Good Remastering
Matt Kennel
huge difference in sound, on the same order as quality of speakers, and
much much more than any difference in electronics.
I recently got a CD recently from the RCA "Living Stereo" new CD remasters,
Brahms & Tchaikovsky violin concerti. This is a 'new' remastering, where
they tried very hard to find the cleanest original masters and took great
care to get good sound.
I have previously released CD versions of the same recording---the new one
is FAR superior. In fact the sound quality is excellent---not 'for 1957'
but 'for 1993'! There is of course, more tape hiss than on modern digital
tapes (the tape hiss is constant and unobtrusive: no modulation with the
signal), but the tonal balance and general texture is marvelous, superior to
most contemporary recordings. Very well captured low frequencies, e.g. low
strings, that is often absent or thin on contemporary recordings. (I bought
an 89 recording of the Berlin Phil (philips) the same day: great clear brass
but the cellos are emasculated!)
In some ways, it points out how *little* stereo recording has progressed, as
these recordings occured when stereo was a studio experiment, before any sort
of consumer stereo playback equipment was available.
If this is what good LP's sound like (produced at the time of the original
mastering in contrast to slapjob remastered CD versions), then I can
understand why some people are dedicated to the LP. (I'm not: I can't
bear the pain of cleaning and thinking of the degradation
every time I play the damn disc)
And I'd also invite LP lovers to listen to this CD, to show that their
dissatisfaction is NOT inherent in the technology, but the recordings.
Too often, of course, you have "audiophile recordings" of mediocre performances
by nobodies.
Safe to say, this isn't the case here, with Fritz Reiner, the Chicago Symphony
and Jascha Heifetz.
The performance is impossible to review adequately, so I won't even try.
I'll just say this:
The definitive recordings of the 20th century.
(ask a violinist)
--
-Matt Kennel m...@inls1.ucsd.edu
-Institute for Nonlinear Science, University of California, San Diego
-*** AD: Archive for nonlinear dynamics papers & programs: FTP to
-*** lyapunov.ucsd.edu, username "anonymous".
Michael Golan
[interesting info on remasters, which I think I can confirm, deleted]
>There is of course, more tape hiss than on modern digital
>tapes (the tape hiss is constant and unobtrusive: no modulation with the
>signal),
Layman's question:
with all the available DSP, can't such hiss be removed or drastically
reduced without a serious effect to the sound? why not?
I'm especially sensitive to hiss, so I just put my digital EQ at around 10khz
with Q=8 or so, with -10db. I loose some of the violin's "edge" (:-) but,
thanks sony, the hiss is mostly gone. I love the ability of the paramateric
EQ to be set "just right".
Is there a good technical reason not to clean up the hiss in the digital
remastering process?
-- Michael
--
Michael Golan | Duel, an addon to gdb, allows "x[..100] >? 0" to
m...@cs.princeton.edu | show the positive elements of x in the debugger.
| annon ftp ftp.cs.princeton.edu:/duel or send me mail!
Gabe M Wiener
>>There is of course, more tape hiss than on modern digital
>>tapes (the tape hiss is constant and unobtrusive: no modulation with the
>>signal),
>
>Layman's question:
>with all the available DSP, can't such hiss be removed or drastically
>reduced without a serious effect to the sound? why not?
To a point. It is a subjective call as to whether or not youi are
damaging the program material.
Most DSP restoration work is done on the Sonic Solutions NoNOISE system,
so for the benefit of the curious, let me run through the steps that one
generally goes through to make a new master.
For the purposes of this discussion, we're only going to talk about
the DSP side. There are other issues involved, such as making sure
your playback machine has similar frequency characteristics to that of
the record machine, and oftentimes you can't know that for sure and
have to make assumptions. And we won't get into the process of
transferring off of vinyl or lacquered 78's or any of that, since
that's even farther afield from the original thread.
Okay. You've got your playback machine all set up and you're sitting
at a Sonic Soulutions digital audio workstation. What happens? The
first thing you've got to do is get it loaded onto the machine. The
Sonic is a hard-drive based system, so you get a time-stamp set up and
create a folder structure for your project, and you then begin to set
your input A/D levels.
