[Carey Carlan]

Bob Olhsson wrote in news:GPHwh.15011$fC2.6662
@bignews4.bellsouth.net:

The only time "normalization" can possibly buy you anything is in the
final D to A conversion. Otherwise it only adds noise or distortion if
you fail dither your gain change.

Yes, you must dither after the amplification, but normalizing in 24 bits
before resampling to 16 bits with dither is harmless.

I normalize every recording I make. As the last step while still in 24
bits, I amplify such that the loudest sample is -1 dBFS. I then resample
to 16 bits and dither, then go to CD.

[Mike Rivers]

Bob Olhsson wrote:

The key word here is "necessary." If you're just going to be changing
the volume again in the final mix, normalizing only adds noise and/or
distortion.

To most people who have a desire to normalize, they're talking about
the final mix. At least until they change something else. g

[Laurence Payne]

On 2 Feb 2007 07:54:53 -0800, "Mike Rivers"
wrote:

To most people who have a desire to normalize, they're talking about
the final mix. At least until they change something else. g

A lot of people who build up music by multitrack recording routinely
normalise every recorded file, before starting to mix.

Bit like driving a car. Those of us who CAN change an old-fashioned
gearbox enjoy looking down on youngsters who rely on synchromesh or
(Lord protect us) automatic gearboxes :-) But you get there just the
same. Somewhat faster, often.

[Laurence Payne]

On Fri, 02 Feb 2007 16:45:18 +0000, Laurence Payne
lpayne1NOSPAM@dslDOTpipexDOTcom wrote:

On 2 Feb 2007 07:54:53 -0800, "Mike Rivers"
wrote:

To most people who have a desire to normalize, they're talking about
the final mix. At least until they change something else. g

A lot of people who build up music by multitrack recording routinely
normalise every recorded file, before starting to mix.

Bit like driving a car. Those of us who CAN change an old-fashioned
gearbox enjoy looking down on youngsters who rely on synchromesh or
(Lord protect us) automatic gearboxes :-) But you get there just the
same. Somewhat faster, often.

A paragraph in the middle of that got lost. I meant to include:

"It's sloppy. But with today's quiet mics and preamps, 24-bit
recording systems and ample processing power, it's sloppiness that
probably doesn't hurt the result."

[Mike Rivers]

Laurence Payne wrote:


A lot of people who build up music by multitrack recording routinely
normalise every recorded file, before starting to mix.

I suppose that gets the noise (electronic and ambient) level up to a
nice even level. I suppose modern DAWs manage internal word length OK
so you don't 'clip' when you sum a bunch of tracks that are running
near full scale, but I know this would require a lot of padding on the
inputs of any of my analog mixers in order to mix 24 tracks coming out
of the D/A converters with peaks at +22 dBu or more.

[Laurence Payne]

On 2 Feb 2007 12:42:23 -0800, "Mike Rivers"
wrote:

A lot of people who build up music by multitrack recording routinely
normalise every recorded file, before starting to mix.

I suppose that gets the noise (electronic and ambient) level up to a
nice even level. I suppose modern DAWs manage internal word length OK
so you don't 'clip' when you sum a bunch of tracks that are running
near full scale, but I know this would require a lot of padding on the
inputs of any of my analog mixers in order to mix 24 tracks coming out
of the D/A converters with peaks at +22 dBu or more.

Yeah. As I indicated in my addendum, there WOULD have been a noise
issue in the old days. But things have changed a lot. How things
worked in analogue often ain't how things work in digital.

[ernest]

Normalizing is often frowned upon by pseudo (and some real)
"professionals," not because it does any damage, but because it
shouldn't be necessary because you should have set the record level so
that it reaches the desired playback level. The damage, if any, has
already been done in the recording process, and you're not making
anything any better by normalizing . . . except for one thing - the
listener is insulated from your "mistake" and doesn't have to turn up
the volume to hear it at at the same level as the last thing he listened
to.

