Back in 2010 I clearly laid out a standardised process (subject of course, to review and improvement) to handle the preparation and presentation of experiments like the one postulated in this thread.
I am completely confused as to what we are aiming at, and considering the hourly charge rate that industry professionals like Pluto and EricW (and I) charge, I really think we've used up all the goodwill we dare. Would someone bring this experiment back on track and explain for the benefit of the average dispassionate reader, what we're trying to prove and to whom in simple language? I sincerely hope that we are not trying to prove to a single individual what common sense told me was obvious from the start, namely, that if your hearing doesn't even extend to 16kHz or so then generating audio at two or three times that frequency is just daft.
In situations like this where engineering seems to be in the driving seat I'm reminded of a conversation I had with Dudley Harwood, our founder. I proudly showed him the great lengths we went to to sort and pair match drive units. He mulled over the piles of the good, bad and the ugly and shocked me saying 'that's bad engineering practice you know'. Quizzing him he explained that the essence of good engineering was to build up to but not beyond the functionality needed for an electro-mechanical system to perform its job. Every penny of cost, complexity and effort expended beyond that point was a waste of natural resources, which included engineering and assembly labour time. In a situation like this, where we are considering precisely these issues of 'where to draw the line', I refer you to another conversation with him in which we explored the technical measurability but complete inaudibility of resonances within speaker cabinets. He was very firm about this and he told me 'If it's inaudible, it's inaudible. End of story. Design complete. Absolutely good enough regardless of the measurements'.
Would it help if I was converted these source files to high bitrate MP3 and placed them in the time honoured way as player applets in a post? That strikes me as by far the simplest way of presenting A v B comparisons which is, of course, why I designed the applet player system here on HUG.
As a point of order for the future, I really don't think we should be quite so open to publishing audibility experiments that do not make a serious effort to understand and control variables. Experiments need thinking and planning time - and that shouldn't be done on the hoof in the glare of public scrutiny. Otherwise the Harbeth User Group cannot add worthwhile value to a subject (which is the sole reason that we exist) and we are no better than hundreds of hi-fi talking shops on the internet. We can do much better than that.
Frustrated of southern England.
Alan A. Shaw
Harbeth Audio UK
If I really believed in HiRez equals better sound then I would have upgraded to HiRez DAC years ago. As I have stated, I couldn’t hear any difference using my DAC. Just plain 16/44 DAC. DSD is another story.
Whatever difference I and many more heard and hearing maybe due to other factors in their playback chain. Getting 9 out of 10 doesn’t prove a thing. Try flipping a coin hundred times and there is good chance you getting heads (or tails) 10 or more times in a row.
Did I hear the difference every time even when using the Laptop's analogue output? The answer is No. Is it possible to hear a difference using 24/96 DAC to hear a 24/96 HiRez recording and another 16/44 DAC to hear a 16/44 recording? I don't know.
Thanks for your time. It was very kind of you to go through the trouble to create the file.
What a condescending remark! That would make me….. ahh..nevermind.
My dear chap, everything in life has a cost and I'm well aware of the status of key contributors here. I have already indicated in previous postings how much time I have devoted specifically to pulling subjects back onto track. This can only be a free service if it is actually working towards the final goal. I cannot work for nothing. I have a mortgage, bills to pay. Family life has no calculatable price - work does - but you cannot contribute effectively here if you are sitting with the family watching TV: you have to isolate yourself to concentrate. That is a real opportunity cost. My accounts department advise me what my time actually costs, and this has to be factored into travel plans, show attendance, time working on newsletters, designing new speakers .... that's normal business thinking. In fact, I am obliged to contribute mostly in the evenings here to escape the 'hourly rate' issue which accounts would surely draw to my attention if they caught me typing away every time they peeked into my office with the pithy comment 'What cost code are you charging that project to ?'.
We are not a chat forum run as a leisure passtime disconnected from any brand. We are not independent. We are the communication arm of the Harbeth speaker company. We exist to sell speakers. I am, more than anyone as I have proved and will continue to prove, willing to make a supreme effort to present what I hope are logical, thought-out contributions. I cannot count the cost of preparing them, but it would be equivalent to employing a full time dedicated professional proxy. Indeed, someone recently presented himself to me as 'the solution to your on-line presence with an easy monthly retainer'. The cost was five figures/year which for the first time put a price on time.
