DSD mastering and quality

[Rushton]

@stevensegal

So because you can’t attenuate DSD, a fatal and incurable flaw…

Just to clarify - yes, you can attenuate DSD. But it must be in what is sometimes referred to by the Sony term “wide-DSD” which is the multi-bit variant of PDM. The marketing term “DSD” is technically single-bit PDM only. For ease of conversation, even though risking some lack of ultimate technical accuracy, I’m simply using the term DSD since that’s how most of us in the listener community think of it. In both single-bit or multi-bit PDM, one is completely within the same PDM domain.

Merging Technologies’ NADAC DAC has long used a PDM (wide-DSD) volume control for attenuation. Mixing DSD via HQPlayer Pro includes attenuation in the process while staying entirely in the DSD domain. So, yes, you cannot attenuate single-bit DSD, but you can attenuate PDM staying in the “DSD domain”, as we laypeople may use the term, and staying completely out of PCM.

Still, PCM via Pyramix is likely always to be the choice of most large recording houses because of convenience, flexibility, vast capabilities, and economy. Time is money, and they can control and deliver a very satisfactory product more economically by staying in PCM. Mixing in DSD is likely always to be a purist solution applied only by small number of perfectionist shops who can afford taking the additional time or don’t otherwise need to apply the vast array of DAW tools that a DVD project in Pyramix provides to them.

Again, for what its worth, we continue seeing Jared Sacks releasing Pure DSD projects, but he is selective about when he can afford to do so. The 2021 Anna Fedorova Shaping Chopin album is an example, HERE. In this case, no post-production was applied, so this is NOT an example of mixing in DSD. It is simply an example of a respected producer choosing to record and release in Pure DSD256. I rather like the sound of this album as I comment in a review, HERE.

Or, if one prefers jazz to classical, take a listen to Jared’s recording of Angelo Verploegen’s The Sweetest Sound in Pure DSD256 HERE. Again, not mixed in DSD but still a Pure DSD256 release from Jared.

[stevensegal]

This is just playing with words. This is all from Merging Technologies’ DSD guide. Note that they say one reason for developing DXD was to process DSD data, but DXD is their default sampling rate.

They explain why you would want to work with DSD:

Of the 4 bullet points at the bottom:
. This is what PSA and most others do - it’s DSD other than the parts that need to be processed, which is done in DXD
. This is the main reason for DSD - archiving - and Merging have a large client base, from the US government to UMG, doing DSD archiving.
. This is what Jared Sachs does
. Using Pyramix as a transport for analog mixing (PSA tried this).

When you say, “mixing in DSD”, it doesn’t exist. It’s not a thing. Pyramix spell it out. DSD files can be mixed using Pyramix’s DXD processing or after analog conversion. The Jared Sachs approach of getting the mix correct in the recording set-up is hugely difficult and time-consuming. It’s like direct cut vinyl, the same process, incredibly difficult but it can be done.

Merging_Technologies_DSD-DXD_Production_Guide.pdf

I cannot play DSD128 or DSD256 files. My system can play DSD64 files, but converts them to DXD.

[stevensegal]

Let’s get real about this.

An effectively “live” capture of a solo piano or acoustic trio (your two recordings) is not going to be overly challenging for a recording engineer.

And if you really need to re-record something, it could probably be arranged without huge cost.

Do you know how much it costs to re-record an orchestra? Paying 100+ people to turn up and play, hiring an orchestral size venue? If the cause of the problem was a couple of microphones being set at the wrong level, which could be fixed on a multi-channel DXD mix in seconds at no real cost … you must be insane.

Here’s a real world example. Channel Classics and Jared Sachs have been recording Florilegium, a London-based baroque chamber orchestra for over 30 years. Florilegium crowdfund preparing the band and recording wages and Channel Classic pay the recording and production costs. It was recorded at St Michael’s, Highgate, a very nice church a few miles from here, they probably paid $200 per day rental. It was recorded in DSD256 and mixed on Pyramix.

YES - JARED SACHS ALSO MIXES IN DXD

Obviously this was because of the budget and the album was released on CD. If it had not been done this way it would never have been made.