Please help me out here Mike. It sounds as if you're saying
during tracking time, one should record at the desired
playback level. On the preamp, I always do it peak 0dBVu.
But on DAW, I kept hearing other say I should hit it between
-15 to -18 dBFS and that's sufficient. But when you say
'desired playback level' it almost mean to hit it much
higher in terms of dBFS. Or you are implying people don't
wanna turn up their volume these days, and/or people are
recording/mixing way too loud? Or whoever said -15 to -18
dBFS is sufficient is simply wrong?

[Laurence Payne]

On 2 Feb 2007 14:37:56 -0800, "ernest" wrote:

Please help me out here Mike. It sounds as if you're saying
during tracking time, one should record at the desired
playback level. On the preamp, I always do it peak 0dBVu.
But on DAW, I kept hearing other say I should hit it between
-15 to -18 dBFS and that's sufficient. But when you say
'desired playback level' it almost mean to hit it much
higher in terms of dBFS. Or you are implying people don't
wanna turn up their volume these days, and/or people are
recording/mixing way too loud? Or whoever said -15 to -18
dBFS is sufficient is simply wrong?

We seem to be jumping between the issues of normalising tracks in a
mixing environment an of normalising a finished mix.

In the day, we recorded analogue tracks hot in order to get good s/n,
and maybe to achieve the particular distortion peculiar to slightly
overloaded tape. Now we place the record level conservatively within
our 24 available bits, and the only sin is overload. Noise floors
are low, so if we find it convenient to trim a channel up (or down),
it's a perfectly acceptable technique.

A completed mix may be nudging full-scale, may be several dB down. The
music will have a finite dynamic range, but we've got a lot of bits in
which to position that range. If it's convenient to position it
higher (or lower) at this stage, so what?

We haven't got an in finite dynamic window, or an infinitely low noise
floor. We can still store up trouble by using too sloppy a gain
structure. But we have much more usable headroom than in the analogue
days. (Rather more useful to look at it as "bottom-room" though.)

[Mike Rivers]

Laurence Payne wrote:

Yeah. As I indicated in my addendum, there WOULD have been a noise
issue in the old days. But things have changed a lot. How things
worked in analogue often ain't how things work in digital.

I actually had my tongue planted in my cheek when I made the comment
about noise. I wasn't talking about the same noise as Bob was talking
about, I was talking about raising the volune level of the computer
fan being picked up by the microphone, and the hum from the guitar
pickups, and the ground loop in the keyboard. Presumably those weren't
a problem at the level that they were recorded but crank them (along
with everything else) up by 10 or 12 dB with normalization and when
you solo a track you say "What's all that noise that didn't used to be
there?"

[Mike Rivers]

ernest wrote:

Please help me out here Mike. It sounds as if you're saying
during tracking time, one should record at the desired
playback level. On the preamp, I always do it peak 0dBVu.
But on DAW, I kept hearing other say I should hit it between
-15 to -18 dBFS and that's sufficient.

Generally they should occur at about the same point. When your preamp
or mixer's meter is showing around 0 VU (assuming you have 20 dB or so
of headroom above that) then your A/D converter sensitivity should be
such that your're getting a digital level of around -15 to -20 dBFS.
That way, both the mic preamp and the A/D converter reach their
maximum level at about the same point.

But when you say
'desired playback level' it almost mean to hit it much
higher in terms of dBFS.

No, not really. If you're mixing a bunch of tracks that run in the -15
dBFS range, unless you're just compressing the life out of them before
you record, they'll have peaks that get pretty close to full scale.
And when you add them together, they'll be plenty loud. You can always
make the mix louder once it's established if that's what you want to
do, but if all your tracks are running pretty hot and then you mix
them, you'll have to pull the levels down to prevent the sum from
"overloading." And raising the level with normalization and then
cutting it back with the mix faders is counterproductive.

Or you are implying people don't
wanna turn up their volume these days, and/or people are
recording/mixing way too loud? Or whoever said -15 to -18
dBFS is sufficient is simply wrong?

-15 to -18 dBFS nominal level is good. It leaves 15 to 18 dB of
headroom for peaks that naturally occur in music.

