PCM v DSD Comparison: 16/44.1, 24/96, 24/192, 64x DSD, 128x DSD

A file format shootout. A DSD versus PCM grudge match. Bit wars. A sample rate smackdown. Can 24/96 take 24/192? Does DSD stomp PCM?

There's a lot of buzz about DSD, talk of CD's demise, and music offered in various PCM formats from CD-quality up to 24/192. What's a person who cares about the quality of their music reproduction experience to do? And the answer is, listen. I got the idea to listen to various files in different resolutions and formats after receiving some comments on my Mytek Stereo192-DSD DAC review asking about double rate DSD or 128x DSD. Namely, is 128x DSD better than 64x DSD. It sure sounds better, mathematically! But the only way I know to figure out if one thing sounds better than some other thing is to listen. So that's what I set about doing.

Let's get a few things straight before we dive into listening. First off, this comparison is necessarily limited by a number of factors including the fact that I'm using one DAC, the Mytek Stereo192-DSD DAC (see my review here), 3 tracks each in 16/44.1, 24/96, 24/192, 64x DSD (2.8MHz) and 128x DSD (5.6MHz) all sourced from the same master tapes and all converted using the same process (see below). I do not intend to draw any sweeping conclusions since I'm aware of the limitations that the environment I'm listening in necessarily imposes. That said, I believe its fairly safe to extrapolate somewhat but I will try to keep my reaches within an arm's length.

Let me fill out the rest of my setup for this listening session: a PC running Foobar2000 connected to the Mytek Stereo192-DSD DAC with an Audioquest Carbon USB cable, a pair of Kimber Kable Select KS 1126 Balanced cables to the Pass INT-30A, and out to my DeVore Fidelity The Nines. I'm using a PC because at present the Mytek DAC only supports 128x DSD via ASIO.

Credit Where Credit Is More Than Due
I owe a huge thank you to Bruce Brown of Puget Sound Studios (you can read our Q&A here) for providing me with the music under scrutiny. Bruce was kind and generous enough to agree to create these files for the sole purpose of this listening experiment and I want to thank him for his time and for providing some great music to listen to over and over again.

Here's a word from Bruce on the process he used to generate the various file formats:

These were original master tapes that I had of these tracks. I made one tape dub of these 3 tracks that would facilitate the test more easily. The tapes were recorded to RMGI SM900 tape via a Studer A80RC MKII that has been greatly modified. This was also the playback machine. I created test tones so the tracks would have equal volume and the tape was played 5 times into each sample rate. I started off the test at 16/44.1 into the newer Korg MR2000sBLK using Mogami Gold 1 meter balanced interconnects. No EQ, Compression or Gain was used, just a straight transfer. The files were then transfered and labeled to an external hard drive and sent to you via zipped ftp.
Thanks Bruce! I'd also like to thank Mytek for the very extended loan of their DAC for the purposes of this comparison.

More Details
And here's the details of the Foobar2000 setup for 128x DSD playback over USB (this is detailed in Mytek's setup PDF):

  • Upgraded the Mytek to firmware to 1.7.1 (if you experience static along with your DSD music, make sure you've upgraded to this latest firmware and it'll go away)
  • Install foo_out_asio and foo_input_sacd
  • Go to File > Preferences > Output > ASIO and doubleclick foo_dsd_asio
  • Select "Mytek Stereo192-DSD DAC ASIO" for ASIO Driver and "ASIO Native" for DSD Playback Method
  • Open a Dale's Pale Ale (optional)
  • Go to File > Preferences > Output and select foo_dsd_asio
  • Go to Tools > SACD > Asio Driver Mode and select "DSD"
  • Click "Apply" and "OK" and "OK"
Let the listening begin!
There were three sample tracks provided [to avoid copyright issues the tracks names will remain private]. All PCM tracks were provided as WAV files and DSD as .DFF and again we have a 16/44.1, 24/96, 24/192, 64x DSD and a 128x DSD copy of each track. I loaded up all tracks into Foobar2000 and spent time listening. I listened to all tracks all the way through a number of times and then went back and focused on certain parts of tracks that emphasized or highlighted the differences I heard. And I will say up front that the differences between CD-quality and DSD make me want to put that word quality attached to CD in quotes.

