SH Spotlight What sounds just like the analog master tape: CD, Vinyl, SACD or a 1:1 analog Reel tape copy?

This is from an EE/DSP/Comp Sci guy with audio/audio processing background, who doesn't want to sell anyone anything, and who really understands the stuff, not just an audio person...

Anything at 48k/16bits or better (properly implemented) can reproduce a master tape for listening purposes essentially perfectly. There will be people who argue with me that 44.1k is good enough, and I will not argue, but nowadays except for CD, why not use 48k and be done with it? Frankly, I prefer 48k/24bits or better, but that is just a psychological thing. There is NO additional 'granularity' other than imperceptable hiss at 16bits.

The problem when people do comparisons is the differences in source material. Such differences will create a mirage and bias the results.
Higher data rates/bit depths are good for certain kinds of nonlinear signal processing done for production.

I have surprised people when I produce a 44.1k/16bit CD .wav file, where they find that it sounds BETTER than a high res 96k/24bit, because I did the conversions correctly, started with the same or better material, and not out to sell anything. Note that I wrote 'same or better' material? THAT (the source material, and its precise state) is where the differences happen. I am not claiming that a Victorola with perfect material will sound as good, but as long as the transport is truly 48k/16bits or better, it is good enough *for listening*. If you have 192k/floating point or 4096DSD or whatever you want -- so be it -- it will work REALLY WELL!!!! It is nice to have that extra breathing room, but anything above *properly implemented* 48k/16bits for listening is breathing room.

John

 

I think the best CDs can sound really good. I think the best SACDs can sound noticeably better. To my ears, SACD delivers better transients, better detail at low volume levels, better timing and pacing and rhythmic stability and/or less jitter, and an overall more analog and natural sound than CD. Maybe it’s not exactly reproducing the sound of the master tape, I don’t know, because, as you point, I’ve never sat in a studio and heard a master tape played back. I have heard thousands of LPs and CDs in my lifetime, and, in my opinion, SACD is a better format than CD, and as good as or better than all but the most high end turntables with moving coil cartridges. But, once again, with SACD, you don’t have to deal with all of the hassles of finding mint LP pressings, cleaning and maintaining records, fiddling with cartridge alignment, flipping a 45 rpm record after every two or three songs, etc.

The insistence that “it’s all in the mastering, formats don’t matter” makes little sense to me. I understand the theory that 16/44.1 ought to be “good enough” for human hearing, but, to many listeners, it isn’t. The DSD/SACD approach of one-bit delta sigma recreation of the analog waveform at 2.8 million samples per second produces better results. If different sampling rates per second don’t matter, why does 30 ips analog tape sound better than 3 3/4 ips analog tape? Running more analog information per second across the tape head produces a better result. Cutting a record at double LP at 45 rpm produces a better result than cutting it as a single LP at 33 1/3 rpm. Not sure why higher bit depths or sampling rates than 16/44.1 wouldn’t also produce a better result, and that applies to a 24/96 DVD-A as much as it does to an SACD. If CD were really good enough, audiophiles wouldn’t bother with vinyl, SACD, DVD-A, Blu-Ray Audio, reel to reel, or any other niche formats.

 

I understand the theory that proclaims that this is so, but, for whatever reason, to my ears, SACD at its best sounds better than CD at its best. I wish that weren’t so, I’d save a lot of money if I didn’t enjoy SACD as much as I do.

 

It's both. The point though is that when it comes to digital formats, the source and mastering far outweigh the impact of the format. This is different from analogue.

Well you state that but efforts to prove it have fallen a bit short so far. I would not say there is no audible limitation in CD, but the evidence is that it is small and hard to hear.

I know it is difficult because intuitively bigger numbers ought to sound better. But digital just doesn't work like that. It is a matter of science and mathematics - unless of course you hear frequencies above the normal range of human hearing (some claim they make a difference, again it's been hard to prove), or play music so loud (and quiet) that you need an extraordinary dynamic range, in which case you probably live in the wilderness or get complaints from neighbours - and in any case, music is mastered to reduce its dynamic range because a super wide dynamic range is so inconvenient. Which doesn't mean I like the DR crushed to oblivion, far from it.

