statistical rigor; generally speaking, we looked to replicate our results on two different systems and duplicate our tests at least once. It is also the reason why these results were submitted for publication to TAS and not the JAES. For the record, we did employ listening panels and numerous replications (at least 4) for what we felt were going to be the most controversial findings.  In the case of the particular experiment we are speaking about, similar results were obtained some time ago with an independent computer and software and at an independent time. Had we tried to repeat everything we did in triplicate with at least 2 weeks intervals between listening tests, we probably would not have lived long enough to complete this study, or at least would no longer be married.
Before we submitted this study, we ran it by a number of industry, computer, and scientific experts. In short, we got a variety of feedback, some arguing your case and others providing possible explanations for the effect we heard. Not included in our already long article is data we have accumulated which strongly supports the sensitivity and accuracy of our listening procedures and which we hope to present in a followup article. So, we favor option option 2) mentioned in one of my earlier posts. If option 1) is proven to be correct, then our results represent one heckuva a good delusion!
 

Chuck Zeilig, The HiFi Doctor

JIMV -- Thu, 01/12/2012 - 19:02

"I contend that if the differences are hearable on any system, they are not meaningless"

That is the essence of the argument. Either there are differences that can be heard or not and if they can be heard on on system, they cam be heard on others of sufficient resolution. I am just not sure from reading the article if the differences can in fact be heard. Someone needs to haul in another group of listeners and limit the testing to something that can be followed...like a WAV v FLAC file playback of one or two files.

Robert Bertrando FB -- Sat, 12/31/2011 - 16:15

That is indeed what I was talking about.  Did you in fact run checksums to determine if the files are identical?
Your argument about differences being noticeable on one (any) system being real differences is certainly true in and of itself.  However, these noticeable differences may be an artifact of some other aspect of that one system.  I'm not saying that is the case here, just pointing out how noticeable differences may have a different cause than the one you suspect, and may not be generalizable.
In any case, the example we are discussing is hardly the only non-replicable finding in the two parts of the article which have been published, just the most outrageous IMO.  I also have over 30 years of "audiophile" listening test experience, although not "professional", and I assume you are as aware as I about the myriad of confounding biases present in your tests as well as mine.
Perhaps you should also mention that these articles were rejected by Stereophile prior to submission to TAS
 
 
 
 

bramankp -- Sat, 12/31/2011 - 17:24

 That's the thing. The article claims EAC said the files were identicle. I'd either like to see the files in question or verify through checksums or another program.

rossop -- Sun, 01/01/2012 - 02:28

This maybe a bit off topic but I ripped the Dolly Parton album Halos and Horns to wav and compared them to the aiff files I did a while back. I played my favorite track, What a Heartache, over and over again and I cannot tell the difference at all. I stream from my pc to a PS Audio PWD in my main, fairly hi-rez, system. Both times I used dBpoweramp in secure mode with bit & sample rates set to "as source". All DSP settings were the same.
 
If I thought there was something else I should do to get a better sound, like upping the sample and bit rates when ripping or converting, I would give it a try.

firedog -- Sun, 01/01/2012 - 02:47

 dbPoweramp allows me to make "uncompressed FLAC" rips. This is a rip with no file/size compression in a FLAC shell. So it is functionally like a WAV file, but allows the full metatag facilities of FLAC. Note that this is NOT FLAC at "0 compression level", which is a compressed/smaller file.  IMO, this is the best of both worlds.

rossop -- Sun, 01/01/2012 - 16:34

I see what you say on my dBp as well. The only thing is you cant up the sample rate or bit depth except you can change the bit depth in the DSP section but not the sample rate.

Robert Bertrando FB -- Sun, 01/01/2012 - 13:31

I'm just guessing here, but I predict that despite WAV > FLAC > WAV producing identical files (both by checksums and audibly), they will sound different for some reason, despite the fact that depending on the operating system, all the codec programs (that I know of) use the same algorithms. I'm specifically NOT saying that playback of WAV and FLAC (even uncompressed) necessarily sound the same.
 

I think that if two supposedly bit-identical files "sound" different, either the files aren't bit-identical or there is a flaw in the listening procedure. There aren't any other likely alternatives.