16-bit/44.1kHz
Let's start on the positive side - music sounds punchy and bold and is clearly well recorded. There is also a sense of dynamic edginess, a hardness that creeps into the sound at dynamic peaks as if you're not hearing the full sound. This holds for each track and lends the music an overall sense of harshness as if it was recorded in too small a space. There's also an unnatural aspect to decay and reverb where they sound stunted reducing both a clear sense of the space of the recording as well as the full natural voice of an instrument or vocal or a finger snap. There's also an emphasis on transient attack that shifts the music's tonal center away from body.

24-bit/96kHz
Everything opens up as if there's more space in and around the recording. Decays sound more natural, musical images in space take on a more solid and fuller feel. Dynamics have a greater sense of ease, tone colors are richer and overall music sounds more natural and relaxed. On the loudest passages which are most prevalent on Sample #2, there is still some sense of hitting a wall as if some of the sound has been cut off.

24-bit/192kHz
The size of the recorded space is once again larger and much more natural sounding, even compared to the 24/96 versions. There's also a better sense of micro-detail or hearing exactly what the musicians are doing. A more intimate picture. This lends the music more drama, more impact because there's more variety to the sounds. Dynamics also appear to have a greater swing from soft to loud and there's an overall sense of ease that is not present in the previous versions. Upper frequencies take on a sweeter sound, cymbals sparkle, and horns sound more fleshed out and with less glare.

64x DSD
Holy crap! The musicians just relocated into real space. Finger snaps are 3D as compared to the PCM versions. Actually every aspect of the music is more dimensional. Dynamics are astoundingly natural-sounding and there's an overall ease to the presentation that translates into an almost uncanny sense of place. Instruments take on a more complex tonal palette and you can hear into the performance much more. Harmonica sounds like a harmonica as opposed to a piece of one. The CD-"quality" version sounds relatively dull and lifeless as if we're missing out on a ton of detail and subtle nuance.

128x DSD
Space is the place. Dynamic swing sounds unrestrained and fluid. There's absolutely no sense of harshness, edginess, or harmonic foreshortening. Music is rich and full. The scale of the recorded space is rock solid and stable and music emanates from this space in a completely natural way. CD-"quality" sounds like a cardboard cutout in comparison. There's a lot of "space" in the Sample #1 recording and the opening vocals resound in that space. With the double-rate DSD you can hear the size of this space whereas with the CD-"quality" version it sounds as if the singer's voice is hitting a wall. There's no depth, no complexity to this reverb tail with 16/44.1. With each step up the PCM bit/sample rate ladder and with DSD this tail gets more and more fleshed out, solidifying a sense of time (distance) and place.

Conclusions
The differences I noted between 16/44.1 and DSD are dramatic and easy to identify. Even though I was dealing with very good-sounding recordings, you could even say the CD-"quality" versions sounded good, when compared to higher rate PCM formats and DSD you get an increased sense of dynamic ease, harmonic complexity, micro detail, and a better sense of the recorded space (and time). A more natural-sounding presentation. The differences between the 24/96 versions and the 24/192 versions were not as significant but differences were there to be heard none-the-less. The same held for the two DSD versions so the most obvious jump was from 16/44.1 to 24/192 and DSD.

I admit that this could all be due to the Korg MR2000sBLK simply being better at converting to DSD and higher rate PCM as opposed to 16/44.1 or the Mytek DAC's ability to reproduce DSD over PCM. But I've also heard similar improvements with other recordings albeit under less controlled circumstances. My feeling is the differences noted between CD-"quality" and 24/192 and DSD are so marked as to suggest that the medium goes a long way in conveying the message.

Sample #2 opens with the band's percussionists playing soft to loud to louder still. In the CD-"quality" version this sounds flat and stunted as if it was recorded in a room just large enough to fit everyone, nothing has room enough to breath. When you listen to the same track in 24/192 or DSD you easily realize the musicians are mimicking the sound of a train approaching going from far to near, to nearer still as opposed to just going from soft to loud. And you realize this because you can hear into the recorded space and the music comes from a deeper, quieter place. This train reference is central to the message of this song, "...they curse this train that brought them to Johannesburg", so there's more drama, more emotional impact and the higher you go up the PCM ladder the more of this drama you get and you really jump right onto the tracks with DSD.

