Pristine Audio Favourites

[Alain Maréchal]

I would never want match richardfinegould for descriptive allusion, but I might introduce a fly in the ointment by pointing out that this site

http://pristineclassics.blogspot.co.uk/

offers some of the same recordings as Pristine Audio, at a fraction of the cost; 100% less, in fact.
It seems to have a lot of partisans, and I have downloaded some of the offerings. Naturally it does not have the scope of the Andrew Rose site, being as far as I can tell a one-man band, but it offers many lost treasures (eg Beecham's mono FNRO recording of the Franck Symphony).

[richardfinegold]

I've promised myself the full CD-set of Mr Rose's restoration of the Schnabel Beethoven piano sonatas set.

C'mon EuroLottery, you can do it

I have the Schnabel set on CD on the Pearl Label, and then the entire 32 Sonatas were offered for $5 as an mp3 through Amazon so I added it for portable listening The mp3s are a tough listen, sounding very cramped and harsh at high frequencies, and I stopped listening when Amazon threw in the second Kempff set as a free download when I ordered the CDs.
The Pearl transfers, though, are quite good. They retain a fair amount of surface noise relative to some other companies that restore recordings of that era, but I find that my ear quickly adjusts to that. It is no worse than listening to lps and having the occasional pop and click. The treble is much more natural sounding than the mp3 and there is a real set of ambience.
I had downloaded one of the Sonatas from Pristine--I think it was Op. 111 (it was saved on a hard drive of a lap top that later completely died on me)--and I thought it sounded distant and kind of boomy compared to the Pearl transfer, which may be due to the "ambient stereo" effects that Rose engages in. I had the same reaction to a Furtwangler Bruckner symphony that I purchased from Pristine when I compared it to the same performance as reissued on Tahra.
To each his own, and those that enjoy the work of Andrew Rose should by all means continue to do so, but I'm not a fan.

[Gordon]

This week's newsletter from Pristine has an interesting article [scroll down the page] on digital sound. Worth a read here:

http://campaign.r20.constantcontact.com/render?ca=9e266f27-03b6-4f0e-b22d-8ded5c5f2dbc&c=5e289290-53ff-11e3-befb-d4ae5292c4bc&ch=5ee4b6a0-53ff-11e3-bf6e-d4ae5292c4bc

[richardfinegold]

This week's newsletter from Pristine has an interesting article [scroll down the page] on digital sound. Worth a read here:

http://campaign.r20.constantcontact....e-d4ae5292c4bc

Thank you for posting that link, Gordon, because it was an interesting read. I have read the same type of content elsewhere, without having to tolerate with the biases of Rose. I have to say that after reading that bit on his website that I am even less inclined to ever sample any of his work. His assertion that 16 bit playback is just as good as 24 bit or 32 bit is just plain crazy, IMO, and it is astonishing that someone so associated with sound reproduction could hold such a position. One wonders if Rose takes time out from listening to scratchy 78s to sample a High Resolution download, such as from HD Tracks, and if so to compare it to "red book" recordings of the same. One doesn't need to be a "Golden Ears" to detect these differences.

[Gordon]

It isn't the first time that some of these points have been made in print. Some of Bob Stuart's writings are worth reading in AES. AR makes a couple of incorrect technical statements - all zeros isn't the quietest sound it is the negative peak level, the equivalent to all 1's at the other peak when a signal occupies the whole quantiser range. Silence is not all 0s, it 1 and all 0's OR 0 and all 1s, these are 1 quantum apart in straight binary. In Gray code these levels are different. Those dB numbers also need a better explanation but do give an idea of the scale differences.

By and large there is a lot of common sense even if you disagree with his conclusions. If so, then simply avoid CD!!

Did you also notice that bit from NOrman Lebrecht about classical music sales? Here it is again:

US Classical Sales Shock

Hilary Hahn's In 27 Pieces was the top-selling classical album in the US this past week. It sold all of 341 copies, a new all-time low ( I feel I know half the people who bought it).

In second spot on the Nielsen Soundscan charts is Barenboim's New Year concert from Vienna. Just 260 sales.