It's true that people don't want to turn their volume control up. And
with their engineer hat on, they feel that if they have to turn up the
listening volume, it will be too low to compete with commercial
recordings. And they'd be right. But the place to make that adjustment
is after the mix is completed, not by rasiing the level of every track.

[Albert]

On Jan 10, 8:51 pm, jtougas wrote:
Hi folks -

In my pursuit of improving my recordings, I've taken a few of my mixes
to a friend of mine, to see what he thinks, and I mentioned I'd
normalized one or two of the channels in the tracks since they were a
little quieter than the others.

If I feel the need to do that I usually increase the gain by 6 dB or a
multiple thereof. Say if the highest peak in the file is - 17dB, I
change gain by +12dB. The thought there is that a 6 dB change is a
simple doubling, whereas any other amount (like an amount that might
be applied by normalising) is more complex, giving a bigger chance of
unwanted artifacts.

Albert

[David Morgan \(MAMS\)]

"Albert" wrote in message oups.com...
On Jan 10, 8:51 pm, jtougas wrote:
Hi folks -

In my pursuit of improving my recordings, I've taken a few of my mixes
to a friend of mine, to see what he thinks, and I mentioned I'd
normalized one or two of the channels in the tracks since they were a
little quieter than the others.

If I feel the need to do that I usually increase the gain by 6 dB or a
multiple thereof.

You've lost me... what do sixes, twelves, eighteens and twenty-fours have to
do with anything?

Are you saying that increasing volume by 200%, 398.11% or 794.33% (etc.) is
somehow beneficial to a basic normalisation?

Say if the highest peak in the file is - 17dB, I change gain by +12dB.

# 1.... why don't you just use a fader at mixtime?

# 2.... why didn't you just normalize to -5 dbfs

# 3.... what's the mathematical difference between # 2 and your gain change?

More later... ;-)



DM

[Don Pearce]

On 2 Feb 2007 23:42:12 -0800, "Albert" wrote:

On Jan 10, 8:51 pm, jtougas wrote:
Hi folks -

In my pursuit of improving my recordings, I've taken a few of my mixes
to a friend of mine, to see what he thinks, and I mentioned I'd
normalized one or two of the channels in the tracks since they were a
little quieter than the others.

If I feel the need to do that I usually increase the gain by 6 dB or a
multiple thereof. Say if the highest peak in the file is - 17dB, I
change gain by +12dB. The thought there is that a 6 dB change is a
simple doubling, whereas any other amount (like an amount that might
be applied by normalising) is more complex, giving a bigger chance of
unwanted artifacts.

Albert

6dB is not doubling - it i just a little less than doubling, so there
is no reason why it should give rise to fewer artifacts than any other
ratio.

If you want to double, it is 6.02059991dB

d

--
Pearce Consulting
http://www.pearce.uk.com

[Albert]

On Feb 2, 10:58 pm, "David Morgan \(MAMS\)" /Odm
wrote:
"Albert" wrote in messagenews: If I feel the need to do that I usually increase the gain by 6 dB or a
multiple thereof.

You've lost me... what do sixes, twelves, eighteens and twenty-fours have to
do with anything?

Are you saying that increasing volume by 200%, 398.11% or 794.33% (etc.) is
somehow beneficial to a basic normalisation?

Say if the highest peak in the file is - 17dB, I change gain by +12dB.

# 1.... why don't you just use a fader at mixtime?

# 2.... why didn't you just normalize to -5 dbfs

# 3.... what's the mathematical difference between # 2 and your gain change?

More later... ;-)

DM

OK... I actually picked up this tidbit on this group. Here's the
quote, from Scott Dorsey.

"Yes. And for a 6 dB increase all you need to do is a right shift, so
there
is no loss of precision.
But for any other level changes that are NOT powers of two, there will
be
some rounding error introduced with the multiplication, and that is
what folks
are trying to avoid. "
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."

