A toast to PGGB, a heady brew of math and magic

[Gavin1977]

@Zaphod Beeblebrox Noticed in your signature the new inclusion of "PGGB.IO (Another way to audition PGGB, with credits towards PGGB purchase)"

 

Would you explain how this works?  I had previously downloaded the software.

 

Also, any way of undertaking volume control (e.g. resample + set volume at a defined level, say -30dB) so I can remove my preamp?

[StreamFidelity]

5 hours ago, Zaphod Beeblebrox said:

Real time processing with 134million taps and more is easily possible,

 

Thanks for the clarification. The application Low Frequency Interpolation has to do something else, otherwise a music track would not be recalculated for hours.

[blueninjasix]

10 hours ago, romaz said:

 

It's a relatively simple function but the degree of SQ improvement, to my ears, is worth the asking price of the SRC-DX.  If you don't like what you hear, you can always return the product but this product fixes what I consider to be a fundamental flaw of the DAVE.

I wonder if a 32 bit version of SRC-DX might eventually become available? Or is it limited to 24 bit by the dual BNC connectors?

[Fourlegs]

7 minutes ago, blueninjasix said:

I wonder if a 32 bit version of SRC-DX might eventually become available? Or is it limited to 24 bit by the dual BNC connectors?

It is the BNC inputs on the Dave and the spdif protocol which are limited to 24bit . . . . . . . .

[chrille]

15 hours ago, Zaphod Beeblebrox said:

Our bad, the wording will be corrected!11 minutes is not a limit, we love our classical music fans.

The answer is is certainly not 42 :)  The truth is, we don't know that yet! We will know when Amazon sends us a bill. The cost is compute plus transfer plus persistent storage plus Paypal fees. We are learning, and figuring out as we go. Running a instance in the cloud capable of running PGGB with its RAM requirements and storage and high bandwidth required to upload and download gigabyte of data can quickly add up. The main purpose of PGGB.IO is to allow those who do not have a capable PC to audition PGGB  with their own tracks without having to depend on others. Our hope is ultimately we may be able to figure out the right model where this will truly become a white-glove service capable of remastering your whole library at reasonable cost and become a true alternative to PGGB desktop.

Thanks again, while I do wish the answer was 42! if we are still "talking cents" here, I am happy to hear that there is not an 11 minute time limit involved.

And if you could work out a reasonable cost alternative that sounds even better to me for the time being that would be great.

But the random weird glitches with the only bigger than 4 gig limit testfile I have are still a problem with my players.

Why does there have to be a 4 GB track limit with PGGB?

And if so, could you recommend a player/solution that works around this  problem  with my current music players still using my old MBP.

I definitely want to be able to listen through a whole symphonic movement without any such glitches disturbing my musical enjoyment.

 

Cheers Chrille

 

[chrille]

15 hours ago, Fourlegs said:

and sometimes players can handle gapless playback but it requires to be enabled in the settings. For instance I use MPD and it was giving a two second or so gap between the parts and then I realised I had simply not enabled gapless playback in the settings menu.

Thanks for your input Nick, the problem I have experienced with the only over 4GB track split into two parts is occuring seemingly at random, and it is not  because I have not engaged gapless play.

As a mainly  classical music listener that was one of the first things I did years ago with both players on my MBP.

I have now also figured out how to play  both 24/705 and 32/768 PGGB via Pure Music. But I can´t figure out why that player insists on playing PGGB from disk instead of memory/ram as Audirvana does.

My normal hi re files load into ram.

Both my players are early versions but I have kept them for one reason, both allow me to load only proxy files not the full file size in  my iTunes library.

Audirvana dropped that VERY smart and useful function long ago, a very stupid move on their part imho.

With only Proxy files in iTunes I can keep most of my hi res files on  two  2TB each portable harddrives and play those files loaded into ram without having my internal harddrive filled.