HIFIDR@AOL.COM -- Mon, 01/02/2012 - 15:23

To Robert Bertrando, FB (NV):
In response to your comment above which said: "Perhaps you should also mention that these articles were rejected by Stereophile prior to submission to TAS"

I am responding to this statement because it is not accurate. I do not want such misinformation to remain in the public domain without correction. I reviewed the email documentation that I have of the timeline and correspondence with Atkinson at Stereophile. We did indeed submit the article to Stereophile on February 4, 2011 after John Atkinson's enthusiastic encouragement upon seeing our data and article outline at the Rocky Mountain Audiofest in October of 2010. He held the article for 5 months and 26 days without giving us any decision. He did eventually ask some questions in an e-mail to me on May 26, 2011. I immediately responded in detail to the questions he posed. After he failed to participate in a scheduled phone conference on May 27th to discuss his questions, and after he failed to respond to any of my subsequent emails or phone calls, we finally concluded he was not interested in publishing it and we formally withdrew the article from Stereophile on July 1, 2011. Perhaps the distinction between us withdrawing the article versus Atkinson rejecting it is one that only the authors and the publishers care about. But it is important to me because it speaks to the motivations behind some of the commentary.

Chuck Zeilig, The HiFi Doctor

Robert Bertrando FB -- Mon, 01/02/2012 - 16:48

I appreciate your clearing the air there, but fail to see that any of the posts here have any relation to that issue, except yours where you mention submitting to TAS rather than IEEE.

I think you need to spend some more time thinking about the implications of my comment above. If in fact the files are identical, it really makes no difference how they were made (in terms of ripping speed); if in fact they sound different, it has to do with how the computer processes them, and the same result might be obtained by merely copying an audio file to a different place on your drive. More likely, IMO, is that there is a different flaw in your playback chain or listening test procedure. If the files are NOT identical, find out why not.

rossop -- Tue, 01/03/2012 - 00:21

I myself dont really understand all this stuff. What I can say is this: My best digital files and CDs when played back through my PS Audio PWT / PWD still dont stir my soul and give me goose bumps like my best vinyl when played back on my good old Simon Yorke S7. Maybe when I update my PWD to the Mk 2  series in a month or so might make all the difference but I wont be holding my breath and will be VERY supprised if it does.

bramankp -- Tue, 01/03/2012 - 07:09

 rossop: This thread got a little derailed but no one is arguing the merits of digital versus analog here.
 
It was simply my original contention that two perfect copies of the same source (that are thus identical to one another) will not sound different when played back on the same media. If, somehow, they might sound different for some truely mysterious reason then it is still independent of the speed at which the files were read from the original source.
 
So far, there are only two questions that remain unanswerered in my opinion: do the files *really* sound different and are the files *really* identical?
 
Chuck Zeilig asserts that the answer to the first question is yes, indeed, they do sound different. That leads me to believe that the answer to the second one is no, the files are not truely identical. It would be pointless to attempt to recreate the experiment because I can't guarantee my resulting files (created using different ripping speeds) are the same as those originally created. We must therefore require either MD5/SHA hashes of the files or use of some other tool to see if the files are, indeed, clones of one another.

HIFIDR@AOL.COM -- Tue, 01/03/2012 - 15:01

To Bramankp:
We are in fundamental ageement with you. We assert the files do sound different and believe that in some manner the two files can therefore not be indentical. If jitter is somehow involved, this would be irrelevant when copying a Word file. However, in a music file, jitter is known to be critically important but is beyond our capability to quantify. 
 

Chuck Zeilig, The HiFi Doctor

Robert Bertrando FB -- Tue, 01/03/2012 - 15:20

Jitter has nothing to do with the information in a file. It occurs during the A>D or D>A conversions. If the two files are bit-identical (if md5 isn't good enough, then use Fdupes to do a bit-by-bit check), then they are the same. They can be copied from one drive to another, from one directory to another, to a flash drive, etc., and they won't change; the computer will always treat them the same way. Depending on your hardware, however, two identical files in different media (e.g., one on a hard drive, one on a flash drive (thumb), one on a SSD, or perhaps if things are screwy enough, two different locations on the same drive) might undergo D>A conversion slightly differently.

proben -- Tue, 01/10/2012 - 01:07

As far as I can tell, Robert is correct - if two files contain the exact same information, then the D/A converter is processing the same information; if the output is different, there must be something wrong with the converter or something after that in the audio chain.  But digital information is digital information  -- "jitter" won't make any difference to the 1s and 0s that are recorded by the system.  (Unless you're saying that there are 2s and 3s being recorded that we don't know about?) 
I just finished reading part three of this series and the claims are even more outrageous.  The article claims a persistent degradation of sound quality after each conversion from WAV to FLAC to WAV again.  If your equipment is working properly -- easily verified with a bit-by-bit check as suggested above -- this does not seem possible. If the claim is true, each subsequent conversion will compound the sound degredation so that eventually you would get a WAV file that is worse than an mp3.  If this is happening consistently in blind experiments I would look at the methodology and setup of your experiment.  