As to your point #1 - one reason might be to have a waveform to look
at where you can see what's happening. Not a big deal, granted, but
sometimes handy for editing. Also, It's something like setting the
faders to unity on a mixer and setting gains to achieve a good basic
mix. It just feels odd to me to have some faders cranked all the way
up and others down at -30.

Albert

[David Morgan \(MAMS\)]

"Albert" wrote in message ups.com...
On Feb 2, 10:58 pm, "David Morgan \(MAMS\)" /Odm
wrote:
"Albert" wrote in messagenews: If I feel the need to do that I usually increase the gain by 6 dB or a
multiple thereof.

You've lost me... what do sixes, twelves, eighteens and twenty-fours have to
do with anything?

Are you saying that increasing volume by 200%, 398.11% or 794.33% (etc.) is
somehow beneficial to a basic normalisation?

Say if the highest peak in the file is - 17dB, I change gain by +12dB.

# 1.... why don't you just use a fader at mixtime?

# 2.... why didn't you just normalize to -5 dbfs

# 3.... what's the mathematical difference between # 2 and your gain change?

More later... ;-)

DM

OK... I actually picked up this tidbit on this group. Here's the
quote, from Scott Dorsey.

"Yes. And for a 6 dB increase all you need to do is a right shift, so
there
is no loss of precision.
But for any other level changes that are NOT powers of two, there will
be
some rounding error introduced with the multiplication, and that is
what folks
are trying to avoid. "
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."

As to your point #1 - one reason might be to have a waveform to look
at where you can see what's happening. Not a big deal, granted, but
sometimes handy for editing. Also, It's something like setting the
faders to unity on a mixer and setting gains to achieve a good basic
mix. It just feels odd to me to have some faders cranked all the way
up and others down at -30.

Albert


Comments Scott? I'm always in learning mode...


--
David Morgan (MAMS)
http://www.m-a-m-s.com
Morgan Audio Media Service
Dallas, Texas (214) 662-9901
_______________________________________
http://www.artisan-recordingstudio.com

[Meindert Sprang]

"David Morgan (MAMS)" /Odm wrote in message
news:fGCxh.1961$5U4.753@trnddc07...

"Albert" wrote in message
ups.com...
On Feb 2, 10:58 pm, "David Morgan \(MAMS\)" /Odm
"Yes. And for a 6 dB increase all you need to do is a right shift, so
there
is no loss of precision.
But for any other level changes that are NOT powers of two, there will
be
some rounding error introduced with the multiplication, and that is
what folks
are trying to avoid. "
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."

As to your point #1 - one reason might be to have a waveform to look
at where you can see what's happening. Not a big deal, granted, but
sometimes handy for editing. Also, It's something like setting the
faders to unity on a mixer and setting gains to achieve a good basic
mix. It just feels odd to me to have some faders cranked all the way
up and others down at -30.

Albert


Comments Scott? I'm always in learning mode...

Scott is right and wrong (:^). Increasing by 6 dB doubles the signal, which
is the same as shifting each sample one bit to the left, which can be
considered as a clean operation. However, shifting one bit left is not
exactly 6 dB, it is 6.02059...... dB. Your DAW software does not even allow
you to set exactly this value. So when you set the gain to 6 dB, the
software has to use an ordinary multiplication anyway, so the gain increase
could be set to any value with the same effect. Further: most if not all
software uses at least 32 bits of even floating point to perform the
calculations and any rounding error in that process is so insignificant
compared to the 24 bit output from your DAW software.

So as long as the software does not offer an exact "sample shift function",
this whole issue is a moot point.

Meindert

[Carey Carlan]

"Meindert Sprang" wrote in
:

Scott is right and wrong (:^). Increasing by 6 dB doubles the signal,
which is the same as shifting each sample one bit to the left, which
can be considered as a clean operation. However, shifting one bit left
is not exactly 6 dB, it is 6.02059...... dB. Your DAW software does
not even allow you to set exactly this value. So when you set the gain
to 6 dB, the software has to use an ordinary multiplication anyway, so
the gain increase could be set to any value with the same effect.
Further: most if not all software uses at least 32 bits of even
floating point to perform the calculations and any rounding error in
that process is so insignificant compared to the 24 bit output from
your DAW software.