Cheers Chrille

 

[chrille]

11 hours ago, happybob said:

I can't speak for Romaz, but in my experience (and over the years I've owned and/or heard many of the very top implementations), I've never yet heard a set of panel speakers (electrostatics, ribbons, etc.) that could match the micro or macro dynamics of a really good horn or traditional woofer/tweeter/etc. set of speakers - even if the panels had subwoofers added. Yes, the panels can cast an amazing image, but there's always been an element missing for me. I'm not saying I don't like the sound of the really great panel speakers, just that I always seem to gravitate to the really good woofer/tweeter speakers even given their crossover limitations. Maybe the only exception I've found to this is with the MBL Extreme speakers that or sort of a hybrid but pretty amazing when done right in the right room...

Hi I agree regarding macrodynamics, very loud big dynamic peaks are not what electrostats thrive at.

And yes I  too used to own a pair of huge HORN coffins which could play much louder than my current electrostats can. But like all horns I have heard including the truly big  Avantgarde Trios,they are just  a bit too colored for my "acoustic music only please" ears.

One of the best speakers I have heard with really  large scale acoustic music where the Gryphon Pendragons,but they are way beyond both my wallet and room capacities.

My compromise  electrostats both economically and dynamically sound quite nice with PGGB and I can play my PGGB  test tracks a bit louder than normal hi res.

And at 1 metre 85 centimetres, taller than me I can stand at my listening position and conduct along without having to be a "couch potato" all day.

Cheers Chrille

 

 

[chrille]

14 hours ago, hanshopf said:

Hello everyone!

 

So today I did the proof-of-concept experiment I recently suggested for PGGB - and the result was not at all what I expected!

 

Let me report directly and present the technical details afterwards: 

 

I recorded a track from Vinyl to digital in 16/44.1 and then had the file transferred to 24bit PGGB. The file was then played through Mscaler plus DAVE (hereby bypassing Mscalers upsampling stage) and compared to the 16bit file, upsampled by Mscaler. 

 

DAVE was connected to Phonitor2, as was the phono amp. Both files this way could be compared directly to the master, which obviously was the LP. 

 

After my initial scepticism towards PGGB, I was surprised to find, that the PGGB file sounded more similar to the LP than the file, which had been upsampled by Mscaler. To be more precise: the space and air around the instruments, which on the PGGB file was much better defined than through "Mscaling", was really there on the LP and not artificially added by PGGB. But what I also found was, that timbre seemed to be slightly more similar to the LP on Mscaler. I perceive a slight added brightness on PGGB, not present on the LP and through Mscaler. I have got no idea, where this could possibly come from. There may be several reasons for this. Is it possible, that the computer, on which the upsampling is done, can be of any influence in this case? 

 

Anyway, against my recent bias, I must conclude, that PGGB ist doing what it is supposed to do - concept proven, at least to my ears! To bullet proof my findings I recorded the track from vinyl as well in 24/192 and compared this file upsampled through Mscaler with the 16/44.1 file upsampled by PGGB. The absolutely stunning finding was: PGGB from 16bit source sounded still more similar to the LP master than the 24/192 file upsampled through Mscaler.

 

The only suggestion for further improvement I can see is in timbre. Mscaler seems to be more genuine to the source in this aspect. Maybe others could make similar experiments in order to either confirm or refute these findings. 

 

In the end the LP still sounded better than its digital copies. But this I think has mainly to do with the non-professional A/D converter I used. Or at least I hope this is the reason, because otherwise this would mean that digital is still not able to make lossless copies from analogue... .

 

 

Technicals:

 

The LP was played on a Technics 1200GR with SME IV arm plus Oyaide BR-12 platter mat and Audio Technica AT750SH cartridge. The signal was sent via XLR to Pro ject Phono Box RS. From here it was sent via XLR to Apogee Duet 2 A/D converter and recorded in CD-format for obvious reasons (and for those to whom it is not obvious: think about it :)).

 

The file was then upsampled to 24/705 in the PGGB cloud and played through Audirvana on a battery driven Macbook Air 2018. The 16/44.1 file was as well played with Audirvana. Both files were sent via USB to a battery driven Mscaler, the PGGB file hereby bypassing Mscalers upsampling stage. 

 

Mscaler was obviously connected to DAVE. I used the latters XLR outs to connect to Phonitor 2, because this amp has got two XLR inputs, so that the Phono Stage could be connected with the same cable to the Amp as DAVE. 

