[Mxsmanic]

If I'm recording something digitally that I will pass on to others for use or
for further processing, what "average" level should I aim for in the recording
level? I've read figures like -6 dB, -3 dB, or 0 dB, so I'm confused.

With a 24-bit recording I calculate a maximum dynamic range of 72 dB, so I
guess in theory I could record at -10 dB and it would still work okay, but I
don't know. What do audio engineers expect if someone hands them a digital
recording?

I have the same question for peak levels. Should the peaks be allowed to reach
0 dB, or -3 dB, or what?

[Don Pearce[_3_]]

On Tue, 17 Jan 2012 17:15:44 +0100, Mxsmanic
wrote:

If I'm recording something digitally that I will pass on to others for use or
for further processing, what "average" level should I aim for in the recording
level? I've read figures like -6 dB, -3 dB, or 0 dB, so I'm confused.

With a 24-bit recording I calculate a maximum dynamic range of 72 dB, so I
guess in theory I could record at -10 dB and it would still work okay, but I
don't know. What do audio engineers expect if someone hands them a digital
recording?

I have the same question for peak levels. Should the peaks be allowed to reach
0 dB, or -3 dB, or what?

Your maths has let you down a bit. Multiply the number of bits by 6 to
get the dynamic range in dB. In practice 24 nit wont give you 144dB
because the analogue circuitry limits you long before this, but a
really good setup may yield perhaps 110dB.

But bear in mind that the noise level is usually set acoustically, and
the equipment plays very little part in it. So leave yourself
somewhere between 10 and 20dB headroom for unexpected peaks and you
will do OK. Of course once it is recorded, you can bring the level
back up digitally so you don't have too quiet a recording.

d

[Arny Krueger[_4_]]

"Mxsmanic" wrote in message
...

If I'm recording something digitally that I will pass on to others for use
or
for further processing, what "average" level should I aim for in the
recording
level? I've read figures like -6 dB, -3 dB, or 0 dB, so I'm confused.

Music has an inherent property called variously crest factor or peak to
average ratio. Once you set the average level or the peak level for
recording a given piece of music, you have implicitly set the other.

Crest factor can be 20 dB or more, but is generally never less than 7 or 8
dB for music. It is 3 dB for pure sine waves:
http://en.wikipedia.org/wiki/Crest_factor

You can change the crest factor of music by using dynamics processing AKA
compression or expansion. Of course, this can have either desirable or
undesirable effects on sound quality.

With a 24-bit recording I calculate a maximum dynamic range of 72 dB,

No, its more like 144 dB. Think of about 6.02 dB times the number of bits.

so I guess in theory I could record at -10 dB and it would still work
okay, but I don't know.

If all audio gear operated with the maximum dynamic range of a 24 bit system
you could give yourself a lot of headroom, but unfortunately nothing we use
is that good.

Things like rooms, microphones, microphone preamps, and real-world digital
converters perform at less-than-ideal levels. In the real world, dynamic
range on the order of 70 dB is really pretty good.

What do audio engineers expect if someone hands them a digital recording?

Peaks below FS, on a good day. ;-)

I have the same question for peak levels. Should the peaks be allowed to
reach 0 dB, or -3 dB, or what?

Setting recording levels for finished recordings so that the peaks are 1 dB
below FS is a good thing.

However, if you make a recording of several different pieces of music, and
you want people to be able to listen to them comfortably, you will end up
adjusting levels so that peak levels in each piece of music varies quite a
bit.

Ever see a digital recording of a concert or church service or seminar or
any other public event? The peaks will vary over a 10 dB or greater range,
depending which part of the event you are talking about.

[Adrian Tuddenham[_2_]]

Don Pearce wrote:

On Tue, 17 Jan 2012 17:15:44 +0100, Mxsmanic
wrote:

If I'm recording something digitally that I will pass on to others for use or
for further processing, what "average" level should I aim for in the
recording
level? I've read figures like -6 dB, -3 dB, or 0 dB, so I'm confused.

With a 24-bit recording I calculate a maximum dynamic range of 72 dB, so I
guess in theory I could record at -10 dB and it would still work okay, but I
don't know. What do audio engineers expect if someone hands them a digital
recording?

I have the same question for peak levels. Should the peaks be allowed to
reach
0 dB, or -3 dB, or what?

Your maths has let you down a bit. Multiply the number of bits by 6 to
get the dynamic range in dB. In practice 24 nit wont give you 144dB
because the analogue circuitry limits you long before this, but a
really good setup may yield perhaps 110dB.