So as long as the software does not offer an exact "sample shift
function", this whole issue is a moot point.

I don't know about your DAW software, but Audition supports volume changes
by percentage (as well as dB), allowing you amplify to exactly 200%. That
truly is a bit shift.

[Meindert Sprang]

"Carey Carlan" wrote in message
...
I don't know about your DAW software, but Audition supports volume changes
by percentage (as well as dB), allowing you amplify to exactly 200%. That
truly is a bit shift.

Whether that is going to be a bit shift or not depends on the optimisations
the compiler performed or whether the author(s) of the software detects this
setting and uses shifting instead of multiplying. So you can never be 100%
sure. You could test it by timing a 199% volume change against a 200% volume
change.

It would also be interesting to see if you can compare a file that is
multiplied using 300% (no shifting) at once against a file that is
multiplied in two steps, one of 150% (no shifting) and 200% (possibly
shifting). There is a good chance that the internal precision is so much
higher that the end result truncated to 24 bits almost equal. When dithering
is used, they will probably be different anyway, which brings us back to the
basic question.

Meindert

[Deputy Dumbya Dawg]

Compare = invert one and sum them so at the null you
either hear nothing ( no change - no difference because
the exact opposites canceled on another out) or the
difference as audible as whatever is left. You tell us.


peace
dawg.


"Meindert Sprang" wrote
in message ...
: "Carey Carlan" wrote in message
: ...
: I don't know about your DAW software, but Audition
supports volume changes
: by percentage (as well as dB), allowing you amplify
to exactly 200%. That
: truly is a bit shift.
:
: Whether that is going to be a bit shift or not
depends on the optimisations
: the compiler performed or whether the author(s) of
the software detects this
: setting and uses shifting instead of multiplying. So
you can never be 100%
: sure. You could test it by timing a 199% volume
change against a 200% volume
: change.
:
: It would also be interesting to see if you can
compare a file that is
: multiplied using 300% (no shifting) at once against a
file that is
: multiplied in two steps, one of 150% (no shifting)
and 200% (possibly
: shifting). There is a good chance that the internal
precision is so much
: higher that the end result truncated to 24 bits
almost equal. When dithering
: is used, they will probably be different anyway,
which brings us back to the
: basic question.
:
: Meindert
:
:

[Scott Dorsey]

David Morgan \(MAMS\) /Odm wrote:
As to your point #1 - one reason might be to have a waveform to look
at where you can see what's happening. Not a big deal, granted, but
sometimes handy for editing. Also, It's something like setting the
faders to unity on a mixer and setting gains to achieve a good basic
mix. It just feels odd to me to have some faders cranked all the way
up and others down at -30.

Comments Scott? I'm always in learning mode...

My comments:

1. A lot of software doesn't do straight left and right shift, even
when it could. So in that case, it doesn't matter.

2. My philosophy is that in general, gain changes are probably a bad idea
if you can avoid them, so make as few as possible. Plan things out,
then do it once.

3. Most workstations now use internal floating point representations anyway,
so your data is already being converted to float and back. Once you are
in the floating point representation, you can do plenty of gain changes
without worry.
--scott

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

[Arny Krueger]

"Carey Carlan" wrote in message

Bob Olhsson wrote in
news:GPHwh.15011$fC2.6662 @bignews4.bellsouth.net:

The only time "normalization" can possibly buy you
anything is in the final D to A conversion. Otherwise it
only adds noise or distortion if you fail dither your
gain change.

Yes, you must dither after the amplification, but
normalizing in 24 bits before resampling to 16 bits with
dither is harmless.

I normalize every recording I make. As the last step
while still in 24 bits, I amplify such that the loudest
sample is -1 dBFS. I then resample to 16 bits and
dither, then go to CD.