 

I listened through Focal Utopia headphones. 

 

 

 

 

 

Hmm, I suspect you still consider LP a better reference point than the actual live sound in the hall?

What  recording did you use? If it was a classical LP, there is a chance I may have it in my  still over 10 shelf meters LP collection now mainly collecting dust since I got my Mscaler. 

Cheers Chrille

[hanshopf]

4 minutes ago, chrille said:

Hmm, I suspect you still consider LP a better reference point than the actual live sound in the hall?

 

Hello, no offense intended, but how on earth would you make the live sound in the hall a reference point of comparison for different upsampling methods? I suppose you slightly missed the point of my trial. :)

 

Apart from that: there is no such thing as "the actual live sound in the hall". On every seat in the hall you have a different sound. And microphones do not "hear" what you would hear, when you are at the same point in the hall. In addition to this in the hall you listen with your eyes as well. 

[chrille]

On 6/28/2021 at 9:11 AM, edwardsean said:

PGGB: A Totally Subjective, Pan-Galactic Gargle-Blasted Review

I found out about PGGB yesterday. I just purchased a license today. ZB lets you trial for 30 days, I didn’t want 30 minutes. I wanted to buy it from the first few seconds I heard it 24 hours ago. I mustered restraint by habit of a long hard won audio discipline. We’ve all learned that things are not always as perfect as they first appear.  But, then again, some things are so good you just know from the get go. PGGB is that good.

 

Nevertheless, I waited because psychology challenges the best of our critical listening. This was going to be the only time I was completely free from any influence of, “You better like it, you bought it.” So, I waited and formed my impressions before I was committed, but all the time thinking, c’mon there’s no way I’m not going to buy this. It is that good.

 

Background: XiSRC to AuI to Roon/HQP to HQPPro and the Joys of Going Off-Line

My system is comprised of the usual suspects: Server+Euphony+customized Farad PS > Fibbr optical (LPS1.2) > Innuos Phoenix > Sablon > DAVE+SJ DC3+SynRes Atmosphere Level 3, Mundorf silver/gold cabling, Orange fuses all around.

 

I started using off-line upscaling about three years ago. There are distinct advantages to preprocessing. Firstly, you avoid all the noise of real time upsampling and all the mitigating strategies that go with it, including having to use a separate server and renderer. I think it's an elegant solution that can potentially cut down multiple noisy components and cables and so improve SQ. Secondly, you aren’t limited by compute power in the way you are by real-time processing and its latency. Thirdly, I don’t know if you’ve tried it, but strapping an Mscaler and a car battery to a Hugo2 is not a good portable solution (I haven’t tried it. I’m just saying). With off-line upscaling you could dissolve the process and bake it weightlessly into your tracks. Once the 2Go came around, this strategy made the Hugo2 a formidable portable performer. Lastly, the files could be duplicated for any and all your devices including little DAPs without any real I/O.

 

Unfortunately, at the time, the only notable SW upscaling was in real time from Roon or HQP. There was virtually no preprocessing options, and certainly nothing that could rival an Mscaler. There was a small company called Xivero that had a decent quality upsampler: XiSRC. It’s since shut down, but back then it provided me with my first results that were more improvement than more error. Next came Audio Inventory’s (AuI) more sophisticated algorithms. Much more expensive, but only slightly better. Of course I paid it, but I started hearing that Jussi of HQP was going to release a pre-processing file converter version. I got my hands on the first demos and this was it. He was bringing his prodigious talents to an offline version. Given where I was coming from the results were stunning. So was the price as it was pitched for pro audio not the consumer market.

 

I’ve corresponded with Jussi about offering a consumer version, but he did not feel there was a real advantage to going off-line for HQP users. However, I’ve spent serious sessions comparing HQP live vs HQPP and, to my ears, preprocessing files with HQPP is superior. The algorithms are the same, but using it off-line in the system made a difference. It would be wonderful if PGGB changed Jussi’s mind and he developed his own consumer version without any real-time constraints.

 

So, for the last 24 hours, minus some sleep, I’ve been comparing HQPPro files to PGGB. I do think this is a fairer fight than HQP desktop to PGGB.