The point is not about cost - that's inescapable - it's about value. And if the value is hidden then who benefits? This should be obvious.
Alan A. Shaw
Harbeth Audio UK
EricW: would you be kind enough to provide a summary thus far for the benefit of the jury? I'm completely lost.
Alan A. Shaw
Harbeth Audio UK
Also, in some contexts, it may make sense to overbuild, at least to a degree. If I'm flying in a jetliner, for example, I want to know that it's capable of coping with the worst possible known conditions and then some. The latter because it's always possible (even if highly unlikely) that my flight will experience conditions even worse than the worst yet known and measured. I don't mind paying a bit of a premium on my ticket to know the plane's overbuilt: it gives me peace of mind and it could be considered a form of insurance. Likewise with a speaker: I don't mind paying a bit extra (how much extra is the tricky part, I suppose) to know it's engineered some point past my ability to hear, that there's a margin of safety. It may not make a difference to my actual listening experience, but it's peace of mind if nothing else.
I suppose at the end of the day it's a judgement call.
Therefore, I though the point of the exercise of Pluto preparing high-res and CD-res files of the same material and providing them to ST was to test this hypothesis. Unfortunately, there appear to be at least two problems with ST's methodology (as far as I can tell): (1) as Pluto pointed out, ST in playing the files through his laptop's sound card subjected some (all?) of the files to additional sample-rate conversion, and (2) in sending the high-res and CD-res files in digital form to an external DAC, he used an older DAC that is limited to decoding a 16/44.1 signal, making it impossible (I think) to detect any difference between 16/44.1 and 24/96, because even the latter would be played back at 16/44.1.
The Meyer/Moran paper in post #1 is what seems to have sparked the whole thing off. It's a strange paper: it seems to provide very robust and solid evidence that the difference between a higher-res digital signal and 16/44.1 was effectively inaudible to all listeners, but then concludes with a final section saying how much "better" SACDs sound than the same recording on CD, completely negating all the experimental data and conclusions drawn earlier in the paper as far as I can tell. Very odd. But maybe it was that last section of the paper that got ST started? Only he can say for sure.
EricW - thanks for your summation, which I think hits the target.
To ST - we are fundamentally testing the audibility - or otherwise - of sample rate conversion, SRC. I'm sure you appreciate, therefore, that any additional SRC performed on the signal over and above that which lies within the parameters of the test will, at the very least, render the test outcome unreliable. If I have misunderstood or misinterpreted what you believe the purpose of this test to be, I apologise. But for the time being, I believe that EricW has, in his previous post, defined the purpose and parameters of the exercise with quite reasonable clarity.
As Alan suggested a few posts back, future such tests need to have the ground rules set in stone before commencement if they are to have any meaning beyond that of simple entertainment...not that there's anything wrong with entertainment!
Originally Posted by Pluto
If the output of the same mastering system is convertedsimultaneously to DSD and PCM (and the job is done properly), you areunlikely to be able to distinguish the two. .....
This is interesting. You may have found flaws in the procedures adapted in the convention paper 6086. What they have done there was, I quote the paper
"As previously stated, one fundamental requirement foran objective, technically valid listening comparison is that the source material which is to be compared must be completely identical and“unprocessed”—it must not be altered in level, subjected to artificialreverberation, edited or otherwise “treated.” Since such material, if it existsat all, was not available, original samples in both two-channel stereo andfive-channel surround were recorded by the authors before the start of thelistening tests. This was done with the help of instrumentalists from theUniversity of Music in Detmold (Hochschule für Musik Detmold) in the “NeueAula” concert hall, under optimal conditions and with the air conditioning system deactivated.
