De-crackling
a 78 - a (relatively) simple method.
Work ongoing on this page, July 2007.
Without any question, the biggest and most chronic problem in playing
many 78 rpm records made in
That this is the probable explanation, is supported mostly by
‘negative evidence’. For example, occasionally a perfectly ordinary, average
looking record, plays very silently indeed. I have a copy of Parlophone R-2127
(Louis Armstrong’s ‘Savoy Blues’, 1927) which plays superbly with no crackle.
While on the other hand, I once had a laminated
pressing of a Joe Venuti record (these are normally very silent) which crackled
horribly. I know that this latter disc had been stored in a garage or outhouse
for 30 or 40 years. The protracted time in this hostile environment must have
been responsible for the crackle. It follows then, that the silent-surface
‘Savoy Blues’ 78 must have been kept in an optimum environment - though I
cannot attest to that fact. That environment logically would be the exact
opposite to an outhouse. Inside such a shed, the temperature would cycle
between say minus 5 degrees Centigrade in the depths of winter, but with
frequent damp spells above zero, to
perhaps 40 degrees C in summer - a British summer also having many damp spells!
(Ironically, this is being written on
23rd July 2007, following almost unprecedented rainfall and truly disastrous
flooding in many Midland counties of England.)
So an optimum environment for storing 78s would seem rather
obviously to be a dry one, in which a moderate and even temperature is
maintained. There is considerable evidence that this is the case. In North
America, there used to be many Record Libraries, e.g. in radio stations, which
often contained large numbers of British 78 rpm discs, especially in Canada.
Obviously these premises would be air conditioned & maintained at a
suitable temperature, in view of the extremes of climate there, which far
exceed those of Britain. Yet, with the advent of vinyl, tape, CD &c., when
these libraries sold off, their 78 holdings, those British pressings were
highly esteemed by the local collectors, who in some cases preferred them to
Canadian or U.S. pressings!
So I think we can legitimately conclude that our ‘78 problem’
here in Britain is simply due to the weather.
Still, having defined the problem, the question remains: ‘What
can we do about The English Crackle’ when transferring 78s to CD?
There is a digital technique that can help.
It was developed by……. Research
& insert name & references here…
It employs relatively simple software, and only two specialised
functions are needed. These are apparently commonplace in this sort of
software, so we make no attempt to list such programs. We merely point out that
the very ordinary (& cheap) programs already in use here for some years, were
found to possess the two required functions.
For the record, these programs are Diamond Cut 32 (v 4.01) and
Sound Forge 6. The crucial functions are (a) the ability to digitally de-noise
a .wav file and retain the residual
signal, and (b) the ability to paste a .wav file into another while inverting the data being pasted in.
DC32 will perform (a), while Sound Forge will perform (b).
The theory behind this technique is almost absurdly simple - once
you have grasped it! Actually, it is far more complicated to describe than to
carry out - especially if you write complicated old-fashioned English as I tend
to do. I will try to become extremely lucid…
Also,
please allow me to introduce an imaginary friend who asks questions!
Question:
You have told us you have DC32. Why do you not simply use the ‘impulse noise
filter’ provided in DC32 to cut out the crackle?
Answer: That filter is extremely effective in many cases, e.g.
intrusive noises on tape, or a difficult but short passage on a 78. But, alas, it
is not very good for decrackling a 78 ‘all in one go’.
Q: Why?
Surely DC32 is primarily intended for restoring 78s?
A: Yes, it is; and DC32, though old, is excellent &
versatile software for working on 78s and I use it for all my audio editing and
virtually all my audio processing. This is also because I do not have £20,000
to spend on some CEDAR hardware modules & the associated software. By the
way, do you have £20,000 to spend?
Q: Er…
no! How much did DC32 cost you?
A: About £65, several years ago. But can I tell you why DC32
cannot decrackle a 78 side ‘in one go’?
Q: By
all means!
Right. Now I have taken a typical ‘78’ record. It features on
another page of this website; it is Columbia 2129, issued in 1913, and is by
the St. Hilda Colliery Band, a famous - but now alas extinct - brass band from
the north-east of England. At this time,
The first groove on side 1 begins 12.1 cm from the centre of this
disc.
So we may assume an effective diameter of 24.2 cm & hence a
circumference of 24.2 = 76.02 cm.
Having used 80 rpm for the transfer, we have 80 ÷ 60 = 1.33
revolutions per second.
Thus the groove velocity on the first turn is 76.02 x 1.33 = 101.1
cm/sec.
But at the end of the side, the groove is only 5.2 cm from the
centre of the disc.
Thus the groove velocity there is only 10.4
x 1.33, i.e. 43.45 cm/sec.
