 
|
Audio Asylum Thread Printer Get a view of an entire thread on one page |
For Sale Ads |
|
|
Audio Asylum Thread Printer Get a view of an entire thread on one page |
For Sale Ads |
81.19.57.162
1. All original 16bit tapes were found, and all of these digital tapes were used recently at British Grove Studios, London -- for the 20th Anniversary edition 5.1 mix -- using the original Sony DASH 3324 machine's analog-outs;2. A separate analog backup-set of the original BIA production was not found;
3. All mixing was done in Nuendo at 96/24.
Follow Ups:
The Sony 33xx DASH machine was the same 'type' of machine as the original used during the Montserrat sessions back in in the 1980s. I didn't ask whether it was exactly the same physical unit (but the studio in London certainly used a unit which could decode the original 24-track DASH tapes).
If the original multitrack machine was in Montserrat, chances are it's still there - buried under a great big pile of volcanic ash :)
I now know the machine they used in London was a 3348. It can read 3324 tapes.
It's nice to see some facts rather than speculation on this topic.And you are right that if the recording had emphasis on, then converting to analog prior to redigitising would be a good idea.
Going back to your original suggestion to converting 44.1 to 176.4 and mixing down to 88.2, the double resampling is a bad idea. ALL resampling generate artefacts, even those that are integer multiples (although arguably not as bad as say 44.1 to 48 which is like a worst case scenario). And doing it twice means you'll incur double the quantization errors from digital filtering.
A much better approach would be to convert to 88.2 and mix entirely in 88.2. However, many DVD-Audio players are not optimised to play back material at this sampling rate, because their clocks are optimised for multiples of 48. I always use the soundtrack to A.I. whenever I'm reviewing DVD players because it has a 88.2 track, and it is a good benchmark for how well a DVD-Audio player handles sampling rates that are not multiples of 48. Lara Fabian's voice in "For Always" tend to sound sibilant and lack body unless the machine is optimised for 44.1 - so far, of all the players I've tested, only one or two was able to play this track well.
So, in the end, I think Ainlay did the right thing - using the DASH to convert to analog and redigitising to 96/24 in my humble opinion will give the best results for the majority of consumers. So I think we can rest assured that this version of BIA is as good as it can be.
Not that I would know, because if it's on DualDisc, I don't plan to buy it.
> > Not that I would know, because if it's on DualDisc, I don't plan to buy it. < <But you could of course get the BIA SACD (but you'll of course be incurring an extra conversion stage from the 96kHz PCM master to DSD ;-)
By the way, you talk of "digital filtering" to get from, say 176.4 to 88.2 (or from 96 to 48 for that matter). But tell me, what's wrong or complicated with simply chopping every other sample word, and then halving the sample clock rate? Surely that's what DVD players do all the time if a flag is set to downsample at the digital output from 96kHz to 48Khz?
e.g. I've heared the results of playing a 96kHz DAD to output from my player into my pre-pro via the SP-DIF at 48kHz [if set the 96 => 48 option]. And no problems sonically (i.e. no artefacts).
"By the way, you talk of "digital filtering" to get from, say 176.4 to 88.2 (or from 96 to 48 for that matter). But tell me, what's wrong or complicated with simply chopping every other sample word, and then halving the sample clock rate?"Maybe nothing, if you started with 44.1k, and didn't introduce any new frequencies (just simple mixing). Why not just stay at 44.1k, though? All you are doing is wasting disk space.
"Surely that's what DVD players do all the time if a flag is set to downsample at the digital output from 96kHz to 48Khz?"
Probably. That doesn't make it right, though.
> > Why not just stay at 44.1k, though? All you are doing is wasting disk space. < <Not really. By increasing the "work-in-progress" resolution in BOTH axes, it makes for much better data interpolating possibilities, DSP operations and mixing etc. And with much smaller rounding errors.
An analogy: It gives the "artist" a bigger "canvas" on which to paint his picture with greater detail. And when this picture is subsequently scaled back down to, say, to fit onto an A4-sized magazine front cover, it looks so much more detailed compared with if he had originally painted it on an A4 sized canvas.
"Not really. By increasing the "work-in-progress" resolution in BOTH axes, it makes for much better data interpolating possibilities, DSP operations and mixing etc. And with much smaller rounding errors."Not really. It depends on exactly what you are doing. And please remember the context. You were talking about downsampling without filtering, after upsampling. That assumes that there is nothing to filter out, which implies that there was no reason to upsample in the first place. Either you need to filter when you downsample, or you didn't need to upsample.
The point of my aside argument about halving or doubling the sample rate (i.e. evenly) was to illustrate the undeniable benefits of 'even', versus the destructive effects of 'non-even' sample-rate multiples. People keep overlooking that, thinking Nyquist digital filtering will solve anything. It won't.
All you've "illustrated" is that you don't understand sample rate conversion.
One-liners are so easy to write.