"How can I be sure..."
I think Bruce Brown was teasing by picking the song for Sample #1, each time I played it I wondered why some doubt was creeping in (kidding). I figured it was worth putting my findings to a test so I put all tracks on random play, turned off my monitor, took off my glasses (which is the same as being blind) and let the music play. Out of 8 random trials (this was all I had the patience for), I was incorrect once thinking that I was listening to was 24/96 when it was in fact 24/192 (this was my first trial and I rushed my decision). All other selections were correct.

But I'd say I got lucky with some picks. The differences between 24/96 and 24/192, for example, are not huge but there are differences if we listen closely. It also helps to listen to complete songs as opposed to switching back and forth between snippets. I do feel confident that I could identify the 16/44.1 version each time as well as identify PCM v DSD but even here Sample #2 and Sample #3 were easier to determine as compared to Sample #1. Then again, if you asked me to do this again in a few months using your computer's speakers all bets are off.

Getting Engaged
This was an interesting and informative comparison and I hope I conveyed a sense of the differences I heard. More importantly I'd like to stress how these differences can affect the way we perceive music and how they can impact the meaning of a song as opposed to just a focus on sound and sound effects. For me its clear that 24/96, 24/192 and DSD are superior to 16/44.1 in many meaningful ways and I've come to this conclusion based on listening to more than today's 3 tracks in different system settings. But the ability to listen to these various samples all generated from master tape using the same equipment and methods, thanks again to Bruce Brown, has helped to solidify this belief.

Where does this leave us? Well, there's the pesky issue of the quality of the original recording which obviously trumps all of the above. Then there are all of the variables, some of which we've touched on here, which makes drawing any firm conclusions potentially misleading. But to my mind when speaking in generalities why not go for ideals? Sure CD-"quality" can sound really good but higher bit/sample rate PCM and DSD can sound better. How much better will come down to the quality of the recording, the quality of the transfer, the playback chain, and your listening habits and preferences. Better still, if you value music first, all of this takes its rightful place in line.

Share | |
COMMENTS
zjaj's picture

I've seen it pop up here and there over the past year or so, but the best rollup of all of that information I've found is here:

http://people.xiph.org/~xiphmont/demo/neil-young.html

As for my own ABX tests, I'd be happy to do so, except for one problem: I can't reliably gather proper source material like you can. Even for my HD downloads, I can't guarantee that they were produced from the same master as the CD rip I'd be comparing them to, so any test result would be suspect. Beyond that, I can't compare DSD to PCM at all because I don't have any source for the former beyond my SACDs, which I can't rip. Finally, not all of us are blessed to have access to as quiet a room, or as high-quality equipment, as would be necessary to do a test like this in a scientifically-valid way.

That's why I think it's important that folks like you, who have the means to do so and are making the claims, to do these tests.

Michael Lavorgna's picture

The author cites 2 papers - Dan Lavry's white paper "Sampling Theory For Digital Audio" from 2004 which talks about an optimal sample rate of 60KHz and "Coding High Quality Digital Audio" by Bob Stuart of Meridian. Here's a quote from that paper's conclusion, "The CD channel with 44.1kHz 16-bit coding (even with noise shaping to extend the resolution) is inadequate...Even 48kHz sampling is not quite high enough..."

So there is clearly some argument to be made on a technical basis to refute Monty's position just by reading the papers Monty cites to support his own argument that is summed up here, "Empirical evidence from listening tests backs up the assertion that 44.1kHz/16 bit provides highest-possible fidelity playback."

In terms of my good fortune, I find myself in this position largely because I value the experience of listening to music on a hi-fi. The enjoyment of listening to music on a hi-fi is not, imo of course, reducible or adequately represented by any white paper or ABX test. The true measure of listening to music on a hi-fi is time. Time spent listening and ideally enjoying music. So if someone like Monty prefers to listen to lossy encoded files or limit his experiences to 16/44.1 who am I to tell him to do otherwise?

I prefer to listen to what I find to be the most musically engaging experience I can get and I have found that higher resolution PCM and DSD formats (and LPs) typically provide a more musically engaging experience.

zjaj's picture

1. There are other articles that discuss this as well. This one, for instance:

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

which discusses the conclusions of this paper (which is behind a membership wall):

http://www.aes.org/e-lib/browse.cfm?elib=14195

2. Those two papers you mention here also contradict you with respect to there being any difference between 24/96 and 24/192, and Lavry's directly argues against 192kHz. So I wouldn't lean on them too heavily.

