Finding clicks
 

On Thu, 11 Sep 2014 09:40:52 +0100, Jim Lesurf
wrote:

In article , William Unruh
wrote:
On 2014-09-10, Jim Lesurf wrote:
In article , William Unruh


So use sox say to impliment the inverse RIAA, then use audacity to
look for those spikes, and remove them, then use the RIAA on the
result. Note that one could just take the derivative, but that would
still leave a finite spreading due to the treble/bass boost.

Wary of that because 'mending' a differential waveform might lead to a
dc offset problem when you re-integrate the result. So I'd use a dx/dt
or

You could always put in a 50 or 30 Hz cutoff in the RIAA curve. Some
advocated that anyway. But those clicks put in a DC bias in the first
place.

Erm. The mechanics and the RIAA don't pass down to dc. So what happens is a
decaying offset. The results shapes are pretty clear. In my case I'm using
a V15 in an old arm that has more mass than ideal. So the peak and fall at
LF is at very low frequency, but not dc.


d2x/d2t to *find* and list click locations. But do any editing on the
actual audio file recorded using RIAA. Avoids the problems of dealing
with the real response curve being rather complicated.

But since the click has been spread out all over hells half acre by
RIAA, that "fixing" either leaves loads of artifacts or also "fixes" a
bunch of the real signal as well.

The pre RIAA is the place to fix it.

Again, looking at the shapes I can see the effects. Adding the 'fix' just
puts in a plausible smooth interpolation anyway.

To deal with it in the way you suggest would require an accurate 'de-riaa'
that also precisely deals with the stylus and arm responses over the full
range down to almost dc. i.e. much lower than 10Hz or so. Even measuring
that isn't trivial. And it differs in the vertical and horizontal planes
anyway. So you'd also have to convert the L and R to V and H first.

So simply applying a reverse riaa preamp curve won't in practice be much
better than a simple integrator if your concern is LF spread.

Given that the mends I've made so far are generally inaudible except for
severe events that clearly lose the waveform anyway. I'm happy enough
despite the nice theory for preferring de-riaa. Life's too short. :-)

Jim

Jim, have you tried differentiating twice - once as a very approximate
RIAA counter, and the second time to pull out the sharp edges into
peaks?

d

Finding clicks
 

In article , William Unruh
wrote:
On 2014-09-11, Jim Lesurf wrote:
In article , William Unruh
wrote:


You could always put in a 50 or 30 Hz cutoff in the RIAA curve. Some
advocated that anyway. But those clicks put in a DC bias in the first
place.

Erm. The mechanics and the RIAA don't pass down to dc. So what happens
is a decaying offset. The results shapes are pretty clear. In my case
I'm using a V15 in an old arm that has more mass than ideal. So the
peak and fall at LF is at very low frequency, but not dc.

Actually, the RIAA curve does go to DC.

It's a nice example of the old engineering maxim. "In theory, theory and
practice agree. But in practice, they don't". The reality is that the
system won't go to dc. And the gain of almost any real-world RIAA pre-amp
won't. And in my system the response doesn't go to dc.


There was a very controvertial
proposal to put another zero/pole at 50Hz to comensate for the cutter
low freq resonance but the problem is that the cutters all have
different resonances to for some it would make thing worse. Anyway,
since your speakers cannot hear 30Hz, you could put it there-- the main
thing is that the unRIAa and RIAA filter be complementary.

....which still won't deal with the point I made that the real world
response will have a bump and fall as you go to lower frequencies. Not a
flat response once RIAA is taken out on the basis of theory. Since that'll
the LF part, it'll be the region you're worried about.

If the spike from the record is before the RIAA then the RIAA filter
will have spread it out all over the place, and "fixing" it after the
filter will leave all that spread out residual in place.

However I find in practice this isn't a worry. I can't hear or see the dc
you claim. I can sometimes see the tail of the decay, but it is small
enough to be inaudible.

And clicks seem to divide into two main classes.

1) fast 'hf ticks' which have a fast rise and quickly fall to nothing much.