Quote from Section 5.4 (Sample Rate Conversion) of the Dolby 5.1 Multichannel Music Production Guidelines:"The converse [upsampling digital masters], however, is discouraged. That is, because a number of 44.1-kHz and 48-kHz sampled digital masters currently exist, it is tempting to zero pad odd samples and integer upsample to be able to claim 88.2-kHz/176.4-kHz or 96-kHz/192-kHz output, respectively. Because the resultant audio stream does not contain additional audio data and is not of higher quality then the original, it is recommended that the music simply be released in its original sample rate (fs) and bit depth."
*** But you could of course get the BIA SACD (but you'll of course be incurring an extra conversion stage from the 96kHz PCM master to DSD ;-) ***Yeah, I was considering it. But to be honest, it's not one of my favourite titles, so I can live without it. I have it on LP (original pressing) and it sounds great!
*** But tell me, what's wrong or complicated with simply chopping every other sample word, and then halving the sample clock rate? ***
In one word, "aliasing".
> > In one word, "aliasing". < <Aliasing? But ONLY in terms of a reduction in resolution.
One shouldn’t overlook a very obvious analogy: Take a straightforward Windows photo bitmap (.BMP).Now, using a very simple bitmap matrix editor, reduce the resolution by an uneven factor of say, 1.7. Now look at the picture: The result looks just looks fuzzy, and tainted by moire-fringes etc. (i.e. basically hacked to pieces and crap). In short: this is "aliasing".
Moreover, because of this aliasing, I would never use such a picture in aviation magazine I work for.
Now take the same large original image, and reduce the size (and hence resolution) by EXACTLY half. Of course the resultant picture looks smaller and some detail is lost, but it STILL looks clear and "right". And sure, I could happily use such a picture in my magazine if, say, the original file was too large in megabytes.
Conversely, if I DOUBLE the size of the image (i.e. scale it by EXACTLY 200%), and if also engage the "scale smooth" function, then it vector interpolates into the added data points. And the result "looks" good. No obvious artefacts. Funnily enough, if I then EXACTLY halve it again, it seems to leave me with an image which looks every bit as good as the original. Again, no obvious artefacts.
Well this applies if you were to depict part of a PCM wave section as a visual bitmap (indeed, I've seen this on digital wave-capture scopes on my computer which can do this easily). Now, if I exactly halve the resolution (i.e. by parsing through it to simply take out every other data column), then the new wave file still looks and sounds as "good" as it can possibly be AT THE NEW RESULTANT LOWER RESOLUTION.
In contrast, if I DON'T scale the wave by exactly even multiples (either up or down), then it NIETHER looks good on the scope matrix, NOR does it sound good when played back.
Anyway, the proof is in the pudding. I have the R.E.M. greatest hits DVD-A. The stereo Group 2 tracks are interpolated & upsampled by factor of four to 192kHz. Now if I tell my player to downsample to 48kHz, it still sounds as good as the corresponding tracks on the "Automatic For The People" DVD-A (which is presented in native 48kHz).
Now, coming back to BIA in 5.1. I don’t think all this digital aliasing stuff matters too much, since Ainlay & Knopfler took the direct analog outs from the DASH. So, I agree with you, Christine, that Ainlay took the very best approach possible given the material he had to work with. And the end result is a superb 5.1 mix of this classic album.
Here's a simple example of aliasing. Say you have a 96Khz sample rate and a 30Khz tone. If you downsample to 48Khz, you should get nothing, since 30Khz is out of range. However, if you downsample by throwing away every other sample, what you get is an 18Khz tone. Totally wrong.
Indeed -- your theory would work exactly as you described if a perfect continuous sinuoidal 30kHz tone (i.e. above the new Nyquist folding frequency) was the input resulting in a continuous 18kHz alias frequency output.But I don’t relax at home by listening to 30kHz sine waves, actually. Do you?
In any case, at low resolutions PCM has shortcomings EVEN when all the rules of Nyquist are perfectly obeyed. That’s precisely why we acknowledge that hirez does sound better than Redbook 44.1kHz across the human audio bandwidth (i.e. up to 20kHz).
But of course there is a trade-off: There are benefits of Nyquist anti-alias filtering prior to sampling (as you state), but there are negative side effects too. Indeed, it has even been suggested that above 88kHz, music actually sounds better and cleaner when no Nyquist filter is applied.
Likewise, the greater the process bit resolution, the less is the need for noise-shaping — since the aural significance of the LSB tends to zero with higher bit depth.
In short, theory and practice are different animals. What I have already described about downsampling is what I have experienced myself. As a result, 'uneven' sample rate conversion does sound worse to my ears -- with music -- than an 'even' one, filtering considerations notwithstanding, IMHO.
If you don't care about accuracy, you can take all sorts of shortcuts.
Absolutely -- People who don't acknowledge that non-integer sample rate conversion shortcuts are destructive, don't care about accuracy.
.
NT
.
.
Engaging in a little speculation, Martin? ;-)As long as we're speculating - it's more likely that the A-D conversion was done with a sound card.
I merely stated what his inventory was. It would not have been impossible for him to fly them over for the session. (I'm not saying he did, though.)
This post is made possible by the generous support of people like you and our sponsors:
Top of Page ]
[ Contact Us ]
[ Support/Wish List ]
[ Copyright © 2025, Audio Asylum, all rights reserved ]