Setting these isn't all that hard. The dynamic range of the original
program is generally quite limited, so you can usually see how your
program material is falling on the meters. Being that the dynamic
range of the input material is considerably less than that of the
digital medium, you can afford to leave a few dB at the top "just in
case." [Generally speaking, when recording live music to digital, you
want to run as hot as possible so that the subtleties of the soft
sections, which today's equipment is fully capable of capturing sans
hiss, are not lost.]
Your levels are now set and the system is ready. You cue up your tape
to the beginning, get the Sonic into record mode, and away you go.
While the program material is loading, you can hear it through the
monitor buss of the Sonic. You take this opportunity to fiddle with
the EQ a little (these changes don't get recorded) and make some notes
about what you might want to do....shelving here, presence cut there,
etc.
So, an hour later your material is loaded. You now have to determine
what sorts of DSP it needs. Generally there are four kinds of processing
one can use. They are:
* Complex filtering -- stackable EQ's...up to 512 per pass.
* Declicking -- Reconstruction of dropouts, removal of
impulsive clicks and thumps.
* Decrackling -- Removal of "grain" or "bacon frying" sounds
(usually only used on vinyl). This is essentially a
very high-precision declicking system that runs on the
microscopic level.
* Broadband denoising -- Removal of hiss, hum, and other
constant noise. The easiest Sonic tool to abuse.
To begin, you find a piece of silence, and you run a frequency
analysis on it. From this graph you can see how well the silence is
following a standard noise curve. The analysis also lets you see any
anomalies such as 60 Hz hum and its harmonics. From these graphs, you
decide how much complex filtering you want to do in order to restore
good linear performance. You have to be careful to avoid the program
material when doing any filtering. Remember, at this stage, you're
not trying to make artistic EQ choices. You're trying to remove
anomalies and restore good performance.
It is a sad state of things that it is impossible to restore what has
been thrown away. Thus if the engineer in 1950 decided to notch out a
band at 7 kHz because there was some horrible noise in there, you
cannot get it back in order to remove the noise with DSP. The
material is gone.
We'll assume now that you've done your complex filtering and you like
the results (which you often don't, and thus you often have to run the
pass again). Now it's on to the tedious work of production
declicking. A click is an impulsive disruption of the waveform. The
software has several interpolator types that rely on different kinds
of contextual information to make a reconstruction. The easiest waves
to reconstruct are those with high rates of periodicity, such as
voice. The amount of clicks on a given recording can vary from none
to many per second. Whether you have one click or thousands, you
want to get rid of them, and the system does this rather well.
If you only have a few clicks, it's probably faster to do them by
hand. You locate them in the source, identify the anomaly, and then
pick an interpolator function based on your assessment of what sort of
waveform it is. If it's periodic, you can use one of the more
intelligent algorithms that will search for periodicity in the context
of the anomaly and repair it. If it isn't, you need to use an
algorithm for non-periodic waveforms, which usually still does a good
job, but works in a very different way.
If you have many clicks (such as on a vinyl record), you probably
don't want to do them all by hand, so you get the production declicker
ready. First you run a click detect pass. What you do here is to
specify parameters that essentially tell the machine what to look
for...duration of the click, wing weight of the click, etc. It then
goes through and finds all the clicks of that "type." At that point
you can actually start the declick pass, where the system will go
through one at a time and repair them, performing the necessary
interpolations.
Of course, you usually have to run several passes...first to get the
"big guys" and then to clean up successively smaller clicks. The
really miniscule clicks will be dealt with in Decrackling. And
even with the automatic declicker, you still have to go in and fix
a few by hand...usually those which the auto declicker has screwed
up. It isn't nearly perfect.
Decrackling is easier to deal with. First, it is usually only neaded
when working with records...it takes care of the grainy, microscopic
clicks that make up the annoying part of record surface noise. This
too usually has to run in several passes, since you usually want to
start with larger degrees of crackle and get smaller as needed.
Now comes the tricky part...broadband denoising. Here is where the
job of remastering ventures away from science and becomes art, since
there is no "right" way to do this. The process works as follows.
You take a "thumbprint" of the background noise, and you create a
denoise curve...i.e. a curve that identifies what the noise *is*.
The theory of removing hiss goes like this. The system performs a
2,048-point FFT on the signal. Using the noise estimate, the system
analyses each point and determines if at that instant it contains
noise (as you have defined noise from the curve) or program material.
Program material is left untouched. Noise is treaded according to
the parameters that you set.