Audiophiles do all sorts of things but the point is, we love music and play it on whatever format it takes to hear it at its best. Which one depends because we are not in an ideal world where every production sounds as good as it possibly could.

Tim

 

Bad mastering supersedes the format. True?

 

How do you know that you have heard CD at its best? That is the problem -- unless you are comparing WRT multi-channel, most comparisons arent' fair. When I have done fair comparisons, previous 'believers' in super high rate tend to start thinking about their opinions.

John

 

BY FAR. Differences in mastering easily cloud the comparisons. Even slight differences in mastering can make a huge difference.

I do NOT do mastering, but I have written software that recovers the signal from some badly mastered CDs -- and the kinds of filters that I use for equalization would probably surprise the most experienced mastering engineers (at least, some of them.) I have various kinds of EQ methods that have the very same superficial specs, but I have certain kinds of configurations that can make improvements over and above the simpler, more common methods. There are LOTS of variations of what can be done -- and changes do seem to be made.

There are so darned many variations in what is done after the original artists/recording engineers mix, the comparisons, unless made on EXACTLY the same material, with EXACTLY the same EQ, those comparisons are questionable at best.

I have several reasons for not claiming to do mastering when I offer improved versions for friends who ask, or when I give away the software that does the work. 1) Mastering is truly a very very fine skill, and I don' thave what it takes. 2) Too often, people who claim to 'master' or 'remaster' have been encouraged to do damage by the intellectual property owners 3) much of the time, re-masters aren't good, because of the demands by the distributors. I'd suspect that most people who do remasters are REALLY competent, but the goal is too often LOUDNESS. Also, it DOES appear that FA latent EQed DolbyA encoding is a requirement. I have heard close-to-correct versions from MFSL, but better than the 'normal' kinds of mastering.(the example that I am speaking of, they appeared to turn off the HF1 band on their DolbyA)

John

 

Yes, and so does great mastering. That's why SH's stuff always sounds great to me on CD and SACD. But, case in point, on his PPM "In the Wind," I prefer the CD layer. Even though the SACD sounds excellent, it somehow loses my attention. But as I said earlier, I've long suspected that my trusty Marantz 8004, which is superb on redbook, isn't a star on SACD. Also, I'm 70, and my ears may respond better to the extra crispness of typical CDs. As someone said earlier, it's a very individual thing. That said, there has to be some objective way of saying one is better than the other. No one doubts that CD sounds better than an MP3 ripped from the same CD.

 

I haven’t. I’ve never heard a Mark Levinson CD player, or a DCS Scarlatti, or whatever the ultimate CD player ever manufactured is.

On the real world components I can afford, SACD sounds better than CD. I’m sure if I bought a Rega Saturn CD player, my CD playback would improve. I’m sure if I could afford a DCS Scarlatti, my CD playback would improve.

I don’t have access to master tapes, and I can’t afford a $30,000 CD player. So disqualify my opinion, but I’ll keep on enjoying SACD.

 

Yes, I AGREE!!! The choices aren't always made freely by the 'engineer', but instead there just might be edicts from the distributors. Doing a truly good job of mastering is an art and skill, but too often recently the mastering person is 'encouraged' to do things that aren't really nice to do.

I as I tried to mention before (my language skills get garbled at times), even the kinds of filters, patterns of EQ, etc can make very audible differences. There is an ART in doing EQ when needed, but in my not-so-humble opinion, it is best to TRY to err on the side of being careful instead of 'tuning the sound'. A little trick that I recently learned, for example... I sometimes couldn't get the 'right sound' from a single filter with Q=0.50, Q=0.707 or Q=0.8409... No Q or even in-between did exactly what I wanted. Note that I am mentioning changes of 0.75dB or 1.5dB in that range. I found that a mix of Q=0.50 and Q=0.8409 often does EXACTLY what I want... Now, I am really an EE (AnalaogEE) and DSP (really sophisticated software), but nothing prepared me for the idea that a mix of Q=0.5 and Q=sqrt(sqrt(0.50)) woud sound better than an average Q=0.707 (sqrt(0.50)), but it does. It makes a BIG difference -- why? Because there is a big difference between jamming something together and making a CD, and trying to 'correct' the CD. Beating on the audio is not usually a good thing, and hyper-compression or even latent DolbyA as is often done -- NOPE, that aint it..