Theoretically the thermal noise generated by a 600-ohm resistor at room
temperature measured at 20 Kc/s bandwidth is -124.9 dBm. Analogue
signal levels up to +20 dBm can be supplied by most good op-amps, so the
theoretical maximum S/N ratio is just about 144 dB.

As you say, in practice it will be nowhere near that.

--
~ Adrian Tuddenham ~
(Remove the ".invalid"s and add ".co.uk" to reply)
www.poppyrecords.co.uk

[Mxsmanic]

Don Pearce writes:

Your maths has let you down a bit. Multiply the number of bits by 6 to
get the dynamic range in dB. In practice 24 nit wont give you 144dB
because the analogue circuitry limits you long before this, but a
really good setup may yield perhaps 110dB.

I can never remember which applications of decibels use 20 log10 and which use
10 log10.

But bear in mind that the noise level is usually set acoustically, and
the equipment plays very little part in it. So leave yourself
somewhere between 10 and 20dB headroom for unexpected peaks and you
will do OK.

So the meters should hover around -10 dB?

Of course once it is recorded, you can bring the level
back up digitally so you don't have too quiet a recording.

What should I bring it up to?

[Mxsmanic]

Arny Krueger writes:

Peaks below FS, on a good day. ;-)

I have the same question for peak levels. Should the peaks be allowed to
reach 0 dB, or -3 dB, or what?

Setting recording levels for finished recordings so that the peaks are 1 dB
below FS is a good thing.

I guess I'll try for -1 dB, then.

Most of what I record isn't music, but just different types of sounds. I don't
always have the luxury of being able to measure the sound levels in advance in
order to set the recording level, but if I do, I can at least try to keep the
peaks at -1 dB. Outdoors in urban environments, though, there always seems to
be something that comes along that's incredibly loud--but if you set the
recording level to accommodate that, most of the rest of the recording is
quiet as a whisper.

Ever see a digital recording of a concert or church service or seminar or
any other public event? The peaks will vary over a 10 dB or greater range,
depending which part of the event you are talking about.

I was looking at some music I was listening to with Windows Media Player
today, and I saw what people mean by loudness wars. Some of the music scarcely
seemed to vary at all between the peaks and valleys in audio levels. And those
recordings did sound a bit muddy.

In contrast, the field recordings I've been trying to make show a huge dynamic
range, and then I have the problem of either cranking up the sound and risking
broken eardrums when the occasional loud peak comes along, or not cranking it
and straining to hear most of the recording over ambient noise in the room.

[PStamler]

Meanwhile, if you want some kind of standard...

There used to be this thing called a VU meter which measured levels
using averaging -- that is, it ignored momentary peaks and
concentrated on the average level of the signal. In the USA it became
near-universal as a level indicator, because two sound clips that gave
the same levels on a VU meter tended to have the same perceived volume
to the human ear. (The correlation wasn't perfect, but it was better
than most metering systems.) Also the peak-ignoring characteristics of
VU meters matched the characteristics of magnetic tape, which came
along a decade or so after the VU became popular -- momentary peaks on
tape tend to be inaudible.

So VU meters became near-universal in the USA; equipment was
calibrated to have some headroom over 0 VU, to allow for varying peak-
to-average ratios; typically this was 6dB in the early years, 8dB
later on, and 10dB still later.

Along came digital recording, where the paramount need is to avoid
having the signal hit 0dBFS, where most digital systems clip with
audibly unpleasant results. Peak meters were now de rigeur, and VU
meters vanished from everything except high-end studio gear (e.g.,
SSL) In some ways that's a pity, as peak meters offer zilch in the way
of audible matching of sound clips.

Anyway, there's still an argument to be made for VU meters in the
digital era, because they're better at matching the loudness
characteristics of the human ear. (As I said, not perfect, but better
than peak meters.) And the industry has adapted; an informal standard
exists for what 0 VU means on a digital system. To wit: in
broadcasting and film production, 0 VU = -20dBFS.

So if you have, say, a console with VU meters on it, and you put tone
through the console, set the tone so the meters read 0 VU, set the
digital recorder so that the console's output records at -20dBFS, then
operate the console so that program material makes the meters jump to
0 VU, you're operating according to this informal standard and should
get good recordings without overload. You're also making recordings
which won't surprise the next person you hand off the recording to, if
they have any familiarity with the standard.

And that standard, informal as it is, is the ONLY standard of which
I'm aware for level setting on digital recordings. Sad, but true.

Peace,
Paul

[William Sommerwerck]

I can never remember which applications of decibels
use 20 log10...