These are shocking stats. There is not much point in making records for so indifferent a market.

[frankwm]

There's not much 'common sense' (presumably 'Bob Stuart' is he of Meridian: I was one of their first customers, in 1977, of the 101/103 amplifier) - and it makes a nonsense of the 'premium' 24 bit downloads: especially as 'quite a few' derive from copying 16 bit CD's: something he admitted-to (if you peruse the internet; as opposed to digesting his 'newsletters') some 7 years back.

I feel sorry for anyone who 'invests' in the Schnabel Beethoven sonatas, as 'samples' he'd provided, some years back, disclosed 'noise-pumping' (the 'hiss/noise' level going up/down according to the volume-levels of each note) amongst other 'anomalies'.

So much more can be written about those transfers; I just provide a small percentage of 'competing'/accurate (AKA: 'uncompromised') transfers.

You can easily compare - but, given various 'affiliations', I doubt most (including 'reviewers') would ever admit/disclose The Truth to 'the public'.

[Bryn]

... 'samples' he'd provided, some years back [my emphasis], disclosed 'noise-pumping' (the 'hiss/noise' level going up/down according to the volume-levels of each note) amongst other 'anomalies'.

Ah, so nothing to do with those currently under discussion here, eh?.

[frankwm]

I only registered to respond to the fact that a link to one of my blogs was made.

The Schnabel sonatas 'under discussion' are 'the same'; with the possible exception they may have been further 'fiddled-with' by using 'Capstan'; you can read some recent professional critiques of that units deficiences on: http://listserv.loc.gov/listarch/arsclist.html

[mikealdren]

I may have misunderstood but I think Mr Rose's argument is fundamentally flawed.

If we take a signal and sample at 24 bits, we have a fine graduation between the different levels from the loudest to the quietest. If we sample at 16 bits, we don't lose (delete?) the quietest 8 bits, we sample in less fine increments. The process of reconstituting the analogue signal then requires more interpolation to recover the waveform.

Mike

[jayne lee wilson]

I may have misunderstood but I think Mr Rose's argument is fundamentally flawed.

If we take a signal and sample at 24 bits, we have a fine graduation between the different levels from the loudest to the quietest. If we sample at 16 bits, we don't lose (delete?) the quietest 8 bits, we sample in less fine increments. The process of reconstituting the analogue signal then requires more interpolation to recover the waveform.

Mike

Yes, that's how I always understood it, that 24-bit recordings are dynamically more subtle, not just (potentially) wider-ranging... any thoughts on this particular aspect, Gordon?

[Bryn]

Yes, that's how I always understood it, that 24-bit recordings are dynamically more subtle, not just (potentially) wider-ranging... any thoughts on this particular aspect, Gordon?

This is the basic misconception which Mr. Ross's piece treats with. Increasing the number of quantization bits 'only' increases the dynamic range available (and consequently the level of quantization error) not the gradation of the dynamics. It's not rocket science, just simple binary arithmetic.

There is plenty to be found on the subject of the mythology of bit depth with little Googling.

[Gordon]

As Bryn has already suggested, there is plenty of explanatory material available on line that will provide the basics of digital sampling and quantising processes, at least at a cursory level, as well as covering noise shaping etc. If you want a text try John Watkinson’s book.

For those that believe that 24 bits are essential to good sound let them seek out source material in that format. For those that find that noise shaped 16 bits are adequate to enjoy recorded sound there is plenty for them.

When it comes to commenting on how much or how little intervention [ref #21] should be employed in restoring older material I am not competent to say. And BTW I still think there is common sense in Rose’s article – I don’t think he’s done a good job on explaining himself.

Firstly this:

I may have misunderstood but I think Mr Rose's argument is fundamentally flawed.

If we take a signal and sample at 24 bits, we have a fine graduation between the different levels from the loudest to the quietest. If we sample at 16 bits, we don't lose (delete?) the quietest 8 bits, we sample in less fine increments.

With you so far. But we could just delete the last 8 bits and leave it at that. The 16 remaining aren't invalid. We do a better job by mapping instead so that some elements of those additional 8 bits find their way into the final 16.