2) round-the-bolders which have a dual spike - first one way, then the
other. Which is what you'd expect for a velocity sensor tracking around a
rock or dipping into wall damage and out again.

So simply applying a reverse riaa preamp curve won't in practice be
much better than a simple integrator if your concern is LF spread.

Agreed that the curve is problematic below 50Hz. But even at 200Hz the
sound is spread out over more than 200 time pixels (400 for 96K
sampling). That's a lot.

Sorry, you'll have to convert that into English I can understand! :-)


Given that the mends I've made so far are generally inaudible except
for severe events that clearly lose the waveform anyway. I'm happy
enough despite the nice theory for preferring de-riaa. Life's too
short. :-)

It is cheap enough to try it. I agree that it may not be an improvement.
Even just a differentiator would be a help (differentiate, fix,
integrate) except you really have to make sure you have enough dynamic
range. Since that is 10 octaves or 60dB emphasis of highs over lows,
which is even larger than the 48dB of the RIAA curve ( which would fit
in another 8 bits that sox stores stuff at.)

Actually I'm more curious about how you'd deal with the constant of
integration problem when snipping bits out of a differential series before
integrating.

Jim

--
Please use the address on the audiomisc page if you wish to email me.
Electronics http://www.st-and.ac.uk/~www_pa/Scot...o/electron.htm
Armstrong Audio http://www.audiomisc.co.uk/Armstrong/armstrong.html
Audio Misc http://www.audiomisc.co.uk/index.html

Finding clicks
 

In article , Don Pearce
wrote:

Jim, have you tried differentiating twice - once as a very approximate
RIAA counter, and the second time to pull out the sharp edges into peaks?

Not yet, but it is what I have in mind to try sometime. However it isn't to
'un-do RIAA' but to enhance the click-to-music-ratio of clicks over the
bulk of the music.

FWIW in practice for *fixing* I find that

A) For *big* clicks / bangs the waveform is nonlinearly distorted as the
stylus clearly is going no-where near the intended waveform for a while as
it's not in contact with the groove walls as cut. The transducer will be
nonlinear in this situation as well even if I avoid clipping the 96k/24
recording. (Which does sometimes happen for these nasty bangs even though
I've set 0dBFS at about +18dBRIAA.)

B) For small ticks I tend to see a fast rise and a decay that looks
exponential at a rate set by the stylus suspension and tip mass. The decay
time is far too short to be set by the roll-off at LF of the replay RIAA
amp or arm-mass LF limit.

(A) isn't going to be truly fixable by any means because there's no data.
Might as well draw in a plausible shape by eye since you can't tell what
was cut on the LP.

(B) is so short that simply doing a sensible interpolate as per Audacity's
'repair' gives a result where I can't here there was any problem.

I'll look at the differential when I can to see how it squares with what
Bill says. But the bottom line so far seems to be that for all but the most
dire clicks I can easily do repairs that render the result audibly fine.
(And the dire ones lack the info you'd need to get the intended waveform.)
The problem is finding some of them because they don't poke obviously out
of the waveform. Takes time. So I have to limit how much fixing I do
because I don't have the time to find all the tiny boogers hiding in the
waveforms.

Jim

--
Please use the address on the audiomisc page if you wish to email me.
Electronics http://www.st-and.ac.uk/~www_pa/Scot...o/electron.htm
Armstrong Audio http://www.audiomisc.co.uk/Armstrong/armstrong.html
Audio Misc http://www.audiomisc.co.uk/index.html

Finding clicks
 

On 2014-09-11, Don Pearce wrote:
On Thu, 11 Sep 2014 09:40:52 +0100, Jim Lesurf
wrote:

In article , William Unruh
wrote:
On 2014-09-10, Jim Lesurf wrote:
In article , William Unruh


So use sox say to impliment the inverse RIAA, then use audacity to
look for those spikes, and remove them, then use the RIAA on the
result. Note that one could just take the derivative, but that would
still leave a finite spreading due to the treble/bass boost.