Voltage or current ratios

...and which use 10 log10.

Power ratios.

[Mxsmanic]

PStamler writes:

So if you have, say, a console with VU meters on it, and you put tone
through the console, set the tone so the meters read 0 VU, set the
digital recorder so that the console's output records at -20dBFS, then
operate the console so that program material makes the meters jump to
0 VU, you're operating according to this informal standard and should
get good recordings without overload. You're also making recordings
which won't surprise the next person you hand off the recording to, if
they have any familiarity with the standard.

And that standard, informal as it is, is the ONLY standard of which
I'm aware for level setting on digital recordings. Sad, but true.

Thanks! Isn't it possible to make digital peak meters behave like VU meters
by just averaging the peaks over a certain period of time, though? Then you
wouldn't need the analog meters.

[William Sommerwerck]

Never, ever let the peak levels hit 0dB. If you don't know where
your peaks are going to hit, it's better to leave more headroom
than less. I tend to set levels for peaks around -12dB and expect
that something sooner or later will be louder than predicted.

When I recorded live, I'd ask the orchestra to hit the loudest they'd play
for that piece, then set the peak level 1 or 2dB below that. Because when
they always played a bit louder in concert.

Recording analog with dbx II was a snap. I set the pre-concert audience
noise to -10dB or -20dB (I forget which) and never had a problem.

[Scott Dorsey]

Mxsmanic wrote:
If I'm recording something digitally that I will pass on to others for use or
for further processing, what "average" level should I aim for in the recording
level? I've read figures like -6 dB, -3 dB, or 0 dB, so I'm confused.

Don't worry about it. Worry about the peak level. Let someone else worry
about the average level.

With a 24-bit recording I calculate a maximum dynamic range of 72 dB, so I
guess in theory I could record at -10 dB and it would still work okay, but I
don't know. What do audio engineers expect if someone hands them a digital
recording?

No, try 144 dB. In actuality you'll be lucky to get more than 16 usable bits
in a typical 24-bit system but that's still 96 dB.

I have the same question for peak levels. Should the peaks be allowed to reach
0 dB, or -3 dB, or what?

Never, ever let the peak levels hit 0dB. If you don't know where your peaks
are going to hit, it's better to leave more headroom than less. I tend to
set levels for peaks around -12dB and expect that something sooner or later
will be louder than predicted.
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."

[Scott Dorsey]

Mxsmanic wrote:

Thanks! Isn't it possible to make digital peak meters behave like VU meters
by just averaging the peaks over a certain period of time, though? Then you
wouldn't need the analog meters.

You don't really need them in the digital world today.

But if you do, RTW and Dorrough will sell you some very nicely made digital
meters with peak and average-reading modes, including some that make them
behave like old-fashioned VU meters.
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."

[Mxsmanic]

Scott Dorsey writes:

Never, ever let the peak levels hit 0dB. If you don't know where your peaks
are going to hit, it's better to leave more headroom than less. I tend to
set levels for peaks around -12dB and expect that something sooner or later
will be louder than predicted.

Is it possible to "repair" clipped peaks so that they don't sound so
distorted, if by accident a recording contains them? Maybe by smoothing out
the waveform or something?

[Mxsmanic]

William Sommerwerck writes:

When I recorded live, I'd ask the orchestra to hit the loudest they'd play
for that piece, then set the peak level 1 or 2dB below that. Because when
they always played a bit louder in concert.

If only I could persuade passing teenagers on ear-busting scooters to do the
same!

Recording analog with dbx II was a snap. I set the pre-concert audience
noise to -10dB or -20dB (I forget which) and never had a problem.

Why wouldn't this work for digital recording?

[Trevor]

"Scott Dorsey" wrote in message
...
Mxsmanic wrote:

Thanks! Isn't it possible to make digital peak meters behave like VU
meters
by just averaging the peaks over a certain period of time, though? Then
you
wouldn't need the analog meters.

You don't really need them in the digital world today.

But if you do, RTW and Dorrough will sell you some very nicely made
digital
meters with peak and average-reading modes, including some that make them
behave like old-fashioned VU meters.

And if you are recording on computer, many DAW software packages have both
peak and VU metering (often to various standards as well since that is easy
to do in software), as well as plug-ins that do the same job.

Trevor.

[Trevor]

"William Sommerwerck" wrote in message
...
When I recorded live, I'd ask the orchestra to hit the loudest they'd play
for that piece, then set the peak level 1 or 2dB below that. Because when
they always played a bit louder in concert.