The process of reconstituting the analogue signal then requires more interpolation to recover the waveform.

I don’t quite understand what this means. In the absence of over sampling the “interpolation” is done by the analogue Nyquist filter in the DAC. That filter doesn’t change with bit depth. If you allow over sampling and bit rate changing there is interpolation for the additional samples that you are manufacturing in the DAC. These are made from the given samples whose bit depth is known. The interpolator may use a greater bit depth internally to accommodate the number crunching and have a number of taps consistent with making the output bit depth at least as good as the input at all points in the reconstructed waveform. In a 24 bit system there is more “interpolation” arithmetic than in 16 simply because there are more bits and so the pipelines and memory have to be wider but the main issue is the additional number of taps needed in a 24 bit system to assure maintenance of the output precision. These issues are not trivial.

And now this:

Yes, that's how I always understood it, that 24-bit recordings are dynamically more subtle, not just (potentially) wider-ranging... any thoughts on this particular aspect, Gordon?

Why ask me!??? There isn't enough space here to cover this issue properly from an engineering perspective - which is all I can do – for one thing I’d need lots of diagrams so I will scratch the surface. Anyway I doubt that anyone is really interested is getting to the heart of it.

You asked particularly about “dynamic subtlety”. I don’t know what that is, the technical theory doesn’t deal in such terms. The following is my understanding of some aspects of bit depth; I can’t do it without some space so if you haven’t the patience or interest to see this through stop reading now. You might want to skip to the end where there is a query to intrigue you. Beacuse the posts are restricted in size this one is in 2 parts, part 2 follows directly.

Digital audio uses two independent processes in tandem: Sampling and Quantising. Bit depth is about the latter. If we assume that the audio is permitted a peak to peak range of R then an N bit PCM sample will resolve this audio to within 1 part in 1/2^N. The range R is divided equally into a large number of small quantum steps each equal to q = R/1^2N. IOW, if the audio amplitude increases or decreases by q in consecutive samples, the binary number used to record this changes by one least significant bit [LSB]. The bit depth question is: How small does q have to be, OR how large does N have to be given an R?

When an ideal sampler [one that has NO instrumental defects] samples and then quantises an audio waveform it makes a small error. The binary number generated by an ADC will cause the reconstruction of the sample at a DAC which is in error by up to + half a quantum, q but no more. The power in the error, when averaged over many samples, will tend to a value given by [q]^2/12. This error is a quantisation distortion [QD], not noise, even though some texts refer to it as "noise". When there is no signal there is no distortion, unlike an analogue path where the noise is additive and never goes away. Clearly the more bits, N, used, for a given R, the smaller this error. This QD is the only impairment that the process of quantisation causes to the sound.

It is worth looking closely at this. The QD error depends ONLY on the size of q and that is a function ONLY of N and R. NO property of the audio itself, eg its loudness, or even the sampling frequency influences it. Once N and R are decided it is a fundamental constant of digital audio. Once the SNR is calculated though, R cancels and becomes theoretically irrelevant.

Typical SNRs for digital paths are defined as the maximum signal power [rms] containable in the range R divided by this error power. Some define it as the ratio of peak to peak signal to the QD which is clearly larger by 9dB for a sinusoid. Some others simply relate the peak to peak range to the smallest quantum q. If a sufficient number of samples is averaged, which it usually is, QD is noise-like, described by well behaved temporal and spectral properties [Gaussian] that spreads it out evenly across the spectrum between the bottom and the top of the wanted “audio band”, ie DC to half the sampling frequency.

If however one considers a decaying loud chord played on a piano, say, then as it decays the level falls gradually until eventually its amplitude gets comparable to a few quanta , q. Then the QD is no longer noise like but takes on a strong harmonic structure very much more like conventional harmonic distortion. This transition from noise to distortion is quite sudden and takes place at very low signal levels when the sound gets very unpleasant. One technique of dealing with this is "dither" which is a small low level signal added to the audio to make sure that it never gets to this distortion phase. This automatically means that the “signal” isn’t just the audio. Microphone and mixer desk thermal noise is not so low that it can’t be used as a “free” dither signal when N is large enough. In this case there isn’t much point in going to higher N, except for number crunching reasons, because the LSBs are just expressing that noise. Dither signals can be synthesised so that they can be removed at the DAC. AFAIK CD does not define a dither signal which in hindsight is a pity especially when they knew about it before the spec was finalised by Sony.