41819.pdf

Pure DSD256 is a sport for a handful of audio aesthetes that for the vast majority of classical recording artists and their audiences are a complete irrelevance.

Florilegium’s last recording was crowdfunded by 85 supporters raising about $9,000, so about $100 each, and they get the CD when produced. Having been to several of their performances, I doubt their typical supporter knows what DSD is. This is the real world. PS Audio may have buried $millions in a DSD recording studio that, by Paul’s account, loses money. He’s living his dream. Go figure.

[Rushton]

@stevensegal

When you say, “mixing in DSD”, it doesn’t exist. It’s not a thing. Pyramix spell it out.

Yes, “it’s not a thing, it doesn’t exist. Pyramix tell us so.” Sorry, that’s not true. Merging’s guide is about what Merging is doing. Not what it possible and not possible to do.

Channel mixing in PDM DOES exist for a Pure DSD result. It is being done–no DXD involved. And, to my ears, the Pure DSD mixed result sounds better. I’ve posted a link to download the sample files. Folks who can play DSD256 can play the files and decide for themselves: same DSD256 source channels, same music selections, different sound quality. One mixed in DSD via HQPlayer Pro with no PCM, the other mixed in DXD via Pyramix.

Heck, one might even hear a difference (better or not better, but a difference) listening to the alternate sample files through a system that converts everything to DXD. I don’t know, I don’t have a way to try that. I’d be interested to know whether one hears a difference between the samples. But, in this experiment there would be yet another conversion. And every conversion creates some artifacts.

But, you are definitely, clearly, absolutely correct that mixing in DSD cannot currently do all the other things to a music file that Pyramix can do within a DXD project. (Shoot, I don’t even know whether it’s possible to frequency EQ a channel. It’s certainly not possible to add reverb and do a whole host of other manipulations that are needed for some recordings.) And, there is no doubt that staying in Pure DSD is a real commitment – it’s not as quick and easy a process as a DXD project in Pyramix. The tool available to mix in Pure DSD is not a DAW, and that’s a big limitation. But, this doesn’t mean that Pure DSD mixing does not exist, because it does, and it is being done.

[stevensegal]

Again, just playing with words.

You can fly to the moon, has been possible for 50+ years, but no one does because it costs a lot of money and few people would be prepared to pay to make it viable.

The thing with HQPlayer seems to be a fix of just one of the numerous things that a mixing engineer usually has at his/her fingertips, and a pretty crude fix at that.

Saying HQPlayer is DSD mixing is a bit like someone who can just about fry an egg calling themself a chef.

The TRUTH is that DSD was developed as a storage format. It was never intended as a production format. If you have DSD files you can mix them by sending them back to analog, or use tools like Pyramix in the digital domain. Or you can do a live capture and not edit it at all.

The latter is not uncommon, but they tend to record several performances, choose the best parts and cut them together.

[stevensegal]

p.s. It’s not me saying “DSD is un-mixable”. Pyramix say so in their literature. So do plenty of other people.

[Rushton]

@stevensegal

Let’s get real about this.

@stevensegal, I don’t disagree in the least about the practicalities and the economics. I’ve talked with Jared about this at quite some length. This all makes sense and it is the reality.

I truly am talking about the small ensemble and solo instrument recordings. And, I am speaking about the niche recording labels. And groups at Florilegium’s size are probably past the size of groups one might apply this to. Certainly would not work with the BFO. And certainly would not work with the small ensemble who feels the need to correct every third note in their performance. A recording of a string quartet requiring over 500 edits is not possible, practical, nor economical.

Yes, I have said, Jared records in DSD256 but he mixes in DXD on Pyramix. And, as a result, I download his DXD files, not his DSD256 files. Jared’s DXD files are today his edit master files. And, on my system, those edit master DXD files sound better than his Pyramix output DSD256 because my DAC happens to do a better job converting DXD to DSD256 than does Pyramix. But the big reason is to avoid yet another file conversion – always download the file that is closest to the edit master because every file conversion adds some artifacts.

@stevensegal

p.s. It’s not me saying “DSD is un-mixable”. Pyramix say so in their literature. So do plenty of other people.