So from the start of the disc, with each rotation, the groove
velocity will decrease in a linear fashion (assuming a constant groove
spacing), from its initial velocity, to only ~42% of that value by the end of
the record.
No wonder ‘old 78s’ sound duller towards the end. The same amount
of ‘information’ has to be squeezed into less than half the space by the end of
the disc!
Equally, the ‘crackle’ of a 78 is due to the granular structure
of the material from which it is pressed.
And if we assume that the ‘grain size’ of this material is uniform (a
generally permissible assumption), then simple geometry informs us that each ‘crackle’
will be, at the end of the disc, at least twice as long in duration as it was
at the start. But it will also be of lower amplitude, as the velocity of the
groove is lower & so the ‘crackle’ registers less on the stylus. It goes
without saying that any de-crackling process we carry out must observe this progressive change in crackle duration/size.
Unfortunately, DC32 cannot do this. You can only set the
parameters for crackle removal at the start of the disc. After some time, as
the groove velocity gradually falls, the duration of the crackles will
increase, and so another setting in DC32 is required.
Q: Can
you not simply process the 78 in a number of sections, changing the settings as
you progress through the disc?
A: Oh yes! That is quite easy - though rather time-consuming -,
and I frequently use that approach on ‘one-off’ transfers. But there’s another
very significant thing: DC32 is only really effective at decrackling a certain
type of sound, so the content of the record is important.
Q: But
surely, all 78s are much the same in content?
A: [Prolonged laughter] Whatever gave you that idea? I don’t
know how CEDAR works it, but the impulse noise filter in DC32 is highly
influenced by the material on the disc. Audio which contains a lot of slow-rise,
low frequency, legato music is quite
easily decrackled in a small number of sections by DC32. But I’m afraid that
sopranos are a perfect beast!
Q:
Here, steady on! I like sopranos!
A: I dare say: but their high notes look like this to a computer
program:
and all those cycles, as you can see, are contained within the
time period 7.3041 to 7.3412 seconds, a duration of just under four hundredths
of a second; so you see the problem?
Q: No;
what is it?
A: Quite simply, that so many ‘events’ in such a short space of
time is likely to confuse the decrackling function of our program. Mainly, the
decrackling function is likely to regard each
and every one of those sinusoidal excursions as… a steep-rise crackle! And
it will therefore try to remove them, which naturally causes catastrophic
distortion in the resulting output file.
Q: I
thought computers were supposed to be extremely good at this sort of simple,
quick, repetitive task?
A: I’m sure they are; as long has one has spent maybe £20,000 on
a CEDAR outfit which tells them what to do.
Q: Can
the technique you are describing ever cater for such things?
A: In one sense it can, to some extent! But I must confess that I
have played a trick on you, because the wave-form above is an extreme case.
Such a wave-form would need no decrackling. The amplitude of this soprano’s
high note (four-hundredths of a second of Joan Hammond, actually) would by far,
over-ride any crackle on the 78 rpm disc. The ‘difficult crackly bits’ would be
the pauses or silences between various sections of the singing.
Q: Pray
proceed with your explanation. I am greatly interested!
Thank you for your courtesy! Here we go…
As we play a 78, there is a sequence of objectionable ‘crackles’.
Now, leaving aside the music: suppose we
could make a copy of each of these
tiny crackles?
It would look something like this:
Each crackle of course is unique, having its own amplitude and
duration, depending on the size & prominence of the particle in the groove
wall that had caused it.
Then, suppose we inverted
these impulses? So that each of them was equal
& opposite?
We would then have this:
Now comes the clever bit! If we could put this ‘equal & opposite’
signal back into the original sound
file, would not each impulse cancel out the original from which it was derived?
Graphically, we are looking at the following:
If the positive and negative excursions shown above are added
together, they will cancel out, and the result should be zero.
And above all, the inconvenient factor that crackles gradually
get longer (though of smaller amplitude) as the disc plays through, is
automatically taken care of. Because each individual crackle is nullified by an
exactly opposite counterpart derived from it, no matter whether it came from
the beginning of the disc, where crackles are short but of high amplitude; or
from the end, where they are of low amplitude, but more than double the
duration.
So let us now begin work on Columbia 2129. It was recorded in
1912 - the first year the St. Hilda Colliery Band won the Brass Band
Championship. It is a selection from ‘William Tell’, arranged by William
Rimmer, and occupies both sides of the disc.
We will use the start of side 2 for our examples. This is
because, curiously, side 2 is more crackly than side 1. We know that Columbia
recorded - nominally - at 80 rpm at this time, so we played the disc at that
speed. A .0035” TE stylus in a Shure M44 cartridge was used. This is perhaps a
slightly larger size stylus than one might have expected, but it seemed to give
a good ‘solid’ underlying sound.