That's only common sense. While *any* change hurts the measured dynamic
range of a file, It makes sense to quantify the size of the effect on sound
quality. If you are working with 24 bit files or better yet 32 bit point
floating files, it takes an immense number of changes to audibly affect the
dynamic range of the recording.

If you're working with 16 bits, any step that attenuates the file will cause
a perfect file to lose some dynamic range. For example, attenuating a
perfect file by 6 dB will cost about 6 dB of dynamic range. However, real
world audio files are far from perfect. Attenuating a q6 bit file with 75
dB dynamic range will have negligable effects on the dynamic range of the
file until the file's peak levels are no higher than about -20 dB FS. Then,
further attenuation will reduce the dynamic range by roughly the amount of
the attenuation.

Amplifiying a 16 bit file has negligable effects on the dynamic range of a
file as long as clipping is avoided.

These are situations that can be tested with DAW software pretty readily,
using RMA55 freeware as your analytical software.

[Arny Krueger]

"Albert" wrote in message
oups.com
On Jan 10, 8:51 pm, jtougas
wrote:
Hi folks -

In my pursuit of improving my recordings, I've taken a
few of my mixes to a friend of mine, to see what he
thinks, and I mentioned I'd normalized one or two of the
channels in the tracks since they were a little quieter
than the others.

If I feel the need to do that I usually increase the gain
by 6 dB or a multiple thereof. Say if the highest peak in
the file is - 17dB, I change gain by +12dB. The thought
there is that a 6 dB change is a simple doubling, whereas
any other amount (like an amount that might be applied by
normalising) is more complex, giving a bigger chance of
unwanted artifacts.

No doubt based on the mistaken idea that 6 dB represents doubling or
halving, and that this will somehow net out to arithmetic that does not
generate new fractional parts. There's at least one fly in that ointment -
doubling or halving corresponds to an uneven number of dB, namely 6.02 dB.

[Romeo Rondeau]

Comments Scott? I'm always in learning mode...

You'll never hear the difference anyway, so who cares?

[Carey Carlan]

"Meindert Sprang" wrote in
:

"Carey Carlan" wrote in message
...
I don't know about your DAW software, but Audition supports volume
changes by percentage (as well as dB), allowing you amplify to
exactly 200%. That truly is a bit shift.

Whether that is going to be a bit shift or not depends on the
optimisations the compiler performed or whether the author(s) of the
software detects this setting and uses shifting instead of
multiplying. So you can never be 100% sure. You could test it by
timing a 199% volume change against a 200% volume change.

It would also be interesting to see if you can compare a file that is
multiplied using 300% (no shifting) at once against a file that is
multiplied in two steps, one of 150% (no shifting) and 200% (possibly
shifting). There is a good chance that the internal precision is so
much higher that the end result truncated to 24 bits almost equal.
When dithering is used, they will probably be different anyway, which
brings us back to the basic question.

It appears that Audition never bit shifts.

Test 1:
I recorded a 1 kHz tone for 10 seconds at -10 dBFS at 16 bits.
I then amplified it to 200%.

The original waveform stats revealed max/min samples of +/- 10363.
Double that is 20786.
The doubled signal was +/- 20727.

Test 2:
Same 1 kHz source.
Amplify left channel to 300%. Min/Max are -31090 + 31090
Amplify right channel to 200%, then 150%. Order is important. If I had
used 150% first, all odd values would have to be rounded. After the 200%
amplification there should be no odd values.
Min/Max values -31090 + 31090 (3 x 10363 = 31089)
Invert right and add to left.
Min/Max samples -32 + 31. About 64 down at 1 kHz. Definitely not
perfectly canceled.

This test impacted the bottom 5 bits rather than 3.

[ernest]

And raising the level with normalization and then
cutting it back with the mix faders is counterproductive.

I just did a mix and it sounds ok mixed, but the overall levels are
just too low. If I mixed down stereo, I cannot get too much by
putting on comp/limit and pushing it up - coz it'll sound crushed.
I've found 'a way' - but you guys probably kick me for this: I mixed
down each sub mix (guitar,vocal,drums...), and then re-import them
into the sequencer, and then there I normalize all of them and the
push it back to get the balance mix back. I know it's bad/wrong/evil,
but what is the alternative? I'm not sure what is 'post-gain' in
cubase term is.