 

HQPPro vs. PGGBlaster

It’s been a while since I’ve done such extensive, intensive critical listening of HQPP files. I remembered anew why I was so impressed with HQPP. Especially, rendered off-line, the lower noise floor really allows Jussi’s fine-grained work to shine through. Compare to redbook, it wonderfully increases the resolution of micro-details in both the sonic components and the spatial cues in between. HQPP really brings a healthy slice of Mscaler-like improvements into files you can take anywhere with you.

 

I have to say though, going from HQPP to PGGB is a bigger jump than from redbook to HQPP. The experience went like this. At first I noticed how much more planted vocalists were in front of me. They held center sound-stage in a more definite and confident way. In fact all the performers and instruments gained substance, definiteness. This struck me so immediately that I first thought that the soundstage had closed in. It wasn’t the space that had been reduced but rather that hazy, diffuse quality that can feel like space. PGGB populated the environment with notes of greater solidity. This might feel more cramped except that, as there was more of the players, there was also more of the stage. Once I was immersed into the venue it was a joyous thing to open my eye/ears and see/hear that every dimension had expanded around me. It is one of the great pleasures of upgrading audio to hear both more close-up detail—and—more space simultaneously.

 

It is these kind of “impossible” improvements that prove, at least to me, that I’ve made a genuine upgrade. The picture PGGB creates is both more dense—and—more spacious, more clear—and—more smooth. You can use all sorts of algorithmic tricks to make audio sound more full/smooth, but sonic shortcuts will also reveal muddiness under scrutiny. You can also artificially boost a sense of space but are left with an image stretched thin and hollow. Moreover, distortion/noise can alternatively excite a signal, masquerading as clarity, or it can mask the signal making it sound thicker. In due time, both effects reveal themselves as unnatural through a telltale fatigue.

 

The only way you can create a sound that is truly clearer and denser and smoother and more precise and more spacious is by actually improving SQ. That is what PGGB has done, and why, for my part, I believe it is a real advancement. The most efficient way to describe the results of PGGB is that its improvements are organic and holistic. It improve sound all the way around: staging, texture, timbre, timing, etc. etc. Build a better waveform and the sonic world will beat a path to your door in perfect rhythm.

 

 

The Answer to the Ultimate Question of Life, the Universe, and Everything (Or, at least why I like PGGB for, you know, listening to music)

I kept thinking to myself, “This is why people like vinyl,” though I will always prefer the benefits of digital. It brings me back to why I listen to music in the first place. There is an entire academic discipline dedicated to the philosophy of music. What exactly is music and why is it so important to people anyway? The most illuminating answer I’ve found comes from the work of Peter Kivy. He poses something called “contour theory.” Without getting into the weeds, it states that music is able to replicate the “shapes” of life in a tangible manner. A rapturous arpeggio of rapidly ascending notes doesn’t just sound like elevation and release. You feel its contour palpably through sound. The visual arts are just as powerful, but they have no ability to press upon us with volume. I don't mean "volume" as measured by dB, but in the way real things have mass and dimension. Music can reproduce the geometry of our experiences of life in ways that actually move the air around us and is, in turn, able to move us, literally and emotionally.

 

In order to do this you can’t just produce a clear window into music. You have to enter in and sculpt the sound in such a way that gives substance to its shapes. Some designs seek to accomplish this by, what feels like, a sonic thickener spread across the spectrum. PGGB’s success is in how effectively it is able to surgically carve out sonic components. The parts gain weight as the space between them is removed, all the while preserving minor details. The vibrations of voices, reeds, and rosin resonate more viscerally into a space that better conveys their reflections. This is what makes music real, that is, life-like, in the way that Kivy muses that music is like life itself. By addressing the problems of audio, at the waveform level, PGGB better reproduces the contours of music which reproduces the contours of life.

 

To give all credit to where it’s due, ZB is quick to point out how others pioneered the field, but all credit to where it’s due, PGGB is cutting one gorgeous path forward. As someone who gets to journey in these audio trails, I’m grateful to everybody. Since 24 hours ago though, I really want to thank ZB, and Roy, Rajiv, and Ray (or was it 42 hours ago?). Does your name have to start with an ‘R’ to achieve audio greatness? Another of life’s big questions. 