I also read your original response to the xiph paper, and besides the obligatory snarkiness (hey, that's why they invented the internet, right?), I didn't see anything you wrote to actually refute anything he wrote. And with respect to Neil Young, I love the man, I love his music, and I believe that he believes that he hears a difference. But he has also been playing electric guitar on stage at ear-shredding volumes for four decades or so, so I don't know whether I would lean on his opinion on the relative audio quality of 16/44.1 vs 24/192. Honestly, I think I'd take your opinion over his. :-)

You say you "appreciate his... advocacy of double-blind and ABX testing," but not enough to actually do it. Considering that there are some reasonably smart people arguing against your position from a scientific standpoint, I'd think you'd want to take a crack at it. Why not go for it?

Michael Lavorgna's picture

Who offer 24/96, 24/192 and DSD (music and compatible gear) argue against all of the "reasonably smart people" who argue against it.

My interests are in the enjoyment of listening to music so double-blind and ABX testing do not apply.

zjaj's picture

Because they have something to sell. And they don't really "argue" against it in any scientific way, they put out marketing fluff and lean on consumer assumptions.

In the context of "enjoyment of listening to music," I agree completely that double-blind and ABX testing do not apply. But that is not the context of this article. This article is about which sounds better, and in that context, double-blind and ABX testing do apply, more than your opinion does.

I'm beginning to think the emperor is feeling more of a breeze than he's letting on.

Michael Lavorgna's picture

So all of the engineers that design DAC chips, for example, capable of handling 24/192 and DSD actually know that there's no point to it other than hoping that some DAC manufacturers who also know that there's no benefit from higher sample rates will make DACs using these chips so they can sell some stuff that they know sounds worse than 16/44.1. And of course the recording engineers who offer HD music are all in on this conspiracy as well and the only reason people believe they hear an improvement with HD music is because they've been told they will in "marketing fluff" when in fact it sounds worse than 16/44.1. And you believe this based on a few things you've read that contradict what you yourself have experienced....

Here's the thing: I want to agree with you. I have DSD recordings that, to me, sound better than their red-book counterparts, and HD downloads that, to me, sound better than an equivalent ripped CD. But I also think it's important for me to know whether I'm fooling myself or not.

Your emperor appears to be less well dressed than mine.

btw - there are free downloads available from SoundKeeper Recordings that offer files in various bit/sample rates (16/44.1, 24/96, and 24/192) all sourced from the same master so you can perform your tests yourself.

zjaj's picture

I didn't say they knew there was no point. There are entire industries where the people who are selling something geniunely believe they are solving a problem and helping their customers, but it's still based on misinformation and assumptions. I'm not putting this in that category per se, but given the papers we've referred to here, I'm keeping an open mind about certain aspects of it.

And yes, the things I've read have made me question what I have experienced. How could it not? The science in them makes sense to me, and there isn't really any contradictory evidence out there. Do you really believe you are immune from confirmation bias? It's part of being human. The only way to prove it is to take your foreknowledge out of the equation. The amazing thing to me is that you don't seem to be the least bit curious about it.

(Thanks for the free downloads. They don't solve my space/equipment issues, though.)

It's clear that you're going to use every rhetorical method you can think of to avoid an ABX test, and still be able to maintain that you can tell the difference between 24/96 and 24/192. There is science that says you can't, and you're unwilling to prove you can in any scientifically meaningful way, so.... It's a shame, because after stumbling onto this site not too long ago, it seemed like a place where I could learn a few things. But you have better things to do than chase one reader.

Michael Lavorgna's picture

I did in fact listen to a random sampling as I discussed and was able to correctly pick 7 out of 8 times.

The issue I have with ABX tests is they do not prove anything significant in this case, imo. For example, let's say I correctly pick DSD v PCM every time (which I did without knowing what file was playing) and 24/96 versus 24/192 the majority of the time (which I did). And let's say we can also easily show how these files differ from one another in terms of their underlying data (which we can). The most important question remains - what are the perceived differences and are they worth it? And the only way anyone can know the answer to this question is to listen because of the number of variables involved including music choice, system, room, and personal preference. That's because we're talking about listening preferences in music, not some objective criteria that is either met or not.

But the real question to address your point is, why didn't my - for all intents and purposes blind test - suffice for you?

It's a shame, because after stumbling onto this site not too long ago, it seemed like a place where I could learn a few things.

If the only way you feel can learn something is through reading about ABX tests, then you will have to look elsewhere and I agree that is a shame.