I have a bunch of related questions. What is 'applying gain', or 'post
gain' in Cubase term? I'm always confused with 2 things in Cubase:

1. The channel fader - it has markings from -inf to +6. What does it
do when I push it above the fader 0? Is that the 'post gain'?
2. Each saved .wav segment has a 3 blue color points, front/back and
middle. The front/back is for fading and the middle one, I can drag it
up or down - for -inf to +6 again, while the waveform changes. Is that
another 'post gain'?

- what are the difference between the two in terms of the gain it
created?
- given digital nature of the signal, why can't one push it much
higher than +6 based on user request?
- what are other ways of applying user specified 'post gain' in the
signal chain? Set the comp to 1:1 and push the makeup???

Thanks...


Or you are implying people don't
wanna turn up their volume these days, and/or people are
recording/mixing way too loud? Or whoever said -15 to -18
dBFS is sufficient is simply wrong?

-15 to -18 dBFS nominal level is good. It leaves 15 to 18 dB of
headroom for peaks that naturally occur in music.

It's true that people don't want to turn their volume control up. And
with their engineer hat on, they feel that if they have to turn up the
listening volume, it will be too low to compete with commercial
recordings. And they'd be right. But the place to make that adjustment
is after the mix is completed, not by rasiing the level of every track.

[Arny Krueger]

"Carey Carlan" wrote in message

"Meindert Sprang" wrote in
:

"Carey Carlan" wrote in message
...
I don't know about your DAW software, but Audition
supports volume changes by percentage (as well as dB),
allowing you amplify to exactly 200%. That truly is a
bit shift.

Whether that is going to be a bit shift or not depends
on the optimisations the compiler performed or whether
the author(s) of the software detects this setting and
uses shifting instead of multiplying. So you can never
be 100% sure. You could test it by timing a 199% volume
change against a 200% volume change.

It would also be interesting to see if you can compare
a file that is multiplied using 300% (no shifting) at
once against a file that is multiplied in two steps, one
of 150% (no shifting) and 200% (possibly shifting).
There is a good chance that the internal precision is so
much higher that the end result truncated to 24 bits
almost equal. When dithering is used, they will probably
be different anyway, which brings us back to the basic
question.

It appears that Audition never bit shifts.

Test 1:
I recorded a 1 kHz tone for 10 seconds at -10 dBFS at 16
bits.
I then amplified it to 200%.

The original waveform stats revealed max/min samples of
+/- 10363. Double that is 20786.
The doubled signal was +/- 20727.

Test 2:
Same 1 kHz source.
Amplify left channel to 300%. Min/Max are -31090 + 31090
Amplify right channel to 200%, then 150%. Order is
important. If I had used 150% first, all odd values
would have to be rounded. After the 200% amplification
there should be no odd values.
Min/Max values -31090 + 31090 (3 x 10363 = 31089)
Invert right and add to left.
Min/Max samples -32 + 31. About 64 down at 1 kHz.
Definitely not perfectly canceled.

This test impacted the bottom 5 bits rather than 3.

Not sure about that. It looks to me like the gains came out different by an
amount that needs 5 bits to be expressed, but the results can easily be
spread oven some or all of the 16 available bits.

If you want to concentrate your analysis on the impact on the low order
bits, you need to actually measure dynamic range, or something like it.

[Carey Carlan]

"Arny Krueger" wrote in
:

Invert right and add to left.
Min/Max samples -32 + 31. About 64 down at 1 kHz.
Definitely not perfectly canceled.

This test impacted the bottom 5 bits rather than 3.

Not sure about that. It looks to me like the gains came out different
by an amount that needs 5 bits to be expressed, but the results can
easily be spread oven some or all of the 16 available bits.

If you want to concentrate your analysis on the impact on the low
order bits, you need to actually measure dynamic range, or something
like it.