To avoid any influence of a mixer on the sound quality, the stereo music examples were recorded with two microphones and the surround examples with five. All the microphones had extended frequency response to 40or 50 kHz (Schoeps MK 2S, MK 4 and MK 41 capsules with CMC 6-- xt amplifiers,and Sennheiser MKH 800); one microphone was simply assigned to each playback loudspeaker. The microphones were connected to microphone preamplifiers (LakePeople F/35 II) which raised the signals to line level, then these signals were sent to the control room via 50-meter low-capacitance cables (Klotz M1 series).At that point the five analog signals were split via “Y” adapters and converted to digital, with one set of three two-channel dCS 904 units used for DSD andanother such set used for 176.4 kHz, 24-bit PCM. "
Do you think this method of capturing the sound could haveresulted in the sound difference that some perceived in the experiment? Frankly, I am bit unclear with your statementthat you can find difference at one stage but not once it is in the digital domain.
Added (26/3/12):- I will accept your challenge. Please see new thread.
Originally Posted by EricW
I did indicate the paper title clearly in the earlier post.You can get the paper 6086 from here. There is only one Paper 6086. I cannot post the PDF here because that would be infringement of copyright.
Read my post # 37. I wasn’t referring to Moran’s paper and may have escaped your attention.
A computer analogue output functions exactly as a dedicated DAC. The audio quality may be lacking but there is no difference in the principle behind those two. In any event, it often emphasized here that the soundcard of a computer is good enough for our ears. A statement that I disagree unless it is something like Asus Xonar Essence One.
To Pluto, actually you understood me correctly. I was taking one step at a time. That's the reason I titled the topic as Interim Report. Unfortunately, the findings did not sit well with Harbeth's policy.
Perhaps it's me, but as someone who has only a passing interest in hi-bit audio, I think I've missed something fundamental here. What exactly is "DSD"? For that matter, what is SACD?
I also note that every time an abbreviation is used (such as SRC) I actually have to pause and try and figure out what is meant. So maybe we should either avoid abbreviations, list them once for easy referral or list them in every post.
I'm still really confused about introducing any secondary sample rate conversions. (SRCs). Surely that's like taking a photo of a photo and trying to judge the colour of the original scene from that. Is it appreciated just what sort of data reduction is involved in a downsample? Wouldn't it be better to reduce the std and hi-bit recordings to 48k MP3 320kb meaning that with a one-step conversion we can play and compare the samples here side by side? At best it feels like we are acting as unpaid promotional agents for a third party music label, and what benefit that brings to Harbeth I cannot see. The most likely outcome is that we are dragged into denigrating a third parties product as inaudible from the bog standard CD, which I'd rather leave to individual buyers to make up their own mind about.
One thing to caution about - I may have missed this in the discussion thus far. I use PC based audio test software which talks to internal or external sound cards. Mentioned in the User Manuals is the observation that although Windows can appear to be communicating with a suitably spec'd card at, say, 96k, in fact due to internal issues in Windows OS and/or the software driver there may be a hidden sample rate conversion which is not at all flagged to the user. For this reason alone, I would strenuously caution you about any conversion or even playout of hi-bit material because you just cannot be sure of what you are playing from the sound card.
There are so many variables here that should have been firmly strapped down before take-off.
Alan A. Shaw
Harbeth Audio UK
I am aware of the issues with computer based audio. That's why I burned the original files to a CDR so that I am absolutely sure that I am hearing the real sound of 16/44 of my player. To my dismay, the PC didnt burn the audio files like how it supposed to. So another variables. See my post # 12 above.*
It seems Windows OS have their own way to interpret audio files. It is also said that Mac or Apple's OS processes the audio files differently which is said to be better than windows OS. Many swear ITunes sounds better in a Mac than a Window.*
The same thing with the embedded JW Player that Harbeth uses in HUG. It is said it employs floating or variable bit rate for the audio in Flash. So are we hearing actual 320 vs 256 or some other interpretation of the Jw Player? Recently, I also read that music over wifi is limited to 48k resolution. Though some are said to be capable of 96k.*
I'll rise to the bait and attempt to explain some of the technological background to all this, and throw in a few observations along the way - which I hope will form the basis of a stimulating discussion of the issues involved.