You expect them to play a bit louder, and only leave "1 or 2dB" headroom? I
guess that's analog you are talking about then.

Trevor.

[Trevor]

"Mxsmanic" wrote in message
...
Is it possible to "repair" clipped peaks so that they don't sound so
distorted, if by accident a recording contains them? Maybe by smoothing
out
the waveform or something?

Yep, both by predictive interpolation and smoothing. FAR better to avoid
needing such drastic measures as much as possible.
Of course most modern music is deliberately clipped to hell and back after
recording/mixing, so it doesn't matter much if it's clipped a bit in the
recording process anyway! :-(

Trevor.

[William Sommerwerck]

"Mxsmanic" wrote in message
...
William Sommerwerck writes:

Recording analog with dbx II was a snap. I set the pre-concert audience
noise to -10dB or -20dB (I forget which) and never had a problem.

Why wouldn't this work for digital recording?

No, for two reasons. The dynamic range was compressed 2:1. And you had about
6dB headroom above the nominal 0dB level.

[William Sommerwerck]

"Trevor" wrote in message
...
"William Sommerwerck" wrote in message
...

When I recorded live, I'd ask the orchestra to hit the loudest they'd
play
for that piece, then set the peak level 1 or 2dB below that. Because
they always played a bit louder in concert.

You expect them to play a bit louder, and only leave "1 or 2dB" headroom?
I guess that's analog you are talking about then.

Nope, it was digital. And that was the way it worked.

[Frank Stearns]

Mxsmanic writes:

Scott Dorsey writes:

Never, ever let the peak levels hit 0dB. If you don't know where your peaks
are going to hit, it's better to leave more headroom than less. I tend to
set levels for peaks around -12dB and expect that something sooner or later
will be louder than predicted.

Is it possible to "repair" clipped peaks so that they don't sound so
distorted, if by accident a recording contains them? Maybe by smoothing out
the waveform or something?

Yes, but it can be damned tedious.

Say you have some "clean" clips, where the wavetop for 10-30 samples is flat across.
(If the clips are wider, you've got more problems than you probably want to deal
with using this manual method.)

Using your favorite higher-performance digital editor, first reduce the overall file
level by, say, 6 dB. That will give you some room to play.

Zoom to sample resolution at the first clip (some editors have search functions to
find clips), then use the pencil tool to draw your best guess as to where the
waveform would have gone had it not hit the ceiling. If it looks like that 6 dB
reduction hasn't given you enough room, undo what you just tried to draw and reduce
the file level another 3-6 dB.

Given that you're zoomed at the sample level, don't make any sudden excursions in
your redraw, or you'll get a nasty pop or click. Keep what you draw smooth and
graceful. (You still might get a "whump" so be careful.)

You might even look at a few cycles before and after the clip to see what those
waveforms look like, as they are likely close "relatives" (in terms of the
information recorded) to the neighboring cycles that clipped.

Repeat for the next clip occurrences, positive and negative. There might be quite a
few, depending on what happened.

Now, there are some automatic clip repair tools in some of the editing platforms,
but I've not played with the newer ones. The older ones didn't work all that well
(some were just plain brain dead).

A common mistake is trying to reshape the slope leading to and from the clip point
so as to avoid adjusting the overall level in the file. Don't do it! Reduce the
level first, then with your redraw try to be true as to where the waveform would
have gone.

Better, of course, to never have clipped in the first place (particularly in this
age of 24 bit). Headroom is your friend.

But, sometimes crap happens and you can get away with a little "body and fender"
work to save that one special recording.

Good luck with it,

Frank
Mobile Audio
--

[William Sommerwerck]

No. Once information is lost, you cannot get it back. You
cannot reconstruct the top of the waveform because you
don't know what it looked like before it was chopped off.

But you do know the slopes of the leading and trailing edges. You can
interpolate from this. It might not be perfectly correct, but it would
reduce the distortion.

[Scott Dorsey]

In article ,
Mxsmanic wrote:
Scott Dorsey writes:

Never, ever let the peak levels hit 0dB. If you don't know where your peaks
are going to hit, it's better to leave more headroom than less. I tend to
set levels for peaks around -12dB and expect that something sooner or later
will be louder than predicted.

Is it possible to "repair" clipped peaks so that they don't sound so
distorted, if by accident a recording contains them? Maybe by smoothing out
the waveform or something?

No. Once information is lost, you cannot get it back. You cannot
reconstruct the top of the waveform because you don't know what it
looked like before it was chopped off.

There are some cheesy tricks that will make the clipping less offensive
sounding, but nothing actually restores it to the way it would have
sounded if the levels had been set properly.