Wary of that because 'mending' a differential waveform might lead to a
dc offset problem when you re-integrate the result. So I'd use a dx/dt
or

You could always put in a 50 or 30 Hz cutoff in the RIAA curve. Some
advocated that anyway. But those clicks put in a DC bias in the first
place.

Erm. The mechanics and the RIAA don't pass down to dc. So what happens is a
decaying offset. The results shapes are pretty clear. In my case I'm using
a V15 in an old arm that has more mass than ideal. So the peak and fall at
LF is at very low frequency, but not dc.


d2x/d2t to *find* and list click locations. But do any editing on the
actual audio file recorded using RIAA. Avoids the problems of dealing
with the real response curve being rather complicated.

But since the click has been spread out all over hells half acre by
RIAA, that "fixing" either leaves loads of artifacts or also "fixes" a
bunch of the real signal as well.

The pre RIAA is the place to fix it.

Again, looking at the shapes I can see the effects. Adding the 'fix' just
puts in a plausible smooth interpolation anyway.

To deal with it in the way you suggest would require an accurate 'de-riaa'
that also precisely deals with the stylus and arm responses over the full
range down to almost dc. i.e. much lower than 10Hz or so. Even measuring
that isn't trivial. And it differs in the vertical and horizontal planes
anyway. So you'd also have to convert the L and R to V and H first.

So simply applying a reverse riaa preamp curve won't in practice be much
better than a simple integrator if your concern is LF spread.

Given that the mends I've made so far are generally inaudible except for
severe events that clearly lose the waveform anyway. I'm happy enough
despite the nice theory for preferring de-riaa. Life's too short. :-)

Jim

Jim, have you tried differentiating twice - once as a very approximate
RIAA counter, and the second time to pull out the sharp edges into
peaks?

Lets model the "clicks". They are very localized defects in the grove.
The cartridge is a velocity sensor, so a scratch should produce a rapid
velocity excursion to the left say, and then to the right. This produces a
voltage which is roughly the derivative of a delta function-- The RIAA
curve then integrates this and smears it out with the high boost
(corners 500Hz and 2KHz) As a rough approx, the derivative of the delta
function then gets converted into a sharp spike, plus a smearing out
over about a millisecond. Thus the first thing you would want is to
unsmear it-- Ie, apply a bass boost with corners again at 500Hz and 2000
Hz.
Of course you also have to be careful that the phases also cancel the
original.

I suspect that the RIAA boost desmearing would give the greatest
improvement. But then I have not implimented this.





d

Finding clicks
 

In article , William Unruh
wrote:

Lets model the "clicks". They are very localized defects in the grove.
The cartridge is a velocity sensor, so a scratch should produce a rapid
velocity excursion to the left say, and then to the right. This produces
a voltage which is roughly the derivative of a delta function-- The RIAA
curve then integrates this and smears it out with the high boost
(corners 500Hz and 2KHz) As a rough approx, the derivative of the delta
function then gets converted into a sharp spike, plus a smearing out
over about a millisecond. Thus the first thing you would want is to
unsmear it-- Ie, apply a bass boost with corners again at 500Hz and 2000
Hz. Of course you also have to be careful that the phases also cancel
the original.

That's pretty much what I've had in mind. By observation, the clicks I see
are generally rather shorter than 1 ms. If I use some simple round numbers,
take 0.2 ms for the click and assume the roll-off of the replay at LF is at
about 5 Hz. That implies that only about 0.1% of the amplitude of the click
will find its way into a tail so far as I can see. i.e. about 60dB down on
the click.

i'm basing this on the mechanical process of the interaction where the
impulse starts to move the arm via the stylus compliance. This then creates
a tail. But the impulse is very short compared with the stylus-arm response
time.

Even allowing for the integration (RIAA boost at LF) of a falling tail the
resulting thump (mostly at low frequencies we don't easily hear) will be
well less that the click itself.

This seems to square with what I hear.

That for the small clicks simply 'repairing' the RIAA output leaves no
audible 'bump'.

That for large long-duration bangs you end up having to redraw or snip
anyway as the result has no linear relation with the intended waveform.
Undoing riaa doesn't recover it.