CG's picture

This is a friggin hobby.  For enjoyment.  As in smiles, fun, good feelings, happiness, to name a few possible descriptive terms.  I am constantly amazed how many people don't seem to be having fun.  But, to each their own.

From the engineering, or scientific side if you like, the sampling rate and bit depth for mixed signal fidelity are just a couple (albeit very important) considerations of many.  This is true not just for audio reproduction, by the way.  

earwaxxer's picture

I have been talking about using various software upsampling techniques, filters etc for a couple of years now. Its an easy and cheap (free) thing to do, and anyone can try it (foobar, Sox etc). Anyone can take a redbook file and upsample it, play with the filters, dither, aliasing etc and see what happens.

I have found, that in the tests that I have done, I like the upsampled files better than the native 16/44.1. Not by a ton, grant you. I can listen to them back to back. Back and forth, over and over. It would clearly follow, IMO, that a remaster of an analog tape to these various sample rates/resolutions would sound different, with an advantage to the higher rates. This just makes sense to me.

What I would say to the doubters: do some work of your own. Download foobar, learn how to use it, add in some open source SRC plug ins, and see what you get. Discard it if you dont like it, but more importantly, tell us what you did and what you found to be the result. Then you can intelligently express an opinion.

Brucest's picture

This author suggests that such views of digital are a feature of ingorance of sampling theory and that 192k has the potential for harming the sound.  I wish someone more up this this would comment on this very divergent viewpoinog.

 

 

http://people.xiph.org/~xiphmont/demo/neil-young.html

Michael Lavorgna's picture

Here

It's also simple to test his theories for free if you have a 24/192 or 24/96-capable DAC. Just download some free HD music and listen. Barry Diament and his Soundkeeper Recordings offers free sample tracks, "from the same album, same mastering session, etc." for this exact purpose here.

You can also read about the importance of higher bit/sample rates and digital filters here.

 

Vigna ILaria's picture

In my mind there is one key aspect of any comparison between DSD and PCM that is not yet clear (to me, at any rate). 

DSD is a subset of the generic format Sigma-Delta Modulation (SDM), just as Red Book CD is a subset of the generic format PCM.  In more than 99% of all PCM DACs the actual digital-to-analog conversion is done in SDM format.  It is simply easier to do it that way from an engineering and cost perspective.  The incoming PCM is converted to SDM using a "filter" - but this is not what you or I would understand as a filter.  It is a very complicated conversion algorithm.

Likewise - as I understand it - in more than 99% (and it could actually be 100% - I would like someone to clarify this) of all PCM ADCs, the actual analog-to-digital conversion is done in a SDM engine, and the resultant data stream is converted to PCM using an analogous conversion algorithm to the one mentioned above.

It is important to note that SDM and PCM store the music data using completely different representations, and there is no mathematical equivalence between the two.  Therefore conversion between the two formats is inherently lossy.  Powerful algorithms can reduce the losses for sure, but from a mathematical perspective my understanding is that the lossy aspect is fundamental.

So the issue for me is that in any real-world qualitative comparison between DSD and PCM, is it possible that all we are hearing is the effect of the SDM/PCM conversion algorithms?  In which case PCM will always give something up to DSD.

Michael Lavorgna's picture

So the issue for me is that in any real-world qualitative comparison between DSD and PCM, is it possible that all we are hearing is the effect of the SDM/PCM conversion algorithms?  In which case PCM will always give something up to DSD.

Hmm. My immediate response is - NOS DACs would be the 1% since they avoid the PCM > SDM step, so in this case you could then say that all true NOS DACs (that avoid PCM > SDM) should be closer to DSD playback...While I do not have extensive experience with NOS DACs, I'm not so sure that this would be the case as their presentations vary widely from DAC to DAC.

That said, I very much enjoyed your thought-provoking post and will have to think more about it!

deckeda's picture

I'll summarize my understand of it like this: Delta-Sigma is easier to perform analogous conversions and like you say, that's why you find it used. But PCM has always been easier to edit and distribute.

dmawhinney's picture

Not to dispitue your findings, but they are perfectly matched to what an a priori assessment would have concluded. However, in my readings of various audio publications, I have several times run across articles outlining the noise generated by the DSD process - a noise level that increases in direct proportion to frequency. In fact, I think John Atkinson wrote about this phenomenon a year or so ago claiming that DSD will forever be inferior to 96k PCM. See <http://www.craigmandigital.com/education/PCM_vs_DSD.aspx> for a presentation of my dilemma. Please be very aware that my mind in not made up on this subject; and in fact the degrees of separation may well be far less than the "damage" done in the recording, mixing, mastering process.