This is not 1985 any longer. We have outrageous, positively outrageous
amounts of dynamic range available to us. There is absolutely zero excuse
for clipping anything today.
--scott

--
"C'est un Nagra. C'est suisse, et tres, tres precis."

[Roy W. Rising[_2_]]

PStamler wrote:
Meanwhile, if you want some kind of standard...

There used to be this thing called a VU meter which measured levels
using averaging -- that is, it ignored momentary peaks and
concentrated on the average level of the signal. In the USA it became
near-universal as a level indicator, because two sound clips that gave
the same levels on a VU meter tended to have the same perceived volume
to the human ear. (The correlation wasn't perfect, but it was better
than most metering systems.) Also the peak-ignoring characteristics of
VU meters matched the characteristics of magnetic tape, which came
along a decade or so after the VU became popular -- momentary peaks on
tape tend to be inaudible.

So VU meters became near-universal in the USA; equipment was
calibrated to have some headroom over 0 VU, to allow for varying peak-
to-average ratios; typically this was 6dB in the early years, 8dB
later on, and 10dB still later.

Along came digital recording, where the paramount need is to avoid
having the signal hit 0dBFS, where most digital systems clip with
audibly unpleasant results. Peak meters were now de rigeur, and VU
meters vanished from everything except high-end studio gear (e.g.,
SSL) In some ways that's a pity, as peak meters offer zilch in the way
of audible matching of sound clips.

Anyway, there's still an argument to be made for VU meters in the
digital era, because they're better at matching the loudness
characteristics of the human ear. (As I said, not perfect, but better
than peak meters.) And the industry has adapted; an informal standard
exists for what 0 VU means on a digital system. To wit: in
broadcasting and film production, 0 VU = -20dBFS.

So if you have, say, a console with VU meters on it, and you put tone
through the console, set the tone so the meters read 0 VU, set the
digital recorder so that the console's output records at -20dBFS, then
operate the console so that program material makes the meters jump to
0 VU, you're operating according to this informal standard and should
get good recordings without overload. You're also making recordings
which won't surprise the next person you hand off the recording to, if
they have any familiarity with the standard.

And that standard, informal as it is, is the ONLY standard of which
I'm aware for level setting on digital recordings. Sad, but true.

Peace,
Paul

.... what Mr. Stamler said. Thanks, Paul!

I'll just add that transistor clipping is about the same enemy as 0 dBFS.
Somewhere I have some old ABC Labs research that found allowance for 16 dB
"peak factor" was the minimum for a broad range of un-processed material.
Systems design grabbed onto the convenient number of 18 dB "headroom". It
is interesting that the subsequent 0 VU = -20 dBFS is similarly protective.
It's a crying shame that AVID hit the market with 0 VU = -14 dBFS ...
"unsafe at any speed". ;-) Long live the lowly VU meter!

--
~ Roy
"If you notice the sound, it's wrong!"

[Doug McDonald[_6_]]

On 1/17/2012 6:49 PM, William Sommerwerck wrote:
No. Once information is lost, you cannot get it back. You
cannot reconstruct the top of the waveform because you
don't know what it looked like before it was chopped off.

But you do know the slopes of the leading and trailing edges. You can
interpolate from this. It might not be perfectly correct, but it would
reduce the distortion.



Yes, this works for short, a few samples, overloads. But for longer
ones, even say 10 samples at CD rate, the missing waveform can be
much more complicated than a simple parabola or third order curve.

I wrote a simple program to do what you say, and it is indeed
better than nothing, by quite a bit.

This was done to repair some Bis recording that are otherwise
wonderful. These were some of the first few ever made. Robert
von Bahr didn't even know they were clipped ... but his current
chief engineer did!

Doug McDonald

[Mike Rivers]

On 1/17/2012 3:25 PM, Mxsmanic wrote:

Isn't it possible to make digital peak meters behave like VU meters
by just averaging the peaks over a certain period of time, though? Then you
wouldn't need the analog meters.

There are digital meters that display an average level, but as Paul
explained, what's really important for recording is the peak level. Once
you have the recording, you can start fooling with the peak-to-average
ratio to make it audibly louder. This doesn't change the peak level, it
only increases the average level.

Some DAW programs and aftermarket plug-in meters have a side-by-side bar
graph with one bar representing peak level and the other representig
average. It's educational to watch a meter like this when playing a
commercial "loud" recording then compare it to a recording straight off
the microphone that you made yourself. You probably can't afford a
Dorrough meter, but if there's a standard meter for this sort of work,
that's it.