I’ve taken the time and effort to pull this information together because the conventional wisdom is that it is not possible to mix in DSD. @paul made that statement in this thread because that is what he believed at the time. You quote Merging saying this.

But the problem is, that statement is a myth. It is not true.* And there will be some smaller niche labels who will hopefully experiment with mixing (and EQ’g) in DSD and they will be able to deliver an improved sound quality with the effort should the effort make economic sense for them.

As Gonzalo Noqué replied when I asked him why go to this trouble:

Going this route is definitively not very practical and still pretty limiting compared to the DXD path within Pyramix. But to my ears, it delivers what in my opinion is the best digital sound achievable today when recording Classical music recorded in real acoustics. So, in my mind, I can only think: Why wouldn’t I do it? Whether the difference is large or small, noticeable by many or few, doesn’t really matter. HERE

And, when I asked Jared why he laid down his microphone tracks in DSD256 when he planned to mix them in DXD, he replied: “It sounds better. And someday the tools may be available to remaster in Pure DSD and the files will be here to re-release in better sound quality.”

In the end, my efforts are not about Linn, Hyperion, DGG, or any of the vast number of large recording companies using DXD. It is purely about what sounds better. Some will pursue that goal. If enough small players head that way, perhaps someone will decide that funding development of a DSD DAW will be a good investment. It would only take a million dollars or so. In the meantime, let the perfectionist pioneers have a swing using the tools available. The audience will be small, just as it is today. But I’m not the only one seeking out Pure DSD files to enjoy.

• Added comment for clarity: when engineers say DSD cannot be mixed, this is both technically true and entirely misleading. When Merging says it, it is self-serving and disingenuous. The technical definition of the Sony marketing term “DSD” is single-bit PDM. To mix, one has to be working in multi-bit PDM (which is not PCM). Sony used the term “wide-DSD” to reference 5-bit PDM. Multi-bit PDM can be of any number of bits wide and still be in the PDM domain. Jussi Laako would prefer the technically more correct terms Pulse Duration Modulation or Pulse Width Modulation (PWM). But this is all going down a deeper rabbit hole for a layperson’s discussion. Thus I continue to talk simply about DSD the way most of us listeners would.

[jazznut]

@Rushton

And someday the tools may be available to remaster in Pure DSD and the files will be here to re-release in better sound quality.”

I think this is the key. By common wisdom, DSD is the best digital archiving format for analog recordings and it’s the best digital recording format. DSD recordings, even when mixed in DXD, sound better than DXD (or lower resolution) recorded and mixed music

So if we consider those 3 statements true, from a sound quality standpoint there’s no reason to digitally record or archive in a different format, even when later mixed in DXD.

That should make it the go to digital format (even if differences are small for many), which just seems to be ignored by most of the big players for convenience reasons.

[andrewc]

@Rushton

Sony used the term “wide-DSD” to reference 5-bit PDM.

You mean “DSD-Wide”, and it’s actually 8-bits, not 5.

[stevensegal]

Unfortunately your three criteria should include a fourth, the ability to transform a data capture into a commercial product. It’s not that DSD is bad at this, it just can’t be done.

@jazznut

which just seems to be ignored by most of the big players for convenience reasons.

No, they ignore DSD because it can’t do the job at all. It’s a fatal weakness.

[Rushton]

@stevensegal

Unfortunately your three criteria should include a fourth, the ability to transform a data capture into a commercial product. It’s not that DSD is bad at this, it just can’t be done…

@stevensegal, I appreciate your commitment to great music recordings. But I totally disagree with your absolutist statements like the one above. Your statement is just not true. Can the large companies make the volume of recordings they wish to make at the cost they want? If their answer is “No” that’s fine. But to say “it can’t be done” without further context is just not true.

[Rushton]

Thanks, @andrewc

[dvorak]

@stevensegal

It’s not me saying “DSD is un-mixable”. Pyramix say so in their literature. So do plenty of other people.

Hi, hope you don’t mind me jumping in but I find this topic super interesting. The statement can be both true and false, depending on what exactly you mean when you say “DSD”.

It’s true that you cannot take two 1-bit DSD sample streams and combine them in 1-bit sample space. If your definition of “DSD” extends to the encoding as well as the signal content itself, then the statement holds.