Here is the appearance of the second side on the screen of DC32.
You will note that there are a number of clicks.
Q: What
is the difference between a small click and a large crackle?
A: I knew you’d ask that! Well, I don’t know whether there is a
proper definition of when a large crackle becomes a small click; after all, one
might say that a small crackle is a very
small click? What they do have in common, of course, is that they both rise
steeply, both are objectionable, & we want to get rid of them. A rule of
thumb I use, is that if you get a distinct visible spike, as above, then it’s a
click. The crackles are really small, and not usually apparent unless you zoom
in quite a lot. In any case, it is better to remove these distinct clicks
manually. There aren’t very many of them on this side, so it won’t take long.
And we want to concentrate on the tiny crackles, and optimise our settings to
deal with those, not huge great clicks.
Here is the file after manual declicking. There are one or two
clicks left, but what the heck!
Q: You
are working in stereo, or at least binaural mono. Why?
A: Because the crackles are scattered at random around both
groove walls. If we put the file into mono. at this stage, some crackles will
be masked, and some will cancel each other out - but many will just be
reduced a bit & lurk around &
not be eliminated. We want maximum ‘visibility’ of crackles at this early
stage.
Click here to hear
the first ~30 seconds of an mp3 of the declicked file above.
Now comes the first critical stage of our procedure. In DC32 (or
whatever program you are using), we apply a severe
digital denoising to the above .wav file - but only ‘save the residue’. The
original file would sound dreadful after this sort of denoising; but we’re not
going to use that. We are just pulling out from it as much noise as we can, for
use later on. This noise consists of rumble, waffle, hiss &c., and the crackle, which is what we’re
interested in.
Here is a picture of the extracted noise:
Click here to
listen to the the first ~30 seconds of it.
The second critical stage is to severely decrackle the above file. In DC32, this means using the
‘impulse noise filter’ with very high settings. For this file at this
amplitude, I used settings blah blah….
Having done this, we ended up with the following file:
It doesn’t look much, does it? It’s actually nearly all hiss, as
you can hear if you click here to
listen to the first ~30 seconds of it. But it is this innocuous looking (&
sounding) file which will enable us to isolate the crackle we need!
We now leave aside DC32 - because it does not possess the
critical function we need right now. But Sound Forge does have the ability to
paste-mix and invert at the same
time, which is what we need to do.
So, we come to the third critical stage. In Sound Forge (or
whatever program you’re using that can paste-mix & invert), open file 3
above - the whole noise.
Then, also open file 4 above, the hiss. Copy this file onto the
clipboard & then close it down or minimise it.
Then, paste-mix file 4
INVERTED into file 3.
The noise - hiss &c - will cancel out the noise in file 3,
and leave just the crackle! Click here to listen to the first ~30
seconds of the ‘crackle file’. Amazing, isn’t it? It’s just crackle &
nothing else.
Finally, copy this crackle file (you don’t actually need to save
it) onto the clipboard, and then paste-mix
it, INVERTED, into file 2 - i.e. the original audio file.
Each crackle should be equal & opposite to those in the
original file, and will cancel them out. You will end up with a decrackled file
- here it is:
It doesn’t look very different; but it sounds different. Click here to
listen to the first ~30 seconds of it. H’mm, I notice that one click has survived
about three-fifths along the left hand channel - but again, what the heck.
Now to put it into mono., which will clear up a lot of rumble
& waffle.
Click here to
listen to the same ~30 seconds.
Finally, we take the two decrackled sides, join them up, make a
fade-in and a fade-out, increase the gain to a reasonable level, and we’ve
finished!
The final file sounds a lot better than it did to start with, and
we haven’t done any digital denoising on it at all. That option remains.
Personally, I hate digital denoising. There is one problem, though. Where side
1 runs into side 2, there is an ugly increase in noise and a significant
improvement of sound quality; the groove velocity has at least doubled, and
side 2 was always noisier than side 1 for some reason.
Q: What
are you going to do about that, pray?
A: I don’t know!
Q: But
I thought you knew everything about transferring 78s?
A: Good heavens, no. But I’m very willing to learn! It is obviously
unthinkable to reduce the quality of the beginning of side 2 to match the end
of side 1; and I can’t think of any way to increase the quality of the end of
side 1. There must be an answer somewhere, though; the serious professionals
who transfer 12” (30 cm) 78 sets of classical symphonies, operas &c must
have cracked this problem long ago! 8^)
To hear the full mp3 of this performance, check it out on the
page which I have recently started on Brass Band 78s: click here to go
to it.
Page
written 23rd/24th July 2007.