There's no standard for average level, but there are a couple of
standards and practices for peak level. SMPTE wants no peaks above -10
dBFS, and an "eyeball average" 10 dB below that. Good practice when
you're going to send a file to someone for mastering is -3 dBFS so
they'll have some room to work. And the way to get that is by
controlling the mix or recording level, not letting it fall where it may
and putting a limiter on it to keep the peaks below -3 dBFS. They'll
know you're cheating if you do that.


--
"Today's production equipment is IT based and cannot be operated without
a passing knowledge of computing, although it seems that it can be
operated without a passing knowledge of audio" - John Watkinson

Drop by http://mikeriversaudio.wordpress.com now and then

[Trevor]

"William Sommerwerck" wrote in message
...
"Trevor" wrote in message
...
"William Sommerwerck" wrote in message
...

When I recorded live, I'd ask the orchestra to hit the loudest they'd
play
for that piece, then set the peak level 1 or 2dB below that. Because
they always played a bit louder in concert.

You expect them to play a bit louder, and only leave "1 or 2dB" headroom?
I guess that's analog you are talking about then.

Nope, it was digital. And that was the way it worked.


So less than 1 to 2dB increase in level then. I guess you did say a "bit"
louder, a *very little* bit obviously, and you weren't one for leaving much
headroom either. I'd rather leave a "Bit" as in 6dB myself.

Trevor.

[Mike Rivers]

On 1/17/2012 2:49 PM, Mxsmanic wrote:

Most of what I record isn't music, but just different types of sounds. I don't
always have the luxury of being able to measure the sound levels in advance in
order to set the recording level

It's not a luxury, it's a necessity, and that's why they put knobs on
things - so you can adjust them. If you just want to walk around with a
microphone ane capture whatever comes along, then set the record level
low enough so that you can record something pretty loud without
clipping. If something louder comes along, turn the record level control
down. You might lose the first couple of seconds, but at least you'll
have something usable as long as it's a repeating or continuous sound.


--
"Today's production equipment is IT based and cannot be operated without
a passing knowledge of computing, although it seems that it can be
operated without a passing knowledge of audio" - John Watkinson

Drop by http://mikeriversaudio.wordpress.com now and then

[Scott Dorsey]

William Sommerwerck wrote:
No. Once information is lost, you cannot get it back. You
cannot reconstruct the top of the waveform because you
don't know what it looked like before it was chopped off.

But you do know the slopes of the leading and trailing edges. You can
interpolate from this. It might not be perfectly correct, but it would
reduce the distortion.

And people do that.

But.... maybe what was chopped off was a peak... and maybe it was two peaks...
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."

[Trevor]

"Frank Stearns" wrote in message
acquisition...
Is it possible to "repair" clipped peaks so that they don't sound so
distorted, if by accident a recording contains them? Maybe by smoothing
out
the waveform or something?

Yes, but it can be damned tedious.

Say you have some "clean" clips, where the wavetop for 10-30 samples is
flat across.
(If the clips are wider, you've got more problems than you probably want
to deal
with using this manual method.)

Using your favorite higher-performance digital editor, first reduce the
overall file
level by, say, 6 dB. That will give you some room to play.

Zoom to sample resolution at the first clip (some editors have search
functions to
find clips), then use the pencil tool to draw your best guess as to where
the
waveform would have gone had it not hit the ceiling. If it looks like that
6 dB
reduction hasn't given you enough room, undo what you just tried to draw
and reduce
the file level another 3-6 dB.

Given that you're zoomed at the sample level, don't make any sudden
excursions in
your redraw, or you'll get a nasty pop or click. Keep what you draw smooth
and
graceful. (You still might get a "whump" so be careful.)

You might even look at a few cycles before and after the clip to see what
those
waveforms look like, as they are likely close "relatives" (in terms of the
information recorded) to the neighboring cycles that clipped.

Repeat for the next clip occurrences, positive and negative. There might
be quite a
few, depending on what happened.

Now, there are some automatic clip repair tools in some of the editing
platforms,
but I've not played with the newer ones. The older ones didn't work all
that well
(some were just plain brain dead).


There were some over a decade ago that did a fair bezier interpolation, as
good as you'd do by hand at least.
But nothing beats getting it right in the first place.

Trevor.

[Trevor]

"Scott Dorsey" wrote in message
...
Is it possible to "repair" clipped peaks so that they don't sound so
distorted, if by accident a recording contains them? Maybe by smoothing
out
the waveform or something?