Michael Lavorgna's picture

...are about as useful as a posteriori knowledge at the race track ;-)

Here are two examples of what John Atkinson had to say about listening to DSD:

Cookie Marenco Dems DSD

Sony Shines

...and in fact the degrees of separation may well be far less than the "damage" done in the recording, mixing, mastering process.

I agree.

judmarc's picture

What Monty's essay and its references don't account for is conversions.  

The conversion between DSD, which is in the native one-bit "language" of virtually all chips used in A/D and D/A converters these days, and PCM was referenced in a couple of the comments here.  

Another conversion that has occurred for PCM input in the chips used in nearly all DACs for the past two decades is "8x oversampling."  In order to avoid audible aliasing errors from the filtering necessary to convert digits to music, the conversion from digital to analog is done not at the input sampling rate (44.1, 96, etc.), but at "8x" rates - 352.8 and 384kHz.  8x rates are attained in three rounds of 2x conversion, if necessary.  44.1 and 48kHz inputs must go through all three rounds.  But 88.2 and 96kHz inputs only require 2 rounds, and 176.4/192kHz just one.

Each conversion, even though it is just a simple 2x "multiplication," requires a mathematical filter.  The way these filters are accomplished is by means of something called Fourier transforms.  A property of Fourier transforms is that the more you optimize the performance of a filter in the time domain, the less optimized it is in the frequency domain.  So no filter can be perfect; every one (and therefore every conversion) involves tradeoffs.  Thus your DAC performs this filtering, and time/frequency domain performance tradeoff, three times for a 44.1/48kHz input, but only once for a 176.4/192kHz input.  (And yes, this is true as well in most DACs that say they don't "upsample."  There are relatively rare examples of DACs, R2R "ladder" DACs like those from 47Labs in Japan, or the Phasure NOS DAC, a one-off design, that don't use 8x oversampling.)

So that's the actual science, the objective fact, of what's going on inside your PCM DAC.  The question then becomes whether these performance tradeoffs are audible in two situations: Michael's, where the source material has been recorded at different sample rates; and earwaxxer's (and mine), where we've used sample rate converters in the computer (Sox in earwaxxer's case, iZotope in mine).  As reported by both Michael and earwaxxer, and in my own listening experience, the answer appears to be yes.

Certainly in Michael's case this should not be surprising, but what about earwaxxer's and mine?  Actually, it shouldn't be too surprising there either.  The computing power in today's PCs, combined with the ability to optimize filters as easily as the next software release, means the filtering done in software in a computer can very possibly be better (less audible effect) than that done in firmware in a DAC chip.

Michael Lavorgna's picture

Thanks for that very well considered comment, judmarc. Your point re: 8X oversampling occurring in 2x steps (where necessary) reminds me of something Charlie Hansen talked about in our Q&A:

The advantage of performing the computation in a single pass is that there are always rounding errors at each oversampling operation. If you perform it all at once, the rounding errors are minimized. But if you perform 16x oversampling as a series of four 2x stages (the normal way, as it is the cheapest way), then the rounding errors are compounded four times.

Very interesting....

firedog55's picture

I've done DSD, vs other hi-res and Redbook testing at home. Not truly scientific, but blind tests where I randomize a playlist and then try to guess what version I'm listening to.

Results: the biggest difference for me vs. Redbook is DSD. It just sounds a little warmer and more natural, not forced, and with more detail and space. I can accurately pick it out more than 50% of the time, so for me no other testing is necessary.

I've also put on some of these mixed playlists when listening more casually, and even when in the next room (open floor plan) and not in a direct line to the speakers, I can pick out the DSD. It just sounds different (better to my ears).

I have no problem in principle with someone doing ABX testing, but I'm not sure the results will mean anything. My experience is that we are talking about small differences between the sounds of the formats. Noticeable, but small. Even "audiophiles" who don't have much experience may not pick out the differences. Like most listening, it is a learned skill. With experience, you learn to hear the differences.