You must specify the threshold (whether something is program material
or noise), the amount of attenuation for noise, the speed at which
the noise gate operates, the cohesiveness of each of the noise bands,
and the frequencies over which the system operates (you rarely want
to attempt to remove noise at low frequencies, for instance).
Setting up broadband denoising is very tricky. Take too much noise
away, and you cut into the program material, often in very subtle
ways. Set it up wrong and your noise floor will appear to modulate.
There is no *right* way to do broadband. It's a lot of fiddling with
the controls, comparing before/after, etc. Thankfully it runs in
realtime so you can start the material playing, and then just start
tweaking everything until you start getting comfortable with it.
Eventually, you make a decision and let it run.
When NoNOISE first came out, people were so excited that they tried to
get ALL the noise, and discovered that they lost most of their room
ambience at the same time. Not good. The fact of the matter is that
you cannot get out all the noise, not if you want your program
material to remain fully intact. Denoising is always a trade-off
between keeping your hands off the progam and getting out as much of
the noise as you can. As I said, it is the easiest part to abuse.
After denoising, the only thing left is to make any artistic EQ
choices, to edit the recording as needed, and then to prepare the
new master.
--
Gabe Wiener -- gm...@cunixa.cc.columbia.edu -- N2GPZ
Sound engineering, recording, and digital mastering for classical music
"I am terrified at the thought that so much hideous and bad music
will be put on records forever." --Sir Arthur Sullivan
Dave Platt
>When NoNOISE first came out, people were so excited that they tried to
>get ALL the noise, and discovered that they lost most of their room
>ambience at the same time. Not good. The fact of the matter is that
>you cannot get out all the noise, not if you want your program
>material to remain fully intact.
I have a number of CDs on the German LINE label - the only available CD
issues of the first few albums by an English rock(sorta) group called
Jade Warrior. I understand that these discs, like many of the other
LINE reissues, were processed by the NoNOISE system.
They show just the sort of over-eager broadband-noise elimination that
Gabe speaks of. They are, indeed, very quiet - the original analog tape
hiss is almost impossible to detect. Unfortunately, the de-noising
thresholds seem to have been set rather too high... during periods of
low overall signal level, the upper midrange and treble seem to be
entirely absent.
It's a shame... I'd much rather have had a small amount of residual tape
hiss, if that were the price to having all of the original program
material survive unscathed.
--
Dave Platt VOICE: (415) 813-8917
Domain: dpl...@ntg.com UUCP: ...netcomsv!ntg!dplatt
USNAIL: New Technologies Group Inc. 2470 Embarcardero Way, Palo Alto CA 94303
Gabe M Wiener
>
>I have a number of CDs on the German LINE label - the only available CD
>issues of the first few albums by an English rock(sorta) group called
>Jade Warrior. I understand that these discs, like many of the other
>LINE reissues, were processed by the NoNOISE system.
>
>They show just the sort of over-eager broadband-noise elimination that
>Gabe speaks of. They are, indeed, very quiet - the original analog tape
>hiss is almost impossible to detect. Unfortunately, the de-noising
>thresholds seem to have been set rather too high... during periods of
>low overall signal level, the upper midrange and treble seem to be
>entirely absent.
It is a *very easy* mistake to make, and novice operators of NoNOISE
workstations _always_ start out by setting their denoise thresholds
and reduction settings way too high. It isn't until they really
A/B the two (which the system allows you to do, as a toggle) that
it becomes apparent that you're being too aggressive. Sadly, many
of the earliest operators were poorly trained and botched a lot
of recordings.
The two extremes of broadband denoising are a) no change, and
b) a recording that sounds like it's being played under water.
>It's a shame... I'd much rather have had a small amount of residual tape
>hiss, if that were the price to having all of the original program
>material survive unscathed.
The current school of thought on NoNOISE operation is to remove only
as much noise as can be cut without damaging the source material.
Usually that's quite a lot if you set everything up right.
Bill Vermillion
>In article <1993Jun13....@ntg.com> dpl...@ntg.com (Dave Platt) writes:
>>I have a number of CDs on the German LINE label - the only available CD
>>issues of the first few albums by an English rock(sorta) group called
>>Jade Warrior. I understand that these discs, like many of the other
>>LINE reissues, were processed by the NoNOISE system.