HIFIDR@AOL.COM -- Tue, 01/10/2012 - 20:09

proben and Bertrando:

When we wrote up our findings, we were quite aware that they would prove controversial. Our purpose in publishing was to alert audiophiles and the audio/software industry to problems in computer playback sound quality; it was not to discover or prove mechanism which we felt was better left to the experts. To call our results outrageous, borders on flaming. We stand by our results. And, as we pointed out, we repeated these listening tests several times with different listeners and could reproduce them consistently. They were more audible on the poorer of two different computers. We interpreted this to mean that the effect was caused by some process within the computer and its interaction with the playback software. As we stated clearly in Part 3 of our article, we do not question whether FLAC is a lossless format. But in the process of software-mediated decompression (or conversion from “uncompressed” FLAC format) there is a problem. We speculated, and I emphasize the word "speculated", that it might be mediated by some form of jitter. We ourselves found the 0's and 1's to be the same. Dr. Bertrando, jitter can be generated virtually anywhere in the reproduction chain prior to the DAC. It is hearable only when the conversion process to analog occurs. But, it can be generated in many places and exists prior to this conversion step. Is it so hard to understand that even though the 1's and 0's remain the same, there can be sonic differences that are not due to the digital information itself? Years ago, before jitter was recognized, there were many Flat Earthers like yourself who just knew in their hearts that those crazy audiophiles who heard differences were.....er...crazy, or worse, outright liars or snake oil salesmen! It is even possible that the mechanism for some of the effects we have quantified by single blind testing has yet to be discovered. It would not be the first time in the course of scientific history this has happened, as I am sure you will have to admit. We have yet to explore the influence of quantum physics on sound quality of computers, n'est-ce pas? I do not want to perpetuate the insulting tone of some of the posts we have seen on various forums about our findings. Let's try to be a little more gentlemanly shall we?

Chuck Zeilig, The HiFi Doctor

Robert Bertrando FB -- Tue, 01/10/2012 - 21:16

"Years ago, before jitter was recognized"  If I recall correctly, jitter was recognized pretty early in the game, late '80's at the latest.  It was described (and probably still is best described) as a (mal)function of the D>A interface, specifically of S/PDIF initially.  It was quickly recognized that it also occurred at the A>D interface. The point is that jitter generated in the A>D process is encoded in the data of the audio file, the "0's and 1's"; it's not some other property that's independent of the data, until more jitter is generated in the D>A process.  But more to the point, any computer (and particularly a PC) treats files with identical data as the same.  If you are proposing that a digital audio file has some quality not detectable by bit-by-bit comparison, then the simple act of copying that file (e.g., to another directory, to another hard drive, etc) is just as likely to corrupt the data as a WAV > FLAC > WAV conversion.  Any attempt at backing up the data or just moving it will do the same thing, because as far as the computer is concerned, if the data is the same the two files are identical and that's all it cares about.
 
BTW, virtually all professional audio workstations (on which our recorded music is recorded and mastered) use FLAC in addition to many other audio formats, and even uncompressed files are copied, backed-up and moved frequently.  If what you propose is true, how can any recorded music ever sound good, since it's being degraded from the moment after it's recorded (i know it some ways it is, of course, but not to the extent that your claims would imply), even with minimalist techniques?
 
i've always been in the subjectivist camp of audiophiles, but our physical world still has "laws" which govern the way things work, and my wishing so can't make them change.  I think it's pretty clear that if two digital files have identical data, they are interchangeable apart from whatever properties are imparted by physical media.  For example look at our banking systems; encryption systems used there are FAR more complex than computer audio (128 bits the last time I checked, instead of 24 or 32 bits for audio), and if there were ways to have identical digital data mean two different things, we would be in even more trouble than we already are.
 
When a scientific study produces results which are seemingly contradictory, the next step is to re-examine the study's methods.  There have already been a number of reasonable critiisms and questions posted around the Internet regarding your study's methodology, and you yourself wondered about it.  It might have been wiser to address those issues rather than publish results which seem un-credible.

bramankp -- Wed, 01/11/2012 - 07:11

 Chuck, stop overlooking what I am saying and stop focusing on bits of the argument that are completely irrelevant. Do I really have to pull out a bunch of qualifications just to make you listen, very carefully, at the simple thing I am saying?
 