Mastering is a REAL art, and I know that I am not good at it. However, I do know what is going on -- and our recordings that we love so much are being beat-up.

John

 

There is also a well-known example with rattling keys where if you listen carefully you can hear the difference between hi-res and 16/44. Not such fun to listen to but unfortunately I don't have your Credence SACD. So it is not true that you can never hear the difference. DSD vs hi-res I am more sceptical but maybe there are cases, who knows.

Tim

 

You are the one that very close describe SACD sound in the way I hear it.

For me, how SACD reproduce bass, the lower frequencies is just as important as the higher frequencies. Bass is more full body, there is more «physical weight» to bass on SACD, bass on the CD layer is «thinner» so to say. And there is no doubt in my mind what is most correct.
I think it’s a bit strange that not everyone hear it, cause for me this is so obvious.
Do you hear it?

 

I agree that bass is important. A naked stand up acoustic bass solo, like the ones Jimmy Garrison used to play a lot in Coltrane’s band, is a good test of what a system can do, and the jazz SACDs I own and have heard handle this test well.

 

In order to do the comparisons correctly, it requires careful management of the tools and control the experiments with exactly the same source material. Too often, the SACD will have slightly different mastering, and mastering is REALLY where the big difference is. For example, I have a VERY VERY good Love Over Gold CD, but also a VERY VERY good SACD with the same material. One might initially say, same material, even SOUNDS like the same mastering -- but it is PROVABLY NOT THE SAME MASTERING. I wouldn't be surprised if many golden ears might say that the SACD sounds better -- because it just might, but even though BOTH have good mastering -- SACD will most likely sound different. What is the very obvious/provable difference in mastering? Compression with the knee in a different place. (Yes, SACD stuff is VERY OFTEN compressed, esp POP, even the VERY BEST POP.)

The reason why I don't argue for 44.1k/16bits is that it is plausible that some people can tell some difference between 48k/16bits, but after that -- we are talking hiss is the big difference. When I did my mistaken eureka for my friend (wasn't planning on it), I was using exactly the same source material, with exactly the same mastering, because it came from my EXACT SAME 96k/FP source material.
starting with 44.1k/16 for processing IS problematical, but I can only demonstrate it as problematical for SIGNAL PROCESSING. I can actually eeek by with aggressive signal processing being done at 48k/FP & 48k/DP (I never do signal processing at 16bits or 24bits), only because the algorithms don't allow the distortion products and modulation sidebands reach beyond 24kHz to begin with. Those are NOT typical algorithms though. 44.1k doesn't leave enough headroom in the frequency range for even my fancy algorithms though. The only way that I'd use 48k for serious stuff in PRODUCTION is if there was no serious signal processing, other than highly qualified algorithsm, going on AT ALL.

Too often, people are knowingly or unknowingly not representing reality in their comparisons. The error is not intentional as good comparisons are tricky, and correct manipulation of audio isn't done by pulling out some DAW software, not understanding EXACTLY what is going on -- and assuming that everything is exactly the same. (For example, I don't depend on a slider anywhere when I set up a processing setting.) Even a GUI slider can kill accuracy.

John

 

SH says "Notice how the reverb vanishes much faster and is not as intense? Loss of resolution."
So question, to anyone in layman's terms . Is redbook simply unable to tell the trail from no signal (ie silence), so it decides to to treat it as silence? I guess I am asking, where does the missing trail go? Into nullity, like it never existed?

 

People who don't truly understand DSP are fooled by the resolution myth about 16bits vs 24bits. The 'resolution' is a complex thing, and isn't what most non-math/DSP people understand. Let me give you a very simple existence proof: Cell systems. They typically use 12 bit or 14 bit A/D, but how do they get the incredible sensitivity range, much larger than audio? A receiver with a good quality 'low res' A/D can receive strong and weak signals at the same time. Any idea about the signal being digital and not needing resolution is also faulty -- the digital signals on cell systems are very complex, and pack a lot of data into a relatively narrow bandwidth, but that small resolution A/D can be used to convert into data with much higher res, but how?