The left/right difference files have a maximum value of 32. That's 5 bits.
That's not saying that the signal only changes when it's in the bottom 30
dB. Rather, it's saying that a -64 dB (or quieter) component has been
added to the whole signal.

[Arny Krueger]

"Carey Carlan" wrote in message

"Arny Krueger" wrote in
:

Invert right and add to left.
Min/Max samples -32 + 31. About 64 down at 1 kHz.
Definitely not perfectly canceled.

This test impacted the bottom 5 bits rather than 3.

Not sure about that. It looks to me like the gains came
out different by an amount that needs 5 bits to be
expressed, but the results can easily be spread oven
some or all of the 16 available bits.

If you want to concentrate your analysis on the impact
on the low order bits, you need to actually measure
dynamic range, or something like it.

The left/right difference files have a maximum value of
32. That's 5 bits. That's not saying that the signal
only changes when it's in the bottom 30 dB. Rather, it's
saying that a -64 dB (or quieter) component has been
added to the whole signal.

The way I look at it, if you're off by +/-32 in a system where values can
range over roughly +/- 32,000, you're off by 1 part in 1,000 or +/- 0.01 dB.
That will never be heard, and you'd need about 32 such differences to have
any chance of an audible effect.

[Carey Carlan]

"Arny Krueger" wrote in
:

The left/right difference files have a maximum value of
32. That's 5 bits. That's not saying that the signal
only changes when it's in the bottom 30 dB. Rather, it's
saying that a -64 dB (or quieter) component has been
added to the whole signal.

The way I look at it, if you're off by +/-32 in a system where
values can range over roughly +/- 32,000, you're off by 1 part in
1,000 or +/- 0.01 dB. That will never be heard, and you'd need about
32 such differences to have any chance of an audible effect.

dB is a log scale. Adding a -64 dBFS signal to full scale is minute. You
personally have recorded concerts with peaks more than 60 dB above the
noise floor. In those quiet passages, a -64 dB signal is loud indeed.
That's why we use 24 bits -- to bury that "32 dB above zero" value deep
into the noise floor.

[Arny Krueger]

"Carey Carlan" wrote in message

"Arny Krueger" wrote in
:

The left/right difference files have a maximum value of
32. That's 5 bits. That's not saying that the signal
only changes when it's in the bottom 30 dB. Rather,
it's saying that a -64 dB (or quieter) component has
been added to the whole signal.

The way I look at it, if you're off by +/-32 in a
system where values can range over roughly +/- 32,000,
you're off by 1 part in 1,000 or +/- 0.01 dB. That will
never be heard, and you'd need about 32 such differences
to have any chance of an audible effect.

dB is a log scale. Adding a -64 dBFS signal to full
scale is minute. You personally have recorded concerts
with peaks more than 60 dB above the noise floor. In
those quiet passages, a -64 dB signal is loud indeed.
That's why we use 24 bits -- to bury that "32 dB above
zero" value deep into the noise floor.

More to the point - many including myself do their mixing and other
processing with 24 or 32 bit floating arithmetic, and leave the 16 bits for
distribution only.

[Peter Larsen]

Laurence Payne wrote:

A lot of people who build up music by multitrack recording routinely
normalise every recorded file, before starting to mix.

Depending on just what they do in the mix it could have some sense, so
as to be nicely in the middle of the 32 bits of usable word length and
avoid truncation, but it is not possible for me to defeat Bobs general
"only one step of processing" viewpoint.


Regards

Peter Larsen

[Arny Krueger]

"Laurence Payne" lpayne1NOSPAM@dslDOTpipexDOTcom wrote in
message
On 2 Feb 2007 07:54:53 -0800, "Mike Rivers"
wrote:

To most people who have a desire to normalize, they're
talking about the final mix. At least until they change
something else. g

A lot of people who build up music by multitrack
recording routinely normalise every recorded file, before
starting to mix.

This can facilitate editing, both visually and audibly, especially when the
tracks were recorded with lots of headroom.