But if you think about the content of the 1-bit DSD sample stream it’s nothing more than a regular plot of amplitude versus time for a bandwidth-limited waveform. Literally ANY operation that we can perform on sampled waveforms can be done on the DSD signal if we have the freedom to use more bits to hold the results of the operation. Because the waveform has been produced with sigma delta modulation we know that the SNR is great down low but there’s only noise up high, and that’s why we use terms like PDM or PWM. (And it’s only the fact of noise shaping which distinguishes these from PCM.)

Simple mixing involves linear operations: the scalar multiplication of each signal and the addition of two signals together. The things which make DSD sound like DSD aren’t changed by doing that.

It is necessary to do a final sigma-delta pass to remodulate the output back to a 1-bit format when you’re finished mixing. So the question is, if the DSD waveform shape is preserved during mixing does it really matter if the intermediate encoding has used additional bits? For me the answer is no, but if in your mind “DSD” isn’t DSD unless it’s a 1-bit format then you might stand by the “DSD is un-mixable” assertion.

[Rushton]

@dvorak

if the DSD waveform shape is preserved during mixing does it really matter if the intermediate encoding has used additional bits? For me the answer is no, but if in your mind “DSD” isn’t DSD unless it’s a 1-bit format then you might stand by the “DSD is un-mixable” assertion.

Thanks for jumping in @dvorak. Your comments are precisely to the point. I tried to add this point in my earlier post as an “Added comment added for clarity” (HERE) but my comment was not as well stated as yours.

[jazznut]

But what can’t be done if mixing a DSD recording in DXD with Pyramix? That may not be the best option but if it’s still better than DXD or lower from beginning to end, then it makes sense enough to do the recording in DSD.

[stevensegal]

@dvorak explained it very clearly.

For me and probably 99% of people who’ve heard of DSD, it’s a 1-bit high frequency data stream, as opposed to any other format that is chopped up into multi-bits, whether you call it PCM, PDM, PMF or fried bananas. Whatever it is, it ain’t one bit and the “analog” claims about 1-bit DSD fly out the window, forever lost.

Furthermore, I reckon 99% of people don’t care what format the original tracks are captured in. It was by luck that I picked the Florilegium release that included the capture format in the liner notes. It may have been captured in DSD, it was released on CD, mp3, 16/44 and 24/96 PCM.

I suspect the capture format is only the concern of the recording and mixing engineer who have to edit and convert the recording into the final product.

Personally, I couldn’t care less. I have plenty of recordings from Channel Classics and, to name another favourite, Myrios, that were probably captured in DSD, which I bought in some PCM format. They sound fabulous.

So I think it is agreed, you can’t mix 1-bit DSD, in the sense that what one person at Eudora does via HQPlayer is not what a mixing engineer would expect to be able to do, and what Pyramid say is not disingenuous.

I think we can also agree that people don’t do what Eudora sometimes do, not because they are lazy, but because it is extremely limiting, and throws out all but one of the mixing tools usually available.

Do we agree that a workflow is pretty pointless unless it can handle any program material, not just a solo piano or a string quartet? How about Meldy Gardot’s last album, for which different channels for multiple tracks were recorded in different studios around the world (vocals in Paris, orchestra in London, instrumentals in LA)?

Noque’s approach is a bit like a car with no steering wheel. It might be the best car in the world for straight roads, but any bends and it’s useless.

If you want a DSD DAW, I suggest you crowdfund one with a target of $20-$30m. $1m gets you next to nothing on software these days. PS Audio probably spent close to $1m on a basic streaming platform they abandoned.

Once that is successfully developed, you have to:

• persuade lots of people to pay for downloads that they would normally stream in a monthly plan, and pay more for each download than they pay for a month of streaming;
• buy a native-DSD capable DAC; and
• give up any DSP they use in their audio system.

It’s all perfectly logical to me.

[jazznut]

There are a few topics around DSD/PCM and once nailed to one, you seem to jump to the other Steven

The way I see it:

Most don’t care if they listen to DSD or PCM → correct, most don’t care about most things.