I have never questioned that you could have, indeed, heard differences in the playback of your audio files. I have always left it as a possibility. You and a bunch of your golden-eared posse can hear all the differences you want. The only thing I am saying is that if two WAV files contain identical information and if those same WAV files sound different, you cannot attribute the differences to a FLAC conversion process. You could have done WAV->FLAC->WAV. You could have done WAV->Meridian Lossless Packing->WAV. You could have done WAV->copy->WAV. Given that there are an infinite number of ways of making identical copies of the same file you can't attribute quality differences to that single step in the process.
 
All other steps in the process? Up for debate. Lively debate, no doubt. I just want you to recognize and admit that your supposition is flawed where you attribute differences in quality to the FLAC conversion process. It's that simple. Please assume I took the time to write out how I work for The Nielsen Company as a Software Engineer and as one of the quality testers for our audio watermarking programs, blah blah blah <insert qualificiations here>.
 
Don't reply about how I claim your listening tests must be flawed. I did not. Don't reply about how I don't understand the fundamentals of jitter. Irrelevant. Don't reply about how I doubt your methodology. I do not. Please, stick to my point.
 
 
Paul Braman
Principal Software Engineer
Engineering and Technology
The Nielsen Company

HIFIDR@AOL.COM -- Wed, 01/11/2012 - 12:32

Paul, Thank-you for your reasoned and reasonable reply. Do not think that we didn't find some of these results troubling. We very much want to resolve the apparent discrepancy between our results (which we still stand by) and a rational explanation. Perhaps it might be fruitful to take our discussion offline and communicate directly. You have my e-mail address; why not communicate directly and we can speak by phone.

Chuck Zeilig, The HiFi Doctor

Robert Bertrando FB -- Wed, 01/11/2012 - 14:15

The problem with standing by your results (in a more general sense) is that since there are as yet poorly identified methodologic flaws, ALL your results must be viewed as invalid, even though I suspect that some might be real.  If the methodologic flaws could be identified, then perhaps not all of your findings would be found to be (equally) affected.

returnstackerror -- Wed, 01/11/2012 - 17:19

A lot of heavy hitters on this thread.... but I will dip my toes in anyways.

I will not try and address any specific points raised nor take sides on this. But I think there are several factors related to ALL read/write media (Hard drives/Solid State drives/USB sticks etc) that hasn’t been discussed

Firstly, all of these are required to be formatted (NTFS or FAT for windows) so that they are readable by the OS and by definition, over time, these volumes will become fragmented (less so of course than with day to day general purpose computing as the audio files we are writing and potentially deleting are large and contiguous by nature).

The other factor is that the read/write heads on a hard drive will do more or less work to retrieve disk sectors depending on if the sectors are on the outside of the platters or the inside (this affect may also exist to a much lessor degree on SSD/USB drives). Believe me on this as placing data on specific parts of a set of drives when running large benchmarks (such as TPC benchmarks) is common practice (been there/done that)

So these two factors, taken separately and together could contribute to differences.

All the activties described ( original ripping/re-ripping/converting from one format to another) will land either the master copy or its clone on different parts of the drive and therefore change the access path taken as the OS and the drive in question navigate the chain of disk sectors that represent the file(s) under test.

If this is a cause then the only true way to conduct a test is to:
- create a master image of the file under test on a "golden image" drive
- reformat a "test" drive for each test
- copy the file under test from the "golden image" drive to the "test" drive and run the test from this "test" drive
- for the paraniod, check sector locations for the file under test with analysis software

I suppose the analogy is whether a streaming network player is best connected via hardwired ethernet rather than wifi. Ethernet has latencies and an under pressure wifi network could induce jitter (not audible drop outs as such but a point prior to a drop out where the replay program is placed under stress due to mild packet starvation)

Peter O'Shea
Sometime Computer Geek

PS. I wont bother describing how different degrees of disk latency will affect the execution of a ripping/playback program.. not saying the bits will be different but memory buffer allocations, cpu idle times in the microsecond realm, power demands within the PC etc will vary (jitter anyone?)... but thats is too paraniod and too hard to control in even the most stringent test protocol

bramankp -- Wed, 01/11/2012 - 18:53

 Peter,
 
I really didn't want to delve into the specifics of PC hardware but since you brought it up ...
 