So in an ideal world what we hear at a DAC is the original sound plus some QD and a small amount of dither and possibly the background thermal noise of the sources. Practical factors to do with the specific design of the ADC and DAC and also their filters etc will add their contribution, but we are setting these aside for now and only exploring the theory. DAC over sampling or bit rate conversion interpolating filters are another story; if done inappropriately these filters can make a mockery of theoretical bit depth considerations based simply on quantisation.

As far as bit depth is concerned then the bigger N is the better. That is what the technology says and so, if you consider that low QD is a good thing, you can try and convert that into “dynamically more subtle” - over to you. I only perceive more precision in larger N, the word “subtle” as I understand its meaning doesn’t fit with my conception of what the technology is doing.

BUT, unless you are happy with an incomplete description of the significance of bit depth, this issue cannot stop just there; effective bit depth [especially as Andrew Rose has described it] is not just about N or q.

Remember what we said above about QD being Gaussian and its spectrum being flat? Well, I think we’re all aware that at a ADC/DAC pair techniques are used to re-shape the spectrum of the QD - one is Sony's SBM - so that the power is not uniform. This process affects the effective value of N and employs over-sampling which we are all aware of too.

If over-sampling is used it does NOT affect the QD directly, it remains as described above. What does change is the bandwidth over which its constant power is distributed. If we over sample by a factor of 2 we also REDUCE the power density in the wanted audio band by a factor of 2. If we double it again we reduce it by another factor of 2 and so on. The more we over sample the smaller the in-band QD gets. These reductions are equivalent to an extra bit in the bit depth for each stage of over sampling; x2 - 1 bit, x4 - 2 bits etc. But the bit depth story doesn’t finish there either.

So go to part 2:

[Gordon]

Part 2:

A flat QD spectrum doesn’t match the ear’s sensitivity across the audio band – ie Fletcher-Munson. Analogue systems recognise this by using Pre- and De-Emphasis, for example, to shape the noise; PE improves the “real” SNR in a system – “real” because it does physically exist. There is no close match between the F-M curves and noise shaping in this case, so Weighting is also used to account for the ear and in this case a subjective “improvement” is added to the real SNR to recognise the fact that ears don’t hear as well at the top and bottom of the audio band, especially at low volume levels. As it happens, CD has a simple one pole PE function defined but I don’t think anyone uses it despite the fact that it will reduce the QD.

In noise shaping, as used in digital audio, it is also possible to shift the spectral properties of the QD to better match the ear. Better still is to have the QD removed as much as possible, not just shift the same power around in the audio band. Digital noise shaping now exploits over sampling to move the QD power from the wanted audio band into the super-sonic region [ie 22 to 44 kHz in CD for x2 over sampling] made available by over sampling, leaving the in band QD much reduced in power. So in this more complex case the performance of an N bit system is no longer totally defined by the textbook QD level and its in band distortion is much reduced even though only N physical bits are apparent. One can also apply the weighting factor mentioned above.

All of this is applicable to any number of bits, N, so that 20 bit and 24 bit systems can also be “improved” by around 4 bits. Similarly a 12 bit system can claim to emulate a 16 bit one. Taking this to its limits, at a sufficiently large over sampling ratio, a 1 bit system can emulate a 16 bit one. That ratio is 2^16 but add in some serious noise shaping as well and this ratio decreases. We are now well on the way towards inventing DSD.

So this is just some of the theory behind bit depth; it scratches the surface. Among the salient points are that:

Over sampling x4 itself theoretically gains 2 bits, x8 3 bits etc, and
more complex noise shaping processing does better still and another 2 bits worth are readily achievable.