The answer is this: dither and bandwidth reduction (which is a specific case of a kind of selectivity.) With the dither (which is pretty much the same as audio hiss), there is the ability to detect and utilize a signal WAY BELOW the apparent 16bit or whatever resolution. Your hearing can hear a 1kHz tone below the hiss, cant it? (gain-up when needed.) Think about this for a second. With proper dither, you can get an effective 18 bit resolution (or more) out of a 16bit signal, but the higher resolution will be associated with a bandwidth limited subset of the signal. (Other ways than 'frequency' alone can also be used for SNR improvement.)

* Dont think of dithered 16bit signals as being 'resolution limited', but instead they are 'data payload limited'. A non-dithered or poorly dithered signal will certainly have unending problems with distortion and other such issues -- don't even go there.

* As a matter of 'religion', I never consider using 44.1k for anything unless for reading/creating CDs... 48kHz with well designed software can do many kinds of processing at high quality...

Where we get into trouble with 16bits is not the transmission of audio for LISTENING PURPOSES, but 44.1k/16bits is trouble for PROCESSING PURPOSES, even including for something as so very simple as panning. Some of the aspects of panning in a 16bit environment don't work simply -- when panning with a 16bit selector, in a ham-handed implementation, it can be easy to get in trouble... (I am not sure when zipper noise starts happening, but 16bits seems to be on the edge.) For certain things -- 16bits can be problematical, just as 44.1k. There are all kinds of subtle things going on that comparisons need to be done scientifically, or the ideas/results will be faulty.

44.1k/16bits is totally inadequate for many kinds of audio signal processing, signal processing includes 'modulation' (e.g. AGC, FM synth), 'gain changing' (similar to AGC), nonlinear processing (soft clipping), lots of other things like that. This is one reason, nowadays, since FP math is so easy to do -- might as well do everything internally in FP or DP form, don't worry about the complex issues associated with dithering or bandwidth increases in every place where it must be considered, or perhaps with 44.1k, where there are terrible things when a modulation sideband or distortion product happens to reach above 22.050k, then wrapping into a more audible frequency region. In these cases, it is almost silly not to use 96k (66.15kHz, if well considered) or higher for internal math, and might as well use FP (sometimes really not adequate without work-arounds) or DP. For my purposes, I have found that 66.15/72kHz are good, minimum sample rates for audio DSP, and still easily handle the distortion sidebands & resolution issues. For the internal math format, I use DP when I can, but FP when I must. (DP can make some math issues work better in extreme cases -- basically allows simplifying certain algorithms, like low frequency IIR filters.)

* If you do have to process a 44.1k/16bit signal, just up-convert it, do the processing, then convert back... With proper conversion, the only loss is the extra step(s) of dithering. With the simple conversion methods that I typically use (avoiding more sophisticated ($$$/license-with-strings) libraries), my sw does lose a very small bandwidth, but that is because I haven't invested any time in improving the algorithms -- focusing my efforts elsewhere.

Important: there is no *distortion* difference between 16bits and 24bits (unless botched HW/SW), but there is a 'relative accuracy' and 'noise' difference. (I use the term 'relative accuracy', because most of the time listeners aren't awfully worried about 'absolute accuracy'.)

BTW -- some designs don't dither a conversion of FP/DP to 24bits, but I wouldn't EVER skip dithering, even when it seems to be unneeded. Given this wonderful FP/DP world we live in nowadays, is so very easy to dither at a additive magnitude of a bit or two at 24bits. (Don't try to get by with just flipping the bits -- do a proper dither.)

John

 

John, thanks for the answer, but as a non-math/non-tech/illiterate I was wondering if I could get a totally "For Dummies" explanation. I mean, where did the end of reverb trail go in redbook? Did the processor just decide it was too low in volume to be signal, and record it as silence? I thought it might be like 720p video vs 108op, except that's not right, because even in 720p, the superfine detail does not entirely disappear as the reverb trail apparently does, its just looks slightly blurrier, though still visible in this slightly blurrier form.

