effects - normalize vs amplify

[steve]

but then we would get loads of posts saying "help, I've used 300 GB of disk space and I've only recorded 3 albums".

The warnings about excessive disk space usage would have to be in massively big letters.

[waxcylinder]

I have just recently transferred a feature request from the forum to thre Wiki requesting retention of the Undo trail - switchable on/off - please go to the Wiki and vote if you feel strongly.

WC

[kozikowski]

<<<"help, I've used 300 GB of disk space and I've only recorded 3 albums".>>>

As opposed to the people who save their multi-thousand dollar project and go home only to come back the next day and discover they have no show because the UNDO vanished. I take option A. Audacity is way too talented at destroying your show.

Remember WordStar? I don't either. It was a very early word processor that even in the face of the early ratty machines would very rarely flush your work. It had a design philosophy that under no circumstances was any original work to be damaged or deleted. Ever.

Koz

[kozikowski]

<<<please go to the Wiki and vote if you feel strongly.>>>

...by clicking on this link...

Koz

[waxcylinder]

ok, ok - the link for the Wiki FR page is: http://audacityteam.org/wiki/index.php? ... e_Requests

And you will find it in: Improved Resource Control > Undo Buffer Management > Save Undo History into the project file

WC

[jan.kolar]

They're not quite the same thing, but the difference is easy to miss. Steve is on the money except for one thing:
They behave differently when you apply the effect to multiple tracks.
...

If you apply Normalize to more than one track, each track will be amplified separately in order to get that track to the specified peak value.

Good point indeed.

Now, if the track contains multiple clips, they are normalized together. (Mulitplied by the same amount.)
Now I zeroed part of track (Ctrl+L, 1.3.5.) and normalized whole of it to see zero changed to some almost zero. Whats that?

Uf, just wanted to ask if the two channels of stereo track are normalized together or separately,
and thinking turned out to be dangerous.
Good night.

[steve]

Now, if the track contains multiple clips, they are normalized together. (Mulitplied by the same amount.)

The Normalize effect is applied to the track. Audacity will calculate how much to amplify the track by to achieve the target peak level. (it would be a problem in many situations if it did not work like this).

Now I zeroed part of track (Ctrl+L, 1.3.5.) and normalized whole of it to see zero changed to some almost zero. Whats that?

If you are not working with a 32 bit track format, the Normalize effect will apply dither to the the output. Didn't you recently have a conversation on the forum about how Audacity applies dither to silence?

I think it would be better if Audacity did not apply dither to long periods of silence (even though it is inaudible). It makes sense for Normalize to apply dither, since there may be slight (very slight) clicking as an audio signal fades out due to quantising errors, but I think after a few samples of silence, dither should not be used.

Take this example:
Generate silence (16 bit)
Normalize
Normalize (again)
Logically you would expect silence, but what you actually get is full scale noise. Of course this is an artificial test, (why would you want to repeatedly normalise a silent track?