>>They show just the sort of over-eager broadband-noise elimination that
>>Gabe speaks of. They are, indeed, very quiet - the original analog tape
>>hiss is almost impossible to detect. Unfortunately, the de-noising
>>thresholds seem to have been set rather too high... during periods of
>>low overall signal level, the upper midrange and treble seem to be
>>entirely absent.
>It is a *very easy* mistake to make, and novice operators of NoNOISE
>workstations _always_ start out by setting their denoise thresholds
>and reduction settings way too high. It isn't until they really
>A/B the two (which the system allows you to do, as a toggle) that
>it becomes apparent that you're being too aggressive. Sadly, many
>of the earliest operators were poorly trained and botched a lot
>of recordings. ^^^^^^^^^
I think operators is the key word here. It seems that there are many
who think you can teach a person the fundamentals (and often only some
of them) - tell them what buttons to push - and PRESTO! they become an
audio engineer. I have seen 'engineers' turned out by some of the
'schools' - and I don't know whether I'd let some of them change a
cassette.
And of course a lot of companies are seeing just how great the re-issue
market is - so they have an agressive re-issue campaign - get the
recordings out - but they have lost respect for the music. This of
course even goes back to the golden eras of LPs. Some companies went
back to masters and recovered as much of the performance as possible -
at an expense. Others just grabbed the material and re-issued it.
On the latter I have seen instances where the 'master' for the
re-issued CD was the tape generated for the 'psuedo-stereo' releases of
the '60's - and really had no semblance of the orignal.
>The two extremes of broadband denoising are a) no change, and
>b) a recording that sounds like it's being played under water.
>>It's a shame... I'd much rather have had a small amount of residual tape
>>hiss, if that were the price to having all of the original program
>>material survive unscathed.
>The current school of thought on NoNOISE operation is to remove only
>as much noise as can be cut without damaging the source material.
>Usually that's quite a lot if you set everything up right.
And if you care about the music - instead of treating it as 'some old
record' - you will probably use less NoNOISE.
Again - these are problems associated with the current record industry.
It happened when the musicians who ran record companies were replaced
by lawyers and accountants.
Thankfully the advent of the CD has meant that more small
music-oriented labels exist than in past years.
--
Bill Vermillion - bi...@bilver.uucp OR bi...@bilver.oau.org
Gabe M Wiener
>>It is a *very easy* mistake to make, and novice operators of NoNOISE
>>workstations _always_ start out by setting their denoise thresholds
>>and reduction settings way too high. It isn't until they really
>>A/B the two (which the system allows you to do, as a toggle) that
>>it becomes apparent that you're being too aggressive. Sadly, many
>>of the earliest operators were poorly trained and botched a lot
>>of recordings. ^^^^^^^^^
>
>I think operators is the key word here. It seems that there are many
>who think you can teach a person the fundamentals (and often only some
>of them) - tell them what buttons to push - and PRESTO! they become an
>audio engineer. I have seen 'engineers' turned out by some of the
>'schools' - and I don't know whether I'd let some of them change a
>cassette.
I must agree with this assessment. Far too many people think that
good engineering means little more than pushing buttons and twiddling
knobs. I've seen a lot of people come out of engineering schools,
supposedly ready to go into the 'working world', and not know signal
flow, or impedance matching, or gain structure, or a dozen other
things that go beyone "push this button to do this." It's all
pointless anyway unless you know how to use those two things on the
side of your head, which unfortunately a lot of people in pro audio
don't, which is why there are so many bad recordings out there.
Several writers were quite skeptical of the NoNOISE system when it
first came into use in the late 1980's, mainly because they heard
recordings processed by people who didn't kknow what they were doing.
NoNOISE work is an art, not a science. There is no way I can teach
someone in the fashion of "if you want to get rid of this, you do
this." That isn't how it works. Every situation is different, and
the only way to learn to use the workstation is to first have a
thorough grasp of what every setting is doing, and then....how do you
get to Carnegie Hall again?
Norman P. Tracy
>m...@lyapunov.ucsd.edu (Matt Kennel) writes:
>
>[interesting info on remasters, which I think I can confirm, deleted]
>
>>There is of course, more tape hiss than on modern digital
>>tapes (the tape hiss is constant and unobtrusive: no modulation with the
>>signal),
>
>Layman's question:
>with all the available DSP, can't such hiss be removed or drastically
>reduced without a serious effect to the sound? why not?