Once you've trained your brain to pick out the differences, you hear them when you listen. This isn't some kind of illusion or bias, just training your brain  to notice certain sounds (or lack thereof). I've had the same experience with friends who say mp3 sounds the same as a CD. It does - to them - until... I point out the differences on a few songs, and get them to notice. Once they've trained themselves to notice, they can pick out the difference  between mp3 and CD even on material they are hearing for the first time.

microtone's picture

Hey Michael, your post is highly informative. But i´ve read it 3 times, and sill coudn´t understand if you think that DSD64 is better sounding then 24/192 PCM.

You said that any HD PCM and any DSD is better than CD. Witch I agree positively with.

You said that 24/192 is a little better than 24/96, which I felt the same using my EMU 1616m DAC at home with ADAM A5x monitors.

You also said that DSD128 is a little better than DSD64, which is problably true, due to more information captured by DSD128.

But Do you think DSD64 is better than 24/192 PCM, or they are similar to each other?

Or 24/192 PCM is similar to the sound quality of DSD128?

Of couse, I´m refering to your findings in this specific listening test.

16/44.1  << 24/96 << 24/192 ????  DSD64 << DSD128

Thank you

Michael Lavorgna's picture

In this particular scenario I would pick DSD over PCM.

DSD has a quality that I do hear from PCM which I tried to describe in this piece. I also had an opportunity to hear a number of DSD sources in different systems at CES and they only served to reinforce this opinion. There's an organic quality to the sound of DSD playback that I do not hear from PCM that's more akin to analog tape. This holds for native DSD recordings, which are admittedly few and far between, but also for analog tape to DSD recordings.

That said, I'd like to stress that the quality of the original recording is the most important factor, and ones attachment to the music obviously even more so. So I look at DSD as simply another tool in the computer audio playback tool chest. I've heard great sounding CD rips, wonderful 24-bit 44.1/48/88.2/176.4/96/192 downloads, and superb-sounding LPs. As I said in my recent AWSI for Stereophile, computer audio and a turntable gets you the best of both worlds.

Maxvla's picture

Michael, thanks for this article. I recently acquired a Matrix X-Sabre DAC (ESS 9018) that is capable of both DSD64/128 and DXD (352.8KHz). I began searching for some samples and came across this site: http://www.2l.no/hires/index.html All of their files are from the same master DXD so it proved a great resource, and for free! I picked up a few of the tracks in both DSD64 and DXD to compare. I was really impressed by both, but quickly I found differences in the way the sound was presented. Where the DSD64 track featured violin sections that sounded slightly compressed as if it was a single player but slightly dithered, the DXD track revealed separate players coming together as a whole. Perhaps if the DSD were 128 it would even the playing field, but I'm wondering if this is more of a difference in format. This could of course also be how my DAC is processing the data.

Have you compared DSD to DXD? If so can you comment?

Michael Lavorgna's picture

Congratulations on your new DAC! One thing to keep in mind with the 2L samples is they all are sourced from a PCM format, DXD. I'd suggest trying one of the free samples from Channel Classics which are DSD through and through.

I recently compared DXD to DSD but not with the same source material and I talk about it in an upcoming review that will be published very soon. The bottom line for me is each format can prove to be musically engaging whether we're talking about CD-quality, DXD, or DSD so having a DAC capable of handling all of these formats is the real win win.

Cheers.

ironsienna's picture

 

Michael,

Thank you very much for your great informative article. I have a question. It seems that you compare PCM and DSD played from the same DAC. But what about PCM playback from really expensive DACs? Is the difference of sound between 192 PCM played from an ultra high-end DAC, to the DSD played from a mid priced DAC that great? I am asking because I recently acquired a Weiss MEDEA+ DAC. The quality of playing high rez PCM from this DAC is stunning and I really cant think of what more can be done for digital sound to sound better than that, so real and analogue like.... But it does not play DSD and personally haven't heard DSD played from a native DSD DAC. So is it worth buying a cheaper DAC to play DSD files natively on it, or my DAC plays PCM so fine that there is not noticeable difference.. Can you please comment on that?

Thanks again for the time you took to write such a great review!

Michael Lavorgna's picture

And great question! I turned my answer into a post.

Cheers.

ahuesjp's picture

Michael,

Thank you for the article. I found it very intersting.

As a computer engineer, HiFi audio is a big hobbie of mine, I am trying to figure out the possible causes of the perceived difference between 94/24 and 192/24 in PCM.

Let me start by saying that I do believe in the value of high resolution music, and I can hear the difference between 44.1/16 and 96/24.