DSD won’t get a mass format → correct

DSD can’t be natively mixed with all functions necessary for any kind of recordings → correct

Eudora’s way of mixing DSD natively is pointless generally → wrong, it’s great for where it applies, it’s just pointless for where it doesn’t apply

DSD as a recording format is pointless → wrong, with DXD mixing it’s even suitable for the masses and all kinds of recordings and sounds better already than any PCM path

Recording formats are pointless → wrong, they are most essential because everything behind builds on their quality, they should be as good as possible today.

DSD is pointless because few have DSD capable DAC’s → wrong, because playing DXD or PCM converted DSD recordings is still better than pure DXD/PCM, even if maybe not by much for most.

[stevensegal]

I have nothing against DSD as a capture (input) format for the simple reason downloads are almost always priced on the output format. So the Floreligium recording I mentioned was recorded in DSD, but here is the pricing (bear in mind it’s a double CD release):

Of course there is no additional work in producing 24/96 PCM compared to MP3, it was just pressing another button on Pyramix. Formats are being used more as a pricing model.

The irony is that this recording is now available on Qobuz in 24/192 PCM, a higher resolution than you can actually buy the download.

I did not say Eudora’s way is pointless generally, it’s probably just impractical for 99% of commercial recordings. It requires a very specific type of music, basically a live acoustic programme, the smaller the group the better.

As to the last point, I have no issue with pure DSD64 (which I can play) or things that were captured in DSD and issued in PCM, like Rachel Podger. The latter is considered state of the art for playback quality. What seems nonsense to me is DSD256 output. Few systems can handle it, even dCS can’t, it’s difficult to transmit if you don’t want to use usb, and you can’t do DSP. Lot’s of people use DSP, especially with headphones, I use DSP for active bass management and would not want to give it up. That’s all before the premium pricing and storage requirements of the files.

It’s been widely accepted for decades that the more dynamic range in the capture, the better the final product can be, even if reduced to 16/44 PCM. Having 60+ tracks in 24/352 or 24/384 gives the engineers massive scope to optimise the finished product, but it doesn’t have to be played back in 24/384. Frankly 24/192 is probably massive overkill.

[jazznut]

@stevensegal

What seems nonsense to me is DSD256 output

Well as much nonsense and storage consuming as the high PCM sampling rates. Quality needs space. At least things start to be competitive with vinyl there (also needs a lot of space)

What seems to be common sense meanwhile is, that DSD256 makes probably little sense for most DAC’s in case the content was not unmixed or analog mixed but DXD mixed…then the DXD output format for a DSD recording seems to be more meaningful in most cases. Man…all this has to be communicated…not easy for one who mixes in DXD and wants to sell DSD… to explain why the most meaningful output format still is PCM and only the recording DSD.

I remember when Paul started to promote DSD (recording and output format) as a superior black box. It turned out to be much more complicated and differentiated the more people asked, researched and reflected themselves. Doesn’t make it worse…just harder to understand in its true correlations. This complexity and the often misleading attempt to make simplified statements makes it hard to finally sell DSD.

[stevensegal]

I think you hit the nail on the head in the last paragraph, DSD all sounded very simple at the outset (1-bit in, 1-bit out, “analog” all the way, no sampling), except that’s hardly ever the case.

I also appreciate your second paragraph that once a DSD capture is processed after conversion to DXD, converting back to DSD is pointless. Paul and Pyramix will say that DXD is only used where needed and the rest remains “pure”, but that doesn’t really cut it.

DSD256 would most likely be “pure” mainly for analog transfers, i.e. a file that has already been mixed in the original analog. It’s basically an archive file, which is what DSD was intended for in the first place. Perhaps some unedited live recordings. I’d rather the mixing engineer had the opportunity to edit a recording rather than rely entirely on the skill of the recording engineer.

But the real nonsense about DSD256 output is that so few people can actually play it, even amongst wealthy audiophiles. Many streamers and DACs can’t handle it and most interconnect protocols can’t handle it either. Plus no one streams it. That is not the case with high PCM sampling rates, you can play them on just about anything, even a mobile phone, it’s very easy to stream, the only limitation is some optical cables like Toslink are limited to 24/96. Heavens, even my ceiling speaker/lights operate wirelessly at 24/192 PCM.