No. Once information is lost, you cannot get it back. You cannot
reconstruct the top of the waveform because you don't know what it
looked like before it was chopped off.

Right, you have to guess it is just the top of a single curve, and is only
satisfactory for minor clipping. As with all editing you simply apply the
"fix" and see if it sounds better or not.
(and record it properly in future!)


This is not 1985 any longer. We have outrageous, positively outrageous
amounts of dynamic range available to us. There is absolutely zero excuse
for clipping anything today.

Well not at the recording stage anyway. In the mastering stage they
sometimes do what they are told. At least I assume that's the excuse for
millions of severely clipped CD's currently available! Not all of the
mastering engineers are unaware of what they are doing.

Trevor.

[cedricl[_2_]]

On Jan 17, 2:34*pm, Mxsmanic wrote:
Scott Dorsey writes:

Is it possible to "repair" clipped peaks so that they don't sound so
distorted, if by accident a recording contains them? Maybe by smoothing out
the waveform or something?

BIAS Peak Pro 7 has a DSP tool called, "Declipper" that supposedly
works to do this. I haven't tried the tool but it has different
settings for sensitivity which one could adjust for varying results.
Go to their website and do some research on it. It might be something
you can use.

[Don Pearce[_3_]]

On Tue, 17 Jan 2012 20:41:46 +0100, Mxsmanic
wrote:

Don Pearce writes:

Your maths has let you down a bit. Multiply the number of bits by 6 to
get the dynamic range in dB. In practice 24 nit wont give you 144dB
because the analogue circuitry limits you long before this, but a
really good setup may yield perhaps 110dB.

I can never remember which applications of decibels use 20 log10 and which use
10 log10.


If you compare voltages, use 20 log, an d if it is power, use 10 log.

But bear in mind that the noise level is usually set acoustically, and
the equipment plays very little part in it. So leave yourself
somewhere between 10 and 20dB headroom for unexpected peaks and you
will do OK.

So the meters should hover around -10 dB?

Or a little lower unless you are reasonably certain of the level of
what you are recording.

Of course once it is recorded, you can bring the level
back up digitally so you don't have too quiet a recording.

What should I bring it up to?

That's to taste, and it depends on how you are going to treat it
subsequently. If you are just going to use it exactly as you recorded
it, then bring it up so that the highest peak in the whole session
just touches full scale.

d

[PStamler]

On Jan 17, 10:29*pm, cedricl wrote:
On Jan 17, 2:34*pm, Mxsmanic wrote:

Scott Dorsey writes:
Is it possible to "repair" clipped peaks so that they don't sound so
distorted, if by accident a recording contains them? Maybe by smoothing out
the waveform or something?

BIAS Peak Pro 7 has a DSP tool called, "Declipper" that supposedly
works to do this. I haven't tried the tool but it has different
settings for sensitivity which one could adjust for varying results.
Go to their website and do some research on it. It might be something
you can use.

Adobe Audition has a similar tool. I used it once on some clipped DAT
files. On the mildly clipped ones (a couple of dB into clipping) it
did a fair job of restoration. One the badly clipped ones (6 or more
dB into clipping) the material still sounded clipped, but it sounded
more like analog clipping than digital clipping. Which is to say,
still dirty and distorted, but not as run-screaming-from-the-room
distorted as before the fix. Under the circumstances, that was the
best I could hope for. Going back and redoing the dubs wasn't an
option, since the source tapes were two states away and air time was
in 12 hours.

Peace,
Paul

[Mxsmanic]

Mike Rivers writes:

Some DAW programs and aftermarket plug-in meters have a side-by-side bar
graph with one bar representing peak level and the other representig
average. It's educational to watch a meter like this when playing a
commercial "loud" recording then compare it to a recording straight off
the microphone that you made yourself.

I've tried it with the consumer version of Sound Forge (all that I can
afford), and it was indeed instructive. Some songs that I've pulled from CDs
show an average of perhaps -6 db and a range of only about 9 dB. One nice
exception is a von Karajan CD that spans 45 dB, although I'm not sure what the
average is because the dynamic range is so large (Sound Forge just has two
meters and some lines that show recent peaks, but no average levels).

When I record my own voice or make my own field recordings, the dynamic range
is vast compared to most pre-recorded stuff like music that I've been looking
at.

You probably can't afford a
Dorrough meter, but if there's a standard meter for this sort of work,
that's it.

They look nice, which guarantees that I cannot afford them.

There's no standard for average level, but there are a couple of
standards and practices for peak level. SMPTE wants no peaks above -10
dBFS, and an "eyeball average" 10 dB below that.