This makes sense to me. The difference in sample word size between 16 bits and 24 bits is very large (2^16 = 65.5K vs 2^24 = 16.7M). It gives you a much higher resolution to describe what is recorded. To my knowledge, this makes the primary difference.

The change between 44.1 KHz and 96 KHz sampling rate also makes a difference, but the difference is in the maximum frequency that can be represented. The maximum frequency that you can represent/record on 44.1 KHz sampling is 22 KHz, on 96 KHz sampling rate it is 48 KHz. While I know of no instruments that produce sound above 20 KHz, I know that there are some interaction harmonics produced that can exist above 20 KHz, and we are recording those (Note that I am intentionally staying away of the argument of whether humans can hear it or not).

I read somewhere that is that even analog recordings, the tape bias is in the 30+ KHz range as its upper limit. Which means that the maximum frequency that is being "recorded" is less than 40 KHz.

The one question I have is in the perceived difference between 96 KHz and 192 KHz. It might exist, as you found. But it doesn't make sense to me.

Most microphones can't record much above 30 KHz. The high end microphones might record up to 50 KHz (ex. Sennheiser MKH 800-P48). Your DeVore Fidelity The Nines speakers specification say that their requency response upper range is 40 KHz. They both should be well served with a 96 KHz sampling rate, and its maximum representable frequency of 48 KHz.

For 192 KHz sampling rate to make a difference, there would need to be important information between 48 KHz and 96 KHz (max representable frequency by 192KHz sampling rate). Even without getting into the argument of whether it can be heard by a human, I don't see how there is relevant information in the recordings above 48 KHz (microphone imited), or that it can be reproduced by the speakers we use.

I know it is perception, and I am not trying to prove it wrong. I am just trying to understand where the difference could come from between 96/24 and 192/24. I realize that we might not have an answer, but in your opinion, what do you believe to be the possible cause of difference between 96/24 and 192/24?

Thanks, Juan

Michael Lavorgna's picture

While I know of no instruments that produce sound above 20 KHz...

As a matter of fact, there are lots of instruments that produce sounds above 20kHz. The most noted paper on this subject is There's Life Above 20 Kilohertz! A Survey of Musical Instrument Spectra to 102.4 KHz by James Boyk. Here's the first sentence from the abstract:

At least one member of each instrument family (strings, woodwinds, brass and percussion) produces energy to 40 kHz or above, and the spectra of some instruments reach this work's measurement limit of 102.4 kHz.

So from one perspective, if we want to capture and reproduce a live acoustic event, for example, we would need to capture frequencies above 20k. But frequency response is not the sole reason for higher sample rates. In brief, filtering and noise shaping are better served by higher frequencies by moving these processes out of the audible range.

Big Ups to Mohan V's picture

Enjoyable read and what I would probably think too, but why only take off your glasses after you knew the difference between the sound of each file?  If a test subject feels the DSD128 'should sound better' than the DSD64, listens to each one, hears a difference and is later able to distinguish between the two, an original preconception cannot be excluded from tainting the results.  I understand this was not designed to find how to distribute food across the planet but it may have have saved you a bit of hate from some seemingly very angry personalities! 

NB Angry people; it's ok, Mr Lavorgna is just expressing an opinion, not testing the safety of therapies for your high blood pressure.  Isn't this meant to be for entertainment?  Get angry about poverty or climate change or something else that poses a risk to you and others, not how the author listens to his tunes.

JIMIXY's picture

Hi, I follow a website called 'Trust me I'm a scientist.com' who has issued a golden ear challenge,

"Our promise to write a glowing, feature-length article about whoever becomes the first person on record to show he or she can reliably hear an improvement offered by any super-high-resolution file format under properly controlled conditions...It could very well be that 16/44.1 PCM or 320kbps MP3 represent the very pinnacle of audio quality as far as the human ear is concerned. (At least that’s what the overwhelming weight of science seems to suggest so far.)

But hey, it’s worth trying, right? Especially since so many marketers claim there’s a real difference. And if that difference is real, it should be acknowledged."

I said I would let you guys at audiostream.com know about it as it might offer worthwhile coverage for audiostream were someone to take it up.

Regards

http://www.trustmeimascientist.com/2013/10/07/update-on-the-golden-ear-c...

Pages

X
Enter your AudioStream username.
Enter the password that accompanies your username.
Loading