About 14 years ago, Sony & Philips announced the ‘next generation’ of CD technology, styled SACD. It is vitally important that everything I am about to say needs to be considered in the context of digital audio technology as it was, more or less, at the turn of the 21st century. Real-world analogue to digital conversion was limited to 48kHz, 20 bit operation, although the last two or three bits were, for all intents and purposes, triggered more by random system noise than coherent signal. 96kHz converters existed but they were curiosities, thought of more as lab. instruments to be used under carefully controlled conditions than lugged around the concert hall recording circuit.
Conventional “Red Book” CD operates with 16 bit samples at a rate of 44.1kHz, within a system known technically as Pulse Code Modulation, PCM. The standard audiophile criticism of the CD was then, as it is now, “not enough data”. This argument was usually accompanied by the inaccurate assertion that analogue media such as the ubiquitous vinyl disc offered “infinite” resolution and therefore, by inference, contained an infinite amount of data or, at the very least, a large factor increase beyond the capability of Red Book – which was responsible for the “fact” that vinyl sounded “musical” compared to the “harsh glare” of digital sound. An odd juxtaposition in view of the fact that most people whose livelihood depended on the quality of the recorded product had their hearts and minds captured by all that digital audio offered.
SACD made use of a newly-developed coding technology from Sony – Direct Stream Digital – DSD. The mathematics are complicated but let’s try a non-mathematical explanation, in which there are aspects that non-mathematicians will have to take on trust. Increasing the sample rate obviously increases the bandwidth of the system (look up Nyquist if you don’t get this). Increasing the bit-depth improves the amount of low-level detail that can be captured, up to the point where nature’s random noise starts becoming dominant. Now, trust me. If you increase the sampling rate sufficiently, you can reduce the bit-depth without significant loss of low level detail.
This principle was demonstrated by some very early Philips CD players which contained 14 bit converters operated a 4× the notional sampling rate, a technique known as oversampling. In theory, the noise performance of a 14 bit player should be a few dB worse than its 16 bit counterpart but, with oversampling, a noise performance on par with genuine 16 bit was achieved. Given that, at that time, the least significant two or three bits in the recording were most likely noise-driven, this was no great shortcoming.
DSD is the reductio ad absurdum where the sampling rate is increased sufficiently to allow adequate performance with just one bit of data – in effect, a rather high frequency square wave of varying pulse density at a sampling rate of 2.8MHz. The technique is very similar to that utilised in the wave of CD players with “single bit converters” that emerged – DSD was the application of the same principles to the wider process.
Nature never gives us something for nothing and DSD is no exception. In order to make this rather arcane process sonically acceptable, significant “dither” has to be used on replay. Dither is, put simply, noise added to the replay process to randomize the errors inherent in the conversion and thus make the end result sound far sweeter, but noisier than it otherwise would have been. Another mathematical trick is that this added noise need not all be within the range of our hearing – much of it can be placed well-beyond the range of the human auditory system, thus reducing its audible impact – a technique know as Noise Shaping. The single bit approach of DSD requires rather drastic noise shaping – so much so that the rules (“Scarlet Book”) for SACD require that players have a significant low-pass filter on their output so as not to risk problems due to the large amount of ultrasonic noise otherwise present on the output of a DSD decoder.
SACD was set to take over the world. Early players were expensive, but no more so than the first CD players and everybody know that prices would drop dramatically. SACD addressed all the perceived sonic problems of Red Book, the future was cast.
But there was a snag: concern by Sony – by then, not only an electronics innovator but also a media mogul – that SACD had the potential to give away the family jewels. The perceived wisdom at that time was that the record business had to operate an order of sound quality magnitude (or more) above the consumer format, to protect the industry’s property rights. CD had not conformed to this dictum and the industry realised that it was paying a penalty for not keeping its eye on this particular ball, so SACD was locked up as tight as the proverbial duck’s arse. In short, you could do nothing with your SACDs other than play them through the analogue output of your Sony-licensed player. By this time, people were starting to play with digital technology – cameras were appearing, computer-based manipulation of sound was starting to capture imaginations, portable players appeared and consumers were starting to imagine all kinds of digital possibilities that had, hitherto, been little more than pipe-dreams. Punters were unwilling to buy into a technology that was, by design, locked into the analogue world.