 

Don’t Panic-TL:DR:

PGGB makes everything sound more real, and that is real good.

 

 

 

 

 Very interesting review and thanks for alerting me to Kivy.

It seems he has even dared to write a book titled "De Gustibus" among many other interesting titles.

I´ll definitely have to find some of his works once my nearest Univerity library is fully open again.

Meanwhile I carry within two quotes from Beethoven:

"Vom Herzen möge es wieder zu Herzen gehen" and, "Musik ist höhere Offenbahrung als alle Weisheit und Philosophie."

Cheers Chrille

[hanshopf]

14 hours ago, hanshopf said:

14 hours ago, Zaphod Beeblebrox said:

If you are able to download and try PGGB, there are a few settings I can suggest you to try. If that is not feasible,   you can email me the track (share a link via Dropbox or wetransfer) and I will be happy to provide you the same file with 4 different settings to see what you think.

 

Very much appreciated. I will send you the file tomorrow!

 

 

Thank you, Zaphod, for the files!! It's astonishingly obvious: 32bit without noise shaping is the way to go via Mscaler. It's so much better than 24bit or 32bit noise shaped, that there is no doubt at all. 

 

The noise shaper in Mscaler obviously works extremely well. Does noise shaper inside DAVE work identically? In that case Mscaler may work like a shield between the source and DAVE similarly to inserting SRC-DX between the source and DAVE.

 

I find no need for experimenting with different settings in PGGB. I believe the way I hear it now is optimal. I cannot wait to get the new Macbook Pro in November to transfer my collection to PGGB. 

 

Thank you so much for your amazing work!

 

P.S. Maybe Rob Watts tried PGGB 24bit or 32bit noise shaped into Mscaler or DAVE and therefore was not convinced. He should definitely try again with 32bit without noise shaping into Mscaler!

[Fourlegs]

7 minutes ago, hanshopf said:

 

 

 

Thank you, Zaphod, for the files!! It's astonishingly obvious: 32bit without noise shaping is the way to go via Mscaler. It's so much better than 24bit or 32bit noise shaped, that there is no doubt at all. 

 

The noise shaper in Mscaler obviously works extremely well. Does noise shaper inside DAVE work identically? In that case Mscaler may work like a shield between the source and DAVE similarly to inserting SRC-DX between the source and DAVE.

 

I find no need for experimenting with different settings in PGGB. I believe the way I hear it now is optimal. I cannot wait to get the new Macbook Pro in November to transfer my collection to PGGB. 

 

Thank you so much for your amazing work!

 

P.S. Maybe Rob Watts tried PGGB 24bit or 32bit noise shaped into Mscaler or DAVE and therefore was not convinced. He should definitely try again with 32bit without noise shaping into Mscaler!

 

Ah!!! Success!!!

 

Now you have set me a new listening challenge -

32bit no noise shaping to Mscaler on pass through versus 24bit with noise shaping to SRC-DX

 

Happy days.

 

By the way, I use a DC4 to power my Mscaler because I could not get on with (did not like) battery power to it.

[chrille]

25 minutes ago, hanshopf said:

 

Hello, no offense intended, but how on earth would you make the live sound in the hall a reference point of comparison for different upsampling methods? I suppose you slightly missed the point of my trial. :)

 

Apart from that: there is no such thing as "the actual live sound in the hall". On every seat in the hall you have a different sound. And microphones do not "hear" what you would hear, when you are at the same point in the hall. In addition to this in the hall you listen with your eyes as well. 

Hmm I guess I did miss your point a bit, but since you mentioned that you do work with live classical  music recordings I thought you might want to use as reference point how close to live PGGB or any other upsampling or recording method can get rather than using an LP as starting point  and adding yet another ADC into the equation?

To me your test reminds me a bit of " crossing the brook to fetch water".

 

I am  of course ,like you fully aware that the sound varies a LOT  depending on both where you are and how you record things:

But at the end of the day and in my book the goal of HIFI should be to get as close as possible to how real acoustic unamplifed instruments and the human voice generally/normally  sound in a good hall.