So I'm still in the situation where I can see the theory makes sense as
theory for doing repairs 'without RIAA'. But in practice it doesn't look to
me to be needed. And applying differentials, altering, and integrating
again may have the drawback of *creating* an offset due to the problem of
integration constants.

Maybe I don't notice this because my system has a relatively low LF
rollaway. Result of using a V15 in an arm that nominally has too high a
mass. However I'll need to check the LF roll away of my RIAA amp as well to
decide as I've forgotten the choice I made mumble years ago. That said,
there is another LF pole I stuck in the path between RIAA output and 'tape'
output. Think that's pretty low though.

Jim

--
Please use the address on the audiomisc page if you wish to email me.
Electronics http://www.st-and.ac.uk/~www_pa/Scot...o/electron.htm
Armstrong Audio http://www.audiomisc.co.uk/Armstrong/armstrong.html
Audio Misc http://www.audiomisc.co.uk/index.html

Finding clicks
 

["Followup-To:" header set to uk.comp.os.linux.]
On 2014-09-11, Jim Lesurf wrote:
In article , William Unruh
wrote:
On 2014-09-11, Jim Lesurf wrote:
In article , William Unruh
wrote:


You could always put in a 50 or 30 Hz cutoff in the RIAA curve. Some
advocated that anyway. But those clicks put in a DC bias in the first
place.

Erm. The mechanics and the RIAA don't pass down to dc. So what happens
is a decaying offset. The results shapes are pretty clear. In my case
I'm using a V15 in an old arm that has more mass than ideal. So the
peak and fall at LF is at very low frequency, but not dc.

Actually, the RIAA curve does go to DC.

It's a nice example of the old engineering maxim. "In theory, theory and
practice agree. But in practice, they don't". The reality is that the
system won't go to dc. And the gain of almost any real-world RIAA pre-amp
won't. And in my system the response doesn't go to dc.

Agreed. But clicks are never DC. That is wow and maybe flutter. Ie,
the stylus never wanders over into the next room, whetehr in velocity of
in position space.



There was a very controvertial
proposal to put another zero/pole at 50Hz to comensate for the cutter
low freq resonance but the problem is that the cutters all have
different resonances to for some it would make thing worse. Anyway,
since your speakers cannot hear 30Hz, you could put it there-- the main
thing is that the unRIAa and RIAA filter be complementary.

...which still won't deal with the point I made that the real world
response will have a bump and fall as you go to lower frequencies. Not a
flat response once RIAA is taken out on the basis of theory. Since that'll
the LF part, it'll be the region you're worried about.

Why am I worried about it? clicks are a few time pixels, not 40000 of
them or 90000 of them.

If the spike from the record is before the RIAA then the RIAA filter
will have spread it out all over the place, and "fixing" it after the
filter will leave all that spread out residual in place.

However I find in practice this isn't a worry. I can't hear or see the dc
you claim. I can sometimes see the tail of the decay, but it is small
enough to be inaudible.

I am not worried about the DC. I am worried about spreading out on time
scales of milliseconds.



And clicks seem to divide into two main classes.

1) fast 'hf ticks' which have a fast rise and quickly fall to nothing much.

That would look more like an integrated velocity tick.


2) round-the-bolders which have a dual spike - first one way, then the
other. Which is what you'd expect for a velocity sensor tracking around a
rock or dipping into wall damage and out again.

Except RIAA integrates out that velocity tracking, This would be more
like a scratch displacing plastic to make a moat and wall.

Well on time scales of milliseconds it is more complex.
and it is these types of ticks I would be more concerned with as they
tend to last longer and thus get more spread out by RIAA.



So simply applying a reverse riaa preamp curve won't in practice be
much better than a simple integrator if your concern is LF spread.

For me, let me define LF as millisecond stuff.

Agreed that the curve is problematic below 50Hz. But even at 200Hz the
sound is spread out over more than 200 time pixels (400 for 96K
sampling). That's a lot.