I have an average-speed laptop at work with an average-speed hard drive that is averagely-fragmented. It takes me under 10 seconds to load an entire 10 minute stereo 24/96 WAV into Cool Edit 2000 (really old audio editor we have lying around). That means the entire audio file is available, in RAM, ready for playback. I imagine most playback PCs could perform similar tasks which means the only area that could really be a problem is the timing with which the program feeds audio samples to the device driver (which would cause audio dropouts) and the timing with which the device driver/OS feeds audio samples to the DAC (which is where the jitter comes into play, if at all).
 
If there's really a problem with playback, I imagine it's the fault of the software doing it and not the 90% of hardware that feeds data to said software. That last 10% of hardware interactions are suspect but we can leave that to hear-say and circumstance.
 
Paul Braman

returnstackerror -- Wed, 01/11/2012 - 20:36

Agreed that playing from a pre-loaded memory image would negate this but to prove that with the highest performing programs used in the testing, you would need to see some performance graphs showing cpu utilization,memory utilization,  I/O per second, read/write kb per second, interupts per second etc during playback to prove that indeed the song is completely in memory.

I dont use computer based playback but what you mention raises some questions:
 
(1) if I start one of these playback programs and start to play one song (say the 10 minute 24/96 wave file in your example), does playback of that song take 10 seconds (or so) to start, indicating anecdotally it is pre-loaded completely or does it start in say 1 second. In the later case, while some of it may be played out of memory,  during playback the hard disk must be being accessed to keep the memory buffer full with the remainder of the song

(2) what about a playlist of 15 tracks of say 24/96 or higher. During playback is there a 10 second (or so) gap between each song?.
 
If so then we could conclude that yes indeed, each track is pre-loaded sequentially. If there is say a 1 sec gap between each track, we have to conclude that only partial buffering is occuring and that during playback the hard disk is being accessed to keep the buffer full.

On this second part, it would be possible to pre-load the entire 15 track playlist into memory (which would be indicated by a long pause at the start... at least 2 minutes in total using your example track) but that would assume at least 1gb of memory being allocated for this pre-load buffer (which is a lot for a PC)

(3) following on from (2), what say my playlist is 100 songs (ie impossible to pre-load all songs into memory), if we "suffer" a 10 second delay between the playing of each track then yes each track is probably pre-loaded... if not, then disk reads will occur during playback

I would be interested in any real life experience with the above scenarios.

My assertion still holds water to some degree, even in  memory based playback, if the disk is accessed during the playback of a track

Remember I am not taking sides here, just wanted to see if anyone thought different disk latencies could be a factor in these hither to unexplained playback quality perceptions.

Thanks,

Peter
 

returnstackerror -- Wed, 01/11/2012 - 21:07

As an aside, if the authors of this study didnt capture detailed OS performance metrics like the ones I noted above (and many others), then they missed a great opportunity to potentially explain different playback quality perceptions.
 
How any particular program utilizes a PC's resources could give an insight into this.
 
For example, maybe the worst performing programs used 50% more CPU then the better ones, indicating poor/inefficient programming code on the poorly performing programs. Or maybe they accessed disk much more during playback.
 
Peter
 
 
 
 

ncdrawl -- Fri, 01/20/2012 - 14:49

 this is so dissapointing. Another member of the audiophile press attempting to muddy waters which by right have been crystal clear for decades. The "series" is woefully devoid of relevant data, shows a fundamental misunderstanding of computer architecture, digital audio, sound propagation, the ear-brain system, and (most importantly) human psychology.  For goodness sake, the genesis of what we now know as sampling theory was in the 30s! It is well understood by those that are experts in the related fields. There are no differences in sound here, because they simply cannot exist, not in any way, shape form, no way, no how. 
 
Hifi writers everywhere... PLEASE STOP.  You are defrauding the public, in my  opinion. Atkinson if he did reject the "article", was wise to do so, as it was a bunch of subjectivist, "golden eared" blather. With the retorts being the usual"your system isnt evolved enough to distinguish differences" "you don't hear as well", yadda yadda.  It is science. Period. 
 
Dear god. 
 
http://www.youtube.com/watch?v=BYTlN6wjcvQ

ncdrawl -- Sat, 01/21/2012 - 00:38

 Paul Frindle(DSP Mastermind and Digital Audio Expert) had this to say. 
 