 

Lots of variables involved absolutely, but still a violin should  ideally sound as close to a real violin as possible shouldn´t it?

Ok different violins sound different depending on a lot of things, make, player, steel or gut strings,repertoire hall and what not, and some modern composers even seem to compose against the instrument instead of for it! "Qoute Isabelle Faust" but they also all sound like violins not violas or cellos.

Until SACD and hi res digital entered the scene digital often  made it difficult to distinguish between them in densely scored passages.

Regarding LP versus digital, violin is certainly one of the instruments digital has struggled most to reproduce close to how it really sounds under shall we say reasonably  natural live circumstances is the violin imho.

And yes although Mscaler  and now PGGB have improved things considerably compared to cd ,I can sometimes also feel that a good old LP played on a good system captures violin sound maybe even better in some respects than our digital toys can.

No sampling rates or bit numbers involved, no added digital artifacts or digital filtering, just what a good tape recorder or even better imho, a direct cut LP captured.

PS Sorry for asking once again but what LP did you use in your comparison ?

I am genuinely curious.

Finally  when you mention that the LP  sounded better than BOTH digital versions I still assume your reference point,  when saying so relates to how as a "trained listener" you consciously or not, know  live instruments sound compared to recorded?

Or what else could make you prefer the LP?

Mit Freundlichen Grüssen Chrille

 

[jeti]

22 minutes ago, Fourlegs said:

 

Ah!!! Success!!!

 

Now you have set me a new listening challenge -

32bit no noise shaping to Mscaler on pass through versus 24bit with noise shaping to SRC-DX

 

Happy days.

 

By the way, I use a DC4 to power my Mscaler because I could not get on with (did not like) battery power to it.

Let us know how it goes; my trial period is over. 

[Zaphod Beeblebrox]

6 hours ago, Gavin1977 said:

@Zaphod Beeblebrox Noticed in your signature the new inclusion of "PGGB.IO (Another way to audition PGGB, with credits towards PGGB purchase)"

 

Would you explain how this works?  I had previously downloaded the software.

 

Also, any way of undertaking volume control (e.g. resample + set volume at a defined level, say -30dB) so I can remove my preamp?

If you downloaded the trial version and are able to run it, it is still the best and most flexible option as it is free :)

 

PGGB.IO is for those who do not have a capable PC/Mac/Ubuntu box to trial PGGB desktop application. Right now PGGB.IO is not a substitute to PGGB desktop,  but an alternate way to audition PGGB with a few of your own tracks. It is quite simple to use; create a login ->upload a few reference tracks you have -> specify the output sample rate and bit depth, wait for a few minutes -> download the remaster tracks ->let us know what you think and how we can make the experience better.

 

Regarding volume control, if you have the PGGB desktop application, please refer the configuration section of the manual: PGGB - Zaphod's Guide (remastero.com). It is as simple as moving the gain slider to -30dB. But if you are enquiring if the gain can be set in PGGB.IO, not currently, we can add it if we see more such requests.

 

 

[Zaphod Beeblebrox]

6 hours ago, StreamFidelity said:

 

Thanks for the clarification. The application Low Frequency Interpolation has to do something else, otherwise a music track would not be recalculated for hours.

I looked at the site, it is hard to understand and most of it was 'lost in translation', from what I can glean perhaps a sinc based interpolator is being used but if it is implemented as a simple convolution, it could take for EVER.

[Zaphod Beeblebrox]

2 hours ago, hanshopf said:

 

 

 

Thank you, Zaphod, for the files!! It's astonishingly obvious: 32bit without noise shaping is the way to go via Mscaler. It's so much better than 24bit or 32bit noise shaped, that there is no doubt at all. 

 

The noise shaper in Mscaler obviously works extremely well. Does noise shaper inside DAVE work identically? In that case Mscaler may work like a shield between the source and DAVE similarly to inserting SRC-DX between the source and DAVE.

 

I find no need for experimenting with different settings in PGGB. I believe the way I hear it now is optimal. I cannot wait to get the new Macbook Pro in November to transfer my collection to PGGB. 