Such devices exist on most DAWs (Digital Audio Workstations). Many classical
recordings have been processed via DSP, or simply low pass filtered, to reduce
hiss from 40+ year old master tapes. The problem is that none of the available
systems can remove the hiss without also taking along a big chunck of the signal.
Keep in mind that tape hiss is a broad band white noise that can cover several
of the top octaves.
>
>I'm especially sensitive to hiss, so I just put my digital EQ at around 10khz
>with Q=8 or so, with -10db. I loose some of the violin's "edge" (:-) but,
>thanks sony, the hiss is mostly gone. I love the ability of the paramateric
>EQ to be set "just right".
There you state your preference "I'm especially sensitive to hiss". I would
state mine as "I'm especially sensitive to information loss due to heavy handed
filtering". Now if the master tape sound is left along during transfer to CD
I will argue that its a whole lot easier for the Michael Golans of this world
to filter out the hiss to their taste than for me to recreate that "violin's "edge"",
i.e. to recreate missing information.
>
>Is there a good technical reason not to clean up the hiss in the digital
>remastering process?
I would say its more experence with what the market wants and an aesthetic
decision rather than technical reasons that leads the charge to hands off
minimal signal processing remastering. In the early days of the CD format
it was quite common to appologize for any analog master tape hiss and to
filter, filter, filter it off all reissues. Over the long run this failed
to please both aesthetic camps. The no-hiss-at-any-cost listener only
bought DDD disks anyway. The I-want-all-the-signal-off-the-master-tape
listeners auditioned early filtered reissues and were not impressed. Then
Philips allowed the Mercury Living Presence catalogue to be remastered under
the guidance of Wilma Fine, original coproducer of the series and widow of
the legendary Bob Fine who engineered the series in the `50s &`60s. Philips
original filtered CD versions had been ho-hum sellers while good copies of
the LPs fetched over $200 each among collectors. Wilma Fine's versions used
restored Ampex 300B tape decks matching the original master recorders, a very
high quality signal path and the latest 128x oversampling A/D converters.
And of course used the treasured original master tapes as the source.
No filtering or gain riding at all was used. The result is the Living Presence
legend at last transfered to CD intact. The $$$ result for Philips is a series
of best selling mid-priced CDs with almost no production cost compared to the
$30,000 a day fees for rerecording with a modern orchestra. The result for
lovers of the Golden Age of `60s symphonic stereo recordings is that we can
now own a virtual clone of a priceless master tape.
With Philips raking in $$$$ hand over fist it did not take too long for RCA
and Sony (who owns Columbia's et. al. master tape libraries after buying up
a bunch or record companies) to notice and start purest reissues of their
back catalogue. The RCA Living Stereo disk that started this thread is an
example of RCA's efforts along this path as are Sony's Super Bit Mapped
versions of the Glenn Gould catalogue and jazz masterpieces like Miles Davis'
'Kind of Blue'.
>
>-- Michael
>--
>Michael Golan | Duel, an addon to gdb, allows "x[..100] >? 0" to
>m...@cs.princeton.edu | show the positive elements of x in the debugger.
> | annon ftp ftp.cs.princeton.edu:/duel or send me mail!
---
Norman Tracy
Gabe M Wiener
>
>Such devices exist on most DAWs (Digital Audio Workstations). Many classical
>recordings have been processed via DSP, or simply low pass filtered, to reduce
>hiss from 40+ year old master tapes.
Wrong word. I think you meant "historical" recordings. Classical recording
refers to music, and miking technique above all, not age or quality.
>The problem is that none of the available
>systems can remove the hiss without also taking along a big chunck of the signal.
>Keep in mind that tape hiss is a broad band white noise that can cover several
>of the top octaves.
Well, none of them if used imporperly, anyway.
>There you state your preference "I'm especially sensitive to hiss". I would
>state mine as "I'm especially sensitive to information loss due to heavy handed
>filtering". Now if the master tape sound is left along during transfer to CD
>I will argue that its a whole lot easier for the Michael Golans of this world
>to filter out the hiss to their taste than for me to recreate that "violin's "edge"",
>i.e. to recreate missing information.
Remember, to remove hiss properly, one does not filter. One discriminates
between program and noise through DSP. The skill with which the operator
can manipulate the parameters will determine the ultimate success of the
project.