 

If you have a tailoff problem, and it is due to 16bits, then someone didn't dither well. However, along with the tailoff drooping quickly, with bad dither, you would hear a bit of distortion during the decrease. Also, it might be a difference in the way that the 16bit signal was created. Dither is trickier than just adding dither -- think of this, when you have a frequency shaped dither, then do a lot of EQ -- you might preferentially lose dither, and maybe drop below the minimum to get a high quality signal. This is one reason why I don't do a dither to 16bit output on my software -- but I'll happily dither to 24bits. Since my software is focused on professional applications, then dealing with all of the dither issues at 24bits is easier. First -- be generous with the dither -- it doesn't help to get an extra 3dB of SNR at 90dB, when (for example) you lose something else. It is especially okay to spend 3dB-6dB at 24bits, the cost might be well worth it, or not. If the loss was 3dB at 75dB SNR, then I'd be stingy with the dither. You don't want to do ANY processing on a 16bit signal that is minimally dithered -- any changes might screw up the stats of the dither. So, when I do dither at 24bits, I am pretty generous with a relatively flat dither within the bandwidth, in case the signal is going to be processed in a significant way. Think of this with 16bits -- it might be dithered minimally, but dither is biased towards the HF. Then, someone does an LPF (low pass), with maybe a cutoff at 1kHz, still being 16bits on the output -- where did the dither go? SO, with such an LPF, you might want to work in higher precision (24bits or better -- I prefer at least FP and be done with it), then re-dither on output..

There are LOTS of variables and LOTS of black boxes that aren't necessarily proven or not necessarily technically trustworthy. This is why, when I don't have source code for something, then I write it from scratch, and do the algorithms very carefully -- REALLY!!!! If I don't have time to implement something correctly, then I avoid doing it. Most of the time, except for something that I can avoid, I'll do the research to do something correctly.

So, it MIGHT be valid that a given 16bit file (recording) has defects like a dropoff, or stairstep noise (distortion), but that is the fault of what created the file.

I knew/wrote DSP for many years some at AT&T Bell Labs, but until I really started writing DSP software for audio practically all of the time, all of the little subtle details weren't obvious. It took some time to fully understand these specific issues. Recently, in another forum, I gave a nice explanation why the wobbles created with a fast cutoff FIR filter isn't 'ringing', but is Gibbs effect, which is more of an indication of LOSS of components, not sustained components. The Gibbs moves around with minimum vs. linear phase NOT because of some kind of magic, but is simply because of the varying delay vs. frequency. Any difference in sound is probably more of a matter of arrival delays, not so much hearing phase. Short delay filters are less likely to cause the problem -- and is probably one reason why peopla re wary of long FIR filters -- but linear phase will work consistently, but with the longer const delay across the spectrum, instead of varying across the spectrum like minimum phase.. There is lots of lore in the audio realm -- even EEs can get tripped up in this stuff. I don't have the luxury to make mistakes -- and I am STILL LEARNING, and never stopped learning. Knowing me on the other forums where I lurk -- I am very fast with a mea culpa.

As soon as we quit learning (including me), then what is the purpose? Profiteering isn't my interest -- I want to know how it all works!!! .

John

 

I re-read this thread (as many posts as possible). It is educational and a great thread after these years. Thank you again, Steve.
@Steve Hoffman

 

I feel inclined to say reel-to-reel because the fidelity is so terrific. The last time I heard one was in 1981 when one of my grandmother's renters played "Superstition" from Innervisions for me. It sounded fantastic to my ears.

 

This is really the thank you, amen, game over post for the CD vs SACD discussion. Same source program, two formats, and a road map of exactly what to look for.

Now, of course, the source material and the mastering engineer determine whether a particular SACD will sound better than a particular CD. There are lots of SACDs that sound inferior to the very best CD version of the same program.

 

Always good to go back and read these threads. I think I’m going to give some of the Nat King Cole SACDs a spin.