 

Thank you so much for your amazing work!

 

P.S. Maybe Rob Watts tried PGGB 24bit or 32bit noise shaped into Mscaler or DAVE and therefore was not convinced. He should definitely try again with 32bit without noise shaping into Mscaler!

You are welcome. Based on your prior comments, I expected 32 bit without noise shaping to sound better when played through MScaler, thanks for confirming.

 

DAVE does not noise shape at the input, it noise shapes at a much higher rate before the pulse arrays. Noise shaping is most effective only at higher sample rates (like 16FS or more), so it makes sense for MScaler to noise shape after MScaling and send it out at 24bits 16FS via DBNC to DAVE. Noise shaping is done when there is requantization involved. I.e, when you compute at a higher bit depth (on a computer or within Mscaler in your case), typically 64bits, but then you need to reduce the bit depth to 24 or 32 (because 64 bits is not possible via BNC)

 

Purely based on Math, 32bit noise shaped track direct to DAVE via USB 'should' sound better than truncated 32bit  track to MScaler. Have you tried the 32bit noise shaped version of the LP track I sent direct to DAVE's USB? If yes, it will help to know in what way it was different? Some have reported a slightly 'softer' transient when 32bit 16FS is played direct via DAVE's USB,  while I did not notice it myself (and in fact found it hard to believe), after listening to 16FS 24 bit tracks via SRC-DX to DAVE's USB, when I go back to direct to DEVE's USB, I feel the same way now.

 

Rob Watts tried a few PGGB tracks last December and they were noise shaped to 32bits, but beyond that I do not know how he listened to them. 

 

 

[Zaphod Beeblebrox]

5 hours ago, blueninjasix said:

wonder if a 32 bit version of SRC-DX might eventually become available? Or is it limited to 24 bit by the dual BNC connectors?

It is limited by 24bits into dual BNC, so a 32 bit version is not of much use.

[hanshopf]

5 hours ago, Zaphod Beeblebrox said:

Purely based on Math, 32bit noise shaped track direct to DAVE via USB 'should' sound better than truncated 32bit  track to MScaler. Have you tried the 32bit noise shaped version of the LP track I sent direct to DAVE's USB? If yes, it will help to know in what way it was different? Some have reported a slightly 'softer' transient when 32bit 16FS is played direct via DAVE's USB,  while I did not notice it myself (and in fact found it hard to believe), after listening to 16FS 24 bit tracks via SRC-DX to DAVE's USB, when I go back to direct to DEVE's USB, I feel the same way now.

 

Yes, I tried the 32bit noise shaped version as well. I found it very slightly less "clean" and sharper. I wouldn't have said it has "softer" transients, because it was not only slightly less brilliant, but also slightly harder. But then this may very well be the result of softer transients.

[Miska]

18 hours ago, Zaphod Beeblebrox said:

Real time processing with 134million taps and more is easily possible, I have benchmarked PGGB-RT SDK with up to 1B taps in Realtime. Though for most streamed audio 256M taps will be plenty and if done right, after a few seconds of startup delay (5-15 seconds), gapless playback is possible too. I was able to do this on my Mac M1 pro.

 

By definition your startup delay (lead-in) is half of the filter length if the filter is linear phase. Otherwise you are truncating your transient/step-response.

 

1B taps at 768 kHz means having 651.4 seconds lead-in delay. And 256M taps is 166.67 seconds lead-in delay.

 

Try with a 44.1 kHz track that has dirac pulse, or square step, as second sample and another one at the sample before the last one. First and last sample being 0. Then look at resulting time and frequency domain responses. If your filter pre-ringing is truncated, you are truncating your transient responses.

 

For example try this file: https://www.signalyst.eu/test/dirac2.wav

 

[Zaphod Beeblebrox]

33 minutes ago, Miska said:

 

By definition your startup delay (lead-in) is half of the filter length if the filter is linear phase. Otherwise you are truncating your transient/step-response.

 

1B taps at 768 kHz means having 651.4 seconds lead-in delay. And 256M taps is 166.67 seconds lead-in delay.