Sorry, you'll have to convert that into English I can understand! :-)

One time pixel is the time for one time sample. 44000 times a second for
cds,




Given that the mends I've made so far are generally inaudible except
for severe events that clearly lose the waveform anyway. I'm happy
enough despite the nice theory for preferring de-riaa. Life's too
short. :-)

It is cheap enough to try it. I agree that it may not be an improvement.
Even just a differentiator would be a help (differentiate, fix,
integrate) except you really have to make sure you have enough dynamic
range. Since that is 10 octaves or 60dB emphasis of highs over lows,
which is even larger than the 48dB of the RIAA curve ( which would fit
in another 8 bits that sox stores stuff at.)

Actually I'm more curious about how you'd deal with the constant of
integration problem when snipping bits out of a differential series before
integrating.

The same way you do-- by arguing that it is inaudible. ( and if you want
to put in a 50 Hz cutoff to the RIAA that will mean that that constant of
integration will disappear in about 1/20 second.)

Finding clicks
 

On 2014-09-11, Jim Lesurf wrote:
In article , Don Pearce
wrote:

Jim, have you tried differentiating twice - once as a very approximate
RIAA counter, and the second time to pull out the sharp edges into peaks?

Not yet, but it is what I have in mind to try sometime. However it isn't to
'un-do RIAA' but to enhance the click-to-music-ratio of clicks over the
bulk of the music.

FWIW in practice for *fixing* I find that

A) For *big* clicks / bangs the waveform is nonlinearly distorted as the
stylus clearly is going no-where near the intended waveform for a while as
it's not in contact with the groove walls as cut. The transducer will be
nonlinear in this situation as well even if I avoid clipping the 96k/24
recording. (Which does sometimes happen for these nasty bangs even though
I've set 0dBFS at about +18dBRIAA.)

B) For small ticks I tend to see a fast rise and a decay that looks
exponential at a rate set by the stylus suspension and tip mass. The decay
time is far too short to be set by the roll-off at LF of the replay RIAA
amp or arm-mass LF limit.

What is the that time scale of that decay? milliseconds? tenths of
milliseconds? The first might be caused by the set at 1KHz of RIAA? The
latter could not.


(A) isn't going to be truly fixable by any means because there's no data.
Might as well draw in a plausible shape by eye since you can't tell what
was cut on the LP.

That is always true. But there is data before or after the click. The
question is whether or not there is an effect of the click after the
click due to say the RIAA processing. Certainly any processing of the
data will produce tails. Is the situation improved by removing those
tails? Myy suspicion is yes. But I admit I have not done the experiments,
but would love to hear from someone who did. My theory suggests it would
be better. But ....


(B) is so short that simply doing a sensible interpolate as per Audacity's
'repair' gives a result where I can't here there was any problem.

I'll look at the differential when I can to see how it squares with what
Bill says. But the bottom line so far seems to be that for all but the most
dire clicks I can easily do repairs that render the result audibly fine.
(And the dire ones lack the info you'd need to get the intended waveform.)
The problem is finding some of them because they don't poke obviously out
of the waveform. Takes time. So I have to limit how much fixing I do
because I don't have the time to find all the tiny boogers hiding in the
waveforms.

Jim

Finding clicks
 

In article , William Unruh
wrote:
On 2014-09-11, Jim Lesurf wrote:
In article , Don Pearce
wrote:



B) For small ticks I tend to see a fast rise and a decay that looks
exponential at a rate set by the stylus suspension and tip mass. The
decay time is far too short to be set by the roll-off at LF of the
replay RIAA amp or arm-mass LF limit.

What is the that time scale of that decay? milliseconds?

It varies, so isn't easy to answer. As I've said in another posting I do
see examples where there's an exponential tail added to the waveform with a
time constant of the order of a millisecond or few. But I also see all
kinds of other waveforms. I suspect that the problem is that the wall
damage LPs acquire can end up having a wild variety of shapes and effects.

Small fast ticks sometimes show after-ringing in the 20kHz region. Which I
suspect is the stylus resonance.

tenths of
milliseconds? The first might be caused by the set at 1KHz of RIAA? The
latter could not.