This is why it is possible to claim that, other things being equal, a 16 bit system can deliver QD performance equivalent to a simple 20 bit PCM system. Anyone who refutes that claim needs to find a way of refuting the science.

The practice adds further issues which I will not attempt here, but one such is issue is the over sampling filters used in DACs and let’s not go to the 44.1 to 48 saga either.

Now to the listening: Theory predicts lower QD for straight 24 bits over straight 16. What that means is that the impairment to the source audio caused by quantisation is less and so what one hears is a closer approximation to that source, there is, as it were, less to get in the way. The unprocessed QD difference is 48 dB. Such a large difference suggests very strongly that it should be immediately audible and not just to golden ears. The predicted difference for 24 and 20 bits is 24 dB which is less but still a large number. Again it ought to be readily audible to anyone. So is it?

Experience reported on these boards claims that some people do hear it and find it subjectively obvious – some even to the extent that 16 bit sound is objectionable. Others do not perceive the subjective difference as such a large one as the theoretical numbers suggest. In fact some go as far as to say they can’t hear much of a difference at all and I am one of those. Now what is so wrong with my ears that I can’t reliably and immediately hear a 48 dB difference in QD the way others claim to do? Wouldn’t I have to have really cloth ears not to hear it? Maybe I am more tolerant of differences and find that 16 bits are adequate for a satisfactory experience in listening to recorded sound at home. Maybe I accept more readily that it is recorded and not the “real thing” which I do not have to hand as a reference.

Listening to 16 bit sound already has a low QD, well down below the audio, getting on for 50+dB in fact even for very quiet sound [ie 40 dB down on the maximum volume] so ANOTHER 48dB isn’t going to have as much impact as it would have had if the original 16 bit QD was only 10dB down on that same quiet sound and more obviously audible.

We are after understanding the potential performance differences between technologies, but also to understand why the experience of particular individuals varies. It is obvious that what matters to any one person in their own home is what actually happens there not necessarily some abstract theory; that involves a number of variables. To an individual, pedantic insistence on objective fairness doesn’t count. If people can get passionate, as they do, about small differences in this or that CD player’s quality or this or that DAC, when there is much more control of the listening environment how can we be conclusive about any comparison conducted at a distance?

So when comparing “16 bit” sound with “24 bit” in practical circumstances what are we actually hearing? To attribute any differences glibly to these labels without some clarity about what “16” and “24” actually mean is surely not right? What practical conditions offer a fair comparison between them? It is easy to decide to use the same equipment throughout, and to use it simultaneously in the same location for all listeners. That immediately suggests that we have some difficulties if individual experiences are to be exchanged on boards like this one.

If you have made it this far then thank you for your attention. I hope you found it of some value. If you didn’t please tell me why.

A final experiment and a question: set up a high quality HiFi system and arrange it to play some 24 bit music files so that a suitable listening volume level is established. Keep to that volume setting DO NOT CHANGE IT and now play a file that has ONLY the LSB of the 24 bits changing at, say, a rate of 3 kHz [where the ear is most sensitive]. IE in straight binary to the DAC, alternate every 70 samples between 1000 0000 0000 0000 0000 0000 and 1000 0000 0000 0000 0001. Do you hear anything? What might you hear? Be specific.

Now do the same but alternate between 1000 0000 0000 0000 0000 0000 and 1000 0000 0001 0000 0000. Can you hear that?

After that you could try this: Play some music you know that has a fairly constant volume level and play it at a normal level. Play it again but now cause the LSB to vary at 3 kHz leaving the others alone with the music. Can you hear that 3 KHz in the music? Now do the same with the other test. Can you hear that?

[frankwm]

I find it quite extraordinary that you deem it necessary to go into this elaborate electronic treatise - which has no bearing on the subject; but on a link that you decided was apposite.