But what of DSD, supposedly a leap of magnitude in the quality of digital audio? By this time, the all-digital music production process was here. Although largely based on 16 bit PCM, tools to manipulate audio to 32 bit and greater precision were available, so while the source material and final delivery were limited to 16 bits, all kinds of operations could be performed within the digital domain with minimal loss of resolution. The trouble with DSD is that it is near-impossible to manipulate. Even today, the amount of technical kit available to manipulate DSD without converting to and from PCM remains minimal, specialised and expensive. In the meantime, along come PCM converters running at 24 bit, 192kHz. No need for elaborate noise shaping techniques, special editors or expensive licenses. In short, DSD has been well and truly eclipsed by the inevitable march of technology.
There is no absolute mathematical transform between PCM and DSD domains, so the conversion process is accompanied by a degree of degradation with each pass. No doubt this will improve over time with the availability of more intense processing and deeper algorithms but this will always be an area where 2+2 = 3.999999999999999999999999999!
How does the potential quality of SACD compare with PCM-based techniques? The opinion of most experts is that SACD approximates to 20 bit, 96KHz PCM. As I explained in the previous paragraph, an absolute answer to this question isn’t possible because there isn’t a 100% accurate transform between the two systems.
Had Sony not locked SACD up as tight as the did, DSD may well have caught on. Had they been more liberal with the licensing of the various technologies involved, other manufacturers may well have leapt into the pool and we might now be in a different digital place. But now, it appears to me that SACD and DSD are marginal technologies – fascinating perhaps but thanks to the avarice of their creator, Sony, ultimately condemned to the dustbin of history.
So, let me see if I understand this correctly. The only way to play SACDs is through the analogue output of a licenced SACD player? No direct digital output? No ability to edit the raw digital SACD file as we can with WAV (but can't with licenced-to-hell MP3?).
How the devil then can we ever make a valid comparison between the digital SACD source and the digital anything else? We can't can we. We'd always have to play-out the SACD via the analogue ports on the SACD-equipped/licenced CD player (however good, bad or indifferent the analogue outputs are) and compare that way. Surely that can't be so can it?
If that case, this entire evaluation is at a dead stop through an insurmountable, uncontrolled variable.
Alan A. Shaw
Harbeth Audio UK
- The only way to play SACDs is through the analogue output of a licenced SACD player?
More or less. The drive technology required to decode an SACD is proprietary. The basic physical structure is fairly similar to DVD with an additional physical modulation termed Pit Signal Processing. Sony will only supply PSP capable drives to licensees. Part of the decryption key is contained within an area of the disc only accessible to licensed drives; part is contained internally within the chip set so the discs themselves are highly encrypted. The license prohibits any unencrypted digital output greater than 16/48. Some recent Blu-Ray players appear to carry an unencrypted digital signal embedded within the HDMI output, which can be extracted with a box designed for the purpose. This is typically at 96kHz, reported to be 24bit on Oppo players but 16bit on Sonys. How, exactly, this squares with the bit in red above I do not know. But in any case, you are at the mercy of how well the player's chip set converts the DSD on the disc into PCM.
Recently, it has become possible to extract the raw DSD from an SACD on certain models of old Sony PS3 units but the process is tricky and not for the feint of heart. Without the right care your PS3 can become a brick.
- No ability to edit the raw digital SACD file as we can with WAV
There is some pro hardware that works entirely in the DSD domain with the appropriate options (Pyramix, some Sadie gear) and Korg offer some cheap recorders and conversion software, but that's it really. 'Ubiquitous' is hardly the word.
You're down by $1250 now!
For another fascinating perspective on this history, but one that appears to reach similar conclusions to my own, see this page. The author of this article, written in 2001, should be congratulated on his prognostication. My piece was written with the benefit of 14 years worth of hindsight!
The most interesting point, which I did not know, was that, at one time, there was a working group whose mission was to devise a specification for a player to handle both SACD & DVD-A. How different the world might have been!