What do you mean by the "set at 1kHz of RIAA"? Do you mean the
inter-constant 'shelf' in the departure from a simple integrator?


(A) isn't going to be truly fixable by any means because there's no
data. Might as well draw in a plausible shape by eye since you can't
tell what was cut on the LP.

That is always true. But there is data before or after the click. The
question is whether or not there is an effect of the click after the
click due to say the RIAA processing. Certainly any processing of the
data will produce tails. Is the situation improved by removing those
tails? Myy suspicion is yes. But I admit I have not done the
experiments, but would love to hear from someone who did. My theory
suggests it would be better. But ....

Well, at some point I hope to find out as I do plan to experiment with
differentiating the waveforms. But in practice this is getting a low
priority as I'm happily listening and recording and for most LPs I don't
bother to do any fixing. So I'll do it some time out of curiosity and then
report what I find.

Jim

--
Please use the address on the audiomisc page if you wish to email me.
Electronics http://www.st-and.ac.uk/~www_pa/Scot...o/electron.htm
Armstrong Audio http://www.audiomisc.co.uk/Armstrong/armstrong.html
Audio Misc http://www.audiomisc.co.uk/index.html

Finding clicks
 

On Mon, 08 Sep 2014 03:16:03 +0100, Johny B Good
wrote:

On Sun, 07 Sep 2014 16:44:21 +0100, Jim Lesurf
wrote:

In article , Don Pearce
wrote:
On Sun, 07 Sep 2014 14:33:16 GMT, (Don Pearce) wrote:

On Sun, 07 Sep 2014 15:03:37 +0100, Jim Lesurf
wrote:


However the problem I'm interested in is any algorithm for *finding*
(and
listing the positions of) clicks and ticks. The repair is the easy part,
although I'd always do that manually so I can the waveform before and
after. Sometimes a careless repair is worse that the original. :-)


I can certainly vouch for that effect when such declicking tools are
used indiscriminately!

Like Don, I've been using CoolEdit. In my case the Pro version 1.0,
since before the turn of the century (from around 1997). I still use
it to this day but I've let the audio processing jobs stagnate these
last few years. :-(

CoolEdit Pro does have some fairly comprehensive click and pop
removal tools (and the usual noise elimination facilities as well, of
course!) but these do need to be applied with some care.

I suspect that there has been very little improvement over the past
17 years in this regard.

I've skimmed through this thread and my first impression is that most of
the developments in the past 15 years seems to have gone by unnoticed. One
of the fastest ways to find clicks is to look in the frequency domain
(which is what Jim seems to be doing) but the newer versions of Audition
include a frequency/time view as standard so the worst of the clicks are
obvious. You can even select regions in the frequency/time view and just
tell the de-click algorithm to work in a certain frequency range for a
certain time. Audition 3 can be freely downloaded from Adobe (although
strictly speaking you should own a licence to run it but Adobe don't seem
worried).

Audition's tools seem to influenced by Cedar's Retouch software which
allows the same sort of manipulation but the click removal algorithms are
reputedly more sophisticated. Do Gordon Reid or Chris Hicks still hang out
here at all? If so, they would be the people who would know the full
details although I'm not sure how much they could divulge ;)

The other big player in this field nowadays is Izotope with their RX
software which has recently gone up to version 4. This seems to remove
clicks more transparently than Audition and has some useful visualisation
modes but I've not used it on a real project. I'd take a look at what it
can do before embarking on your own software.

James.
--
JRP Music - http://www.jrpmusic.co.uk

Finding clicks
 

Here is a link:
https://www.adobe.com/cfusion/entitl...=cs2_downloads

As far as I can tell this is a legitimate offer of 'obsolete' software
by Adobe. I understand that Audition 3 is part of cs2. There is also an
old version of Photoshop available on that page. Serial numbers are
included on the page.



Audition 3 can be freely downloaded from Adobe (although strictly
speaking you should own a licence to run it but Adobe don't seem
worried).