People seeking to indicate some 'favourites' can only be left bemused...and might be interested to know how you consider the aforementioned Walter/Mahler 9 'better on the whole' than EMI's CD transfer - which it could easily be copied from (there is 'hidden code' to avoid mentioning that) - and the 'sonic splendour' Alwyn/1812 is from a transfer I have - the same as his - and the end of 1812 is a disaster in terms of modulation-distortion: here is the RMCR link where he also doesn't show any comprehension of the what 'original' meant on the Decca SXL label.

https://groups.google.com/forum/?hl=en#!searchin/rec.music.classical.recordings/alwyn$201812$20pristine/rec.music.classical.recordings/BDzaFmcYko8/BMglwf2rtw4J

The Klemperer/Brahms was copied from a cheap reissue box-set - and even got 'slated' in 'Fanfare'; - which indulges in 'vanity reviewing': you can compare the Brahms 4 here:

http://pristineclassics.blogspot.co.uk/2012/10/otto-klemperer-conducts-brahms.html

IMO, you would be doing a greater service if you were to devote your keyboard-usage to elaborating on the >diminution of resolution< caused by 'noise-removal' processing (of LP/78 source material); and whether the result of such intervention justifies the selling of much more than a badly compromised mp3 - let alone '24 bit' FLAC downloads.

[Gordon]

I find it quite extraordinary that you deem it necessary to go into this elaborate electronic treatise - which has no bearing on the subject; but on a link that you decided was apposite.

Oh dear, bad weekend was it? You’re new here aren’t you, we often go off piste. And anyway I was responding to Jayne’s question which is legitimate regardless how many characters I use. Jayne and I don’t always see eye to eye but our discourse here has always been amicable. Yours isn’t a good start.

People seeking to indicate some 'favourites' can only be left bemused...and might be interested to know how you consider the aforementioned Walter/Mahler 9 'better on the whole' than EMI's CD transfer - which it could easily be copied from (there is 'hidden code' to avoid mentioning that) –

I have both the EMI CD and AR’s version. On the whole I prefer ARs - simple. Guilty as charged your honour. Do I get community service or a life sentence? It would be nice to hear the 78s sometime. So why don’t you?

and the 'sonic splendour' Alwyn/1812 is from a transfer I have - the same as his - and the end of 1812 is a disaster in terms of modulation-distortion: here is the RMCR link https://groups.google.com/forum/?hl=en#!searchin/rec.music.classical.recordings/alwyn$201812$20pristine/rec.music.classical.recordings/BDzaFmcYko8/BMglwf2rtw4J

where he also doesn't show any comprehension of the what 'original' meant on the Decca SXL label.

I also owned that LP once upon a time. It was over modulated both in the cut and on the tape. Decca’s style was to wallop tape for a bright effect but this impressed people as a demo disc and was used as such - hence sonic splendour. If you want some fun from a recording this is the content and Decca are your people for fun. But you don’t do fun do you?

I suspect you don’t have much of a sense of humour, you certainly don’t get irony.

The Klemperer/Brahms was copied from a cheap reissue box-set - and even got 'slated' in 'Fanfare'; - which indulges in 'vanity reviewing': you can compare the Brahms 4 here:
http://pristineclassics.blogspot.co....ts-brahms.html

SO!? I enjoyed listening to those transfers for a change of aural perspective and technical curiosity if nothing else. Is that a crime? So how would you do these? Go back to the SAX’s in early EMI first matrix pressings? OR go back to the master tapes? Or what? You are good at criticism but not that constructive.

IMO, you would be doing a greater service if you were to devote your keyboard-usage to elaborating on the >diminution of resolution< caused by 'noise-removal' processing (of LP/78 source material); and whether the result of such intervention justifies the selling of much more than a badly compromised mp3 - let alone '24 bit' FLAC downloads.

Go see #11 and then perhaps you could comment on my queries there about what degree of intervention is acceptable? This place gives you the opportunity to explain your philosophy and to get some feedback on what people here might think of it. Try it sometime you’ll find it a pleasant and informative experience. You might even learn to disagree with people amicably and less grumpily.

It’s none of your business who buys what from whom. Your posts have a strong tang of sour grapes as if there is “history” between you and AR. If you can do better then do so and let the market decide. If you did, what format[s] would you offer instead? Would you even bother with this new fangled digital stuff anyway?

Have a pleasant evening. One of these might help.