All of these words, pomp and hubris is typical of HiFi audio
discussions, mostly inspired by a commercial desire to sell you
something you never knew you needed. LOL.

There are only 2 things a DAC responds to; 1) the data we feed to it ,
2) the timing information it gets.

For the data; data is data - there is no possibility that identical
data sets could ever produce a different sound, regardless of their
origins. In short, numbers are numbers - and if they are the same,
they are the same - period.

For timing it's a slightly different matter, because it is essentially
and analogue signal - it's properties (i.e. rate) are analogue in
concept.

So there is a slight possibility that interference on the clock signal
can affect the DACs performance, if the timing is modulated in some
way, by slightly changing the rate of plyout with time.

Of course a good DAC system will circumvent this possibility by using
it's own internal clock and some buffering of the data - so that the
DAC's timing follows the filtered average rate of the input timing -
such that short term timing rate variations (due to interference) do
not make it through. I.e. it will synchronise to the input clock,
rather than simply passing the input clock straight into the DAC.. You
can almost think of it as stacking up the data as it ccomes in and
playing it out using it's OWN high quality clock set to the same rate.

However - as you can imagine, price competition tends to rule out
anything that might increase the cost of the product - and so you will
have to spend extra money on your DAC system to get this..

So where does the timing modulation of the clock rate come from
(sometimes called jitter)?

First on the list are wires and connections. Line frequency hum, RF
and interference getting into clock cables will modulate the timing at
the recieving end. If you DAC doesn't reject them (as above)
performance may suffer.

Next on the list is bad design. For instance, in consumer players
where the DAC is within the player box, power supply modulation from
bad design may cause internal circuits to interfere with each other.
For instance if the motor servo is being varied by slightly eccentric
discs and/or wobbly ones that stress the focus servo, the cyclic
changes to current draw my affect the clock oscillator if the power
supply is badly desogned etc.. So although the data is correctly read,
the DAC itself may perform less well than it might, due to internal
design flaws.. And of course this is true of any electronic system...

So there you go - it's a simple as that - no PHD required :-)

In summary:

Data itself coming from different systems cannot cause a change in
sound - if the data values are both identical.

But hardware may perform slightly differently if the timing integrity
is compromised.

Having an external DAC does not avoid the issue - unless it's a very
good one with internal timing re-clocking. In which case it will be
effectively immune from timing errors and will sound the same whatever
way it gets connected, providing the data is correct.

So instead of worrying about data coming from Macs or PCs and/or a
whole load of hot air from HiFi zealots filling pages with 'waffle' -
get yourself a high quality DAC and say bye-bye to the whole
discussion :-)

I hope this helps..

Paul

P.S. - I have no idea why discs are marked as suitable for different
speeds, apart from commerically generated market differentials. It's a
physical process and I can see no reason why a disc should not be
written and read at whatever speed the laser is capable of.. It may
simply be that after manufacture the slightly eccentric ones are sold
as slower - to prevent shaking of your drive in operation!!

Michael G. -- Sun, 01/22/2012 - 23:51

I am about to take the plunge into computer audio so these questions may be basic or discussed before.
Based on the findings of the recent 4-part series in the last issues of Absolute Sound, here is what I need advice on.
1. When ripping CDs to transfer your music to an external hard drive or NAS with their recommended software, say, JMRC or db Power Amp, does the CD/DVD drive on one's computer (assume it is decent quality) provide good sonics or should you be using an external CD-ROM drive like the Samsung unit discussed.
2. Should you rip to 16 bit/44 KHz first, save it, then upsample the saved file to say 24 bit/176 kHz or is it done in a combined ripping/ upsampling step.
 
 

Steven Stone -- Sat, 01/28/2012 - 12:51

1. Use your internal drive. Keep it simple.

2. Do not upsample 44/16 material. Keep it native.

3. Take the plunge, keep it simple.

4. Keep it simple and you will spend time enjoying music instead of screwing around with your computer.

Steven Stone
Contributor to The Absolute Sound, EnjoytheMusic.com, Vintage Guitar Magazine, and other fine publications

rossop -- Sun, 01/29/2012 - 03:04

Best advice ever

thesurfingalien -- Sat, 01/28/2012 - 11:37

My congratulations to the TAS magazine and the authors of the series of articles discussed here.   Over the years I have had some pretty good laughs reading audio-related articles, but the claims made in this article that:
 
a) bit-for-bit identical files are responsible for audible differences because they originate from another source / ripper or;
b) files that have been compressed & decompressed while resulting files remain bit-for-bit identical suffer from audible quality loss;
 
beat all others... hands down!
 