 

Try with a 44.1 kHz track that has dirac pulse, or square step, as second sample and another one at the sample before the last one. First and last sample being 0. Then look at resulting time and frequency domain responses. If your filter pre-ringing is truncated, you are truncating your transient responses.

 

For example try this file: https://www.signalyst.eu/test/dirac2.wav

 

Thanks for your thoughts.

 

When I said 'delay' I did not mean lag in terms of samples. Streamed tracks can still be pre-cached (on disk or on RAM),. When you hit play, if PGGB-RT completes processing the whole track within the first 15 seconds, and loads it into output  buffer, all the user sees is a 15 (or whatever time it took to process the whole track) second delay before the music starts to play. While the music is playing, the next track will be processed and if the track is processed before the first track finishes playing, then it too is ready to be played and the user never sees a gap. So when I said real-time processing, that is what I meant. True, It is not strict real-time when the tracks are cached and played, but many Audio players already do this like TAS on Extreme and I think  Euphony too. Same is true if the files are already on disk and not just for streaming. When files are on disk, there is absolutely no sense in having to treat them like they are streamed in real-time. They can be processed and played.

 

But yes, what you say is true if PGGB receives samples in real-time say from Roon like what HQP does, where you do not have access to the full track when the user hits play, then the lag will be at least half of the filter length. I would never want to implement that way as it will lead to a bad user experience. Who wants to wait around for minutes and then wait again if the samplerate changes.

 

[Zaphod Beeblebrox]

If you are referring to not throwing away and keeping the first 1/2 filter length worth of samples (after convolution) and just treat the successive tracks as one continuous stream, I do not like that approach either and I had already explained why in an earlier discussion. This would result in using samples form disjoint points in time and is not optimal either. I prefer to process one full track at a time, just the way PGGB desktop does.

[Zaphod Beeblebrox]

1 hour ago, hanshopf said:

 

Yes, I tried the 32bit noise shaped version as well. I found it very slightly less "clean" and sharper. I wouldn't have said it has "softer" transients, because it was not only slightly less brilliant, but also slightly harder. But then this may very well be the result of softer transients.

We hear differently, and the terms we use to describe sound are not standardized either, so it is possible we are saying the same thing.  Hopefully in the future if you get a chance to try SRC-DX, I would be interested in hearing your thoughts on how it compares to Mscaler's BNC.

[Miska]

21 minutes ago, Zaphod Beeblebrox said:

If you are referring to not throwing away and keeping the first 1/2 filter length worth of samples (after convolution) and just treat the successive tracks as one continuous stream, I do not like that approach either and I had already explained why in an earlier discussion. This would result in using samples form disjoint points in time and is not optimal either. I prefer to process one full track at a time, just the way PGGB desktop does.

 

That is what should happen if you process one track at the time. It should become half filter length longer at the beginning and end. Otherwise you are truncating transients that happen before and after half filter length to the track.

 

If you process entire album at once, then only beginning of first track is half filter length longer and end of last track is half filter length longer. But then you need to accept that transient response gets broken for half filter length if you seek, or begin playing anywhere else than beginning of the first track.

 

For typical filters this time window would be only couple of milliseconds.

 

[Zaphod Beeblebrox]

29 minutes ago, Miska said:

 

That is what should happen if you process one track at the time. It should become half filter length longer at the beginning and end. Otherwise you are truncating transients that happen before and after half filter length to the track.

 

If you process entire album at once, then only beginning of first track is half filter length longer and end of last track is half filter length longer. But then you need to accept that transient response gets broken for half filter length if you seek, or begin playing anywhere else than beginning of the first track.

 

 

If you process the whole album as one then you are assuming all the tracks are truly continuous in time which is rarely true and we don't even know if it is an album or a playlist with unrelated material. Then you would have to accept that you are reconstructing your new samples borrowing samples from unrelated material when you are near the end of a track or beginning of the next one, that could not be possibly good either. I have tried both approaches, and settled on processing on a per track basis based on both subjective listening tests and a over all better user experience.

 

Quote

For typical filters this time window would be only couple of milliseconds.

 

This thread wont exists if everyone is happy to just settle for 'typical' filter 😏