Of course, the next question that comes to mind is: if you have absolutely no
peaks above -10 dB, then what are you gaining from that top 10 dB?

I suppose it could provide some margin if the material might be mixed with
something else, but if it's going to be used alone ... I don't know.

Good practice when
you're going to send a file to someone for mastering is -3 dBFS so
they'll have some room to work. And the way to get that is by
controlling the mix or recording level, not letting it fall where it may
and putting a limiter on it to keep the peaks below -3 dBFS. They'll
know you're cheating if you do that.

I try to adjust the original recording level to get it into the range I want.

A lot of stuff I use myself, but sometimes I upload interesting sounds to
freebie sound sites, and that's why I've asked if there's some sort of
standard to follow for average and peak levels.

[Mxsmanic]

Don Pearce writes:

If you compare voltages, use 20 log, an d if it is power, use 10 log.

The problem is that I'm never sure whether the quantity in question is being
referenced to voltage or power. I guess sound pressure levels are acoustic
power.

That's to taste, and it depends on how you are going to treat it
subsequently. If you are just going to use it exactly as you recorded
it, then bring it up so that the highest peak in the whole session
just touches full scale.

Is anything below 0 dB generally safe in terms of being reasonably
distortion-free? And how far down can the quiet parts go before they are
likely to be lost, distorted, or drowned out by noise?

I notice that below -50 dB I don't really hear anything in the headphones
(although I keep them set quite low), even though the meters go down to -90 dB
and occasionally show _something_ in that range.

[Mxsmanic]

Mike Rivers writes:

It's not a luxury, it's a necessity, and that's why they put knobs on
things - so you can adjust them. If you just want to walk around with a
microphone ane capture whatever comes along, then set the record level
low enough so that you can record something pretty loud without
clipping.

Is there any disadvantage to recording with a really low recording level (one
low enough to handle just about anything)? Does normalization after the fact
diminish the quality of a recording a lot? I was thinking of just recording
everything at really low levels than normalizing it later, but I don't know
what effect that has on the sound quality.

If something louder comes along, turn the record level control
down. You might lose the first couple of seconds, but at least you'll
have something usable as long as it's a repeating or continuous sound.

I've discovered that there are many unpleasant surprises in urban
environments. Vehicles make a lot of noise, of course, but two-wheeled
vehicles are way worse than anything else. And buses are loud,
surprisingly--it seems that they make a lot of low-frequency noise that
perhaps isn't as subjectively noticeable but is nevertheless quite loud.
Then there are sirens, etc., that seem to come out of nowhere.

[Mxsmanic]

Trevor writes:

Of course most modern music is deliberately clipped to hell and back after
recording/mixing, so it doesn't matter much if it's clipped a bit in the
recording process anyway!

Why is it deliberately clipped? I've been reading that clipping is a terrible,
horrible thing for digital audio.

[Mxsmanic]

Frank Stearns writes:

A common mistake is trying to reshape the slope leading to and from the clip point
so as to avoid adjusting the overall level in the file. Don't do it!

That would leave unwanted harmonics, right? If the slope doesn't match a sine
wave, that is.

[Mxsmanic]

Scott Dorsey writes:

No. Once information is lost, you cannot get it back. You cannot
reconstruct the top of the waveform because you don't know what it
looked like before it was chopped off.

You could guess that it was just an extension of a sine wave, though, right?

Nevertheless, I imagine this would get tedious when adjusting 8 kHz tones for
30 seconds at a time.

This is not 1985 any longer. We have outrageous, positively outrageous
amounts of dynamic range available to us. There is absolutely zero excuse
for clipping anything today.

If I set the recording level so that it averages -40 dB with 24-bit sound, do
I risk losing anything, or is it all just numbers? In other words, if I
reduce the recording level on my little recorder, is it changing something on
the analog side (which could have repercussions), or is it just juggling
numbers during or after the analog-to-digital conversion (which potentially
would not have any consequences, since analog electronic circuits are by then
out of the picture)?

[Anahata]

On Wed, 18 Jan 2012 06:13:53 +0000, Don Pearce wrote:

What should I bring it up to?

That's to taste, and it depends on how you are going to treat it
subsequently. If you are just going to use it exactly as you recorded
it, then bring it up so that the highest peak in the whole session just
touches full scale.

Preferably slightly less, like 0.5dB below FS, as the antialiasing
processing in some D-A converters can overshoot and clip.

--
Anahata
--/-- http://www.treewind.co.uk
+44 (0)1638 720444