Unfortunately, this is not something to laugh about... at all! The only thing claims like this will cause is the birth of yet another fairytale that the gullible endorse and with full conviction spread amongst others, giving the objectivists one more reason to ridicule the subjectivists in general.
 
I truly hope that the TAS magazine management comes to their senses and will stop publication of the articles until some real controlled, verifiable and repeatable double-blind testing have been performed to substantiate the claims made.
 
Regards,
Peter

proben -- Sun, 01/29/2012 - 05:22

^^^^ THIS. I do not believe these studies were double blind. I can't think of any other explanation for the claim of progressive degradation of files that have been compressed and decompressed. If they are compressed and decompressed properly they should be bit-for-bit identical, which is easily verified with software tools (it's unclear if the files were verified in this way, so it's possible the authors were measuring degradation of the files by poor compression software, though highly unlikely, since such problems would more likely manifest in corrupted files than in progressive sound degradation). A useful study would verify bit-for-bit identity of the files before looking at listening experiments. If the files are bit for bit identical, it is highly improbable that listeners would consistently report progressive degradation with each compression/decompression, unless the study was not truly blind (in which case, psychological suggestion could certainly factor in).

In addition, they ought to be able to compress/decompress the file over and over again to the point of unlistenability, the way you could if you continually re-recorded a piece of music on analog equipment, or if you compress/decompressed using lossy compression such as mp3. I don't believe this is possible. I think if you took the file from the beginning of the chain of compression/decompression and the file at the end of the chain (say, after 20 rounds of compression and decompression) and gave them to blind listeners you would not see them consistently identify one as the better sounding file, and whatever differences they perceived between the two would be rather minimal. I think if you played three or four different files out of the chain in no particular order, you would also not see any consistency to which files were preferred.

However, I believe that if you play the files in the chain in order and say "this is the original file," "this is the first copy," "this is the second copy" etc., that you would find that people notice progressive sound degradation over time. Psychology plays a big role. The problem is not that anyone is delusional, but that we condition ourselves to expect certain perceptions and then we perceive them; it reminds me of the "power balance bracelet" tests described in this video - http://www.youtube.com/watch?v=2xBVEM2iMns

The authors suggest that it is possible that there is information being transfered into the file in a copy (or a compression/decompression, whcih amounts to the same thing) that is not measured in a bit for bit analysis. I was joking about it above referring to "2s and 3s" that are copied alongside the 1s and 0s. I'm not ruling out the possibility of such "magic smoke," but I think such claims are quite extraordinary and really demand extraordinary attention to details of observation and analysis, and I don't believe that has taken place here.

thesurfingalien -- Sun, 01/29/2012 - 07:18

Hi Proben,

To start with your last remark about the possibility of information transferred into a file by (de)compression / copying that is not measured... In my 20+ years working in various areas in computer technology I have never ever encountered such behavior. If this really was the case, it would have shown up in other forms. After all, a player is not any different from other applications that reads a file (wherever it resides) and does some processing.

Let me give an example... If I were to publish an article in which I claim that files with raw graphic information, after a few compression/decompression cycles with Winzip, started to degrade in quality, clearly visible to me on my computer setup even though the files are bit-for-bit identical, I would become the laughing-stock of the year. NOBODY WOULD BELIEVE A SINGLE WORD OF IT.

But in all fairness, the above scenario in effect is not all that different than music reproduction: converting digital information to its proper analogue form.

Unfortunately, in the audio-scene it is perfectly acceptable to make the wildest claims without any form of sustainable proof. It is even mentioned (in the article?) that the results of the (some?) listening tests do not hold up statistically and testing methods might be flawed to an extent. Still, the authors maintain the differences must be caused by the most unlikely (read: impossible) cause: the files.

Peter

Edit:

Even if files somehow become "contaminated" with "other" information, music-players only can work with the ones and the zeros in the files, just like any other application, for example, a tool to verify whether or not files are indeed bit-for-bit identical.

proben -- Sun, 01/29/2012 - 22:26

Peter - you are exactly right; I was trying to give the authors the benefit of the doubt that perhaps there was the possibility of some magic smoke that we haven't yet discovered in digital audio files. If such elements exist, only real blind testing will help uncover it; but the reality is, if it existed, it would also affect non-audio files. If I compress and uncompress a Microsoft word file 12 dozen times, the end file still looks exactly like the first one and is just as readable.

Robert Bertrando FB -- Sun, 01/29/2012 - 08:30

Fortunately it appears that in the audiophile world at large (information sharing is indeed wonderful in the Internet Age) these findings are not being taken seriously, except in the sense that computer audio HARDWARE and connections can indeed interfere with the audio signal (in mostly measurable ways), and that some software/hardware interactions also probably affect the sound.

Alan_HK -- Sun, 01/29/2012 - 13:13

I have read the findings published in TAS #219 and subsequent issues co-authored by Charles Zellig and Jay Clawson with great interest, mainly because their experiments mirrored what I have been experiencing in my own set up over the last years, while trying to find the best possible sound from my computer based system. My compliments to the gentlemen for their incredibly comprehensive testing work. Apart from the findings on ripping of CDs, which I am not very experienced in as I mainly listen to downloaded files, my own findings from experimenting with different file formats, playback software and software based upsamplers over some years, are going in the same direction. I came to computer audio by way of a Sony compact hifi system I had bought some 10 years ago for its superb sound, which connected to my Sony notebook via USB. This allowed me to play back audio files I had on my notebook through my Sony hifi system. It all worked fine under windows xp, as Sony had provided the necessary USB drivers and to some extent also under vista, though I had to use the old xp drivers as Sony had faded out compact systems with USB computer connections. The only limitation of this USB connection was that it did not allow for higher quality sound than 44.1 kHz-16 bit. I don't know whether it was this limitation or the trouble with the USB drivers that were not available for my new notebook that prompted me to by an EMU 404. Anyway, the EMU allowed me also to send higher resolution files through USB to my Sony compact system, which I connected via its analogue input to the EMU. After undusting my old Audio Space M6-300PP power amps and an unused music fidelity V3 buffer stage with separate power supply, I started experimenting more seriously with higher resolution files and various file formats. Anyway, through various system permutations, I am now listening to a computer based system that has as source a self assembled windows 7 based music server, connected via firewire to a Brainstorm DCD-8 wordclock synchronised with a trimble thunderbold gps disciplined clock, which outputs to a dCS Delius DAC, also synchronised with the Brainstorm’s clock signal. The Delius is puts out to a Music & Art tube preamp, which is feeding my Audio Space power amps. As alternative I use a Minimax Plus tube DAC with Burson opamps that is fed from the Brainstorm clock through an RME ADI-192. And yes, the sound is better when the DACs are connected to the Preamp, rather than directly to the power amps, which is testimony for this incredibly good but inexpensive preamp built by my friend Peter Lam in Hong Kong. In my current system, based on my own experience with various files and software, I use nearly exclusively wav files, either in their original from as downloaded or converted from flac. Most of my wav files are upsampled by the Voxengo r8brain Pro software to 192 kHz 32-bit (fixed point). This, in my experience, gives me the best sound in my own system, and I am playing back with JRMC and the JPLAY 4.1 plugin.
In essence, my own way on the road to the absolute sound from a computer as audio source, confirms the findings published in the TAS reports by Charles Zellig and Jay Clawson. These findings may currently not be able to be confirmed by unequivocal scientific evidence, but they may be experienced by carefully listening. At least this is my experience, and let’s face it, there were times when people did not believe that power cords can have an influence on the quality of sound.

proben -- Sun, 01/29/2012 - 22:25

The above post is exactly why double blind testing is essential. Sure, you can convince yourself a bit-for-bit identical file sounds different, or that upsampling a digital file somehow adds clarity, but that demonstrates only the power of human psychology rather than anything about audio quality in digital files.

thesurfingalien -- Mon, 01/30/2012 - 04:29

Hi proben,

With regards to up-sampling I must say that I find it not all that hard to believe people may experience differences. During the up-sampling process existing 16-bit information is scaled up and additional information is added. As this is done according to some kind of mathematical / predictive model, the resulting data-stream might result in a different analogue output that the original one.

Now, the question is if this results in an enhancement or just a difference... However, I suspect that the just-a-difference can easily turn into an improvement if one believes that "more is better" (44.1/16 vs. 192/24) while in effect we are talking about an interpretation of "how things should sound like". I guess personal taste also plays a role here.

Peter

All content, design, and layout are Copyright © 1999 - 2011 NextScreen. All Rights Reserved.
Reproduction in whole or part in any form or medium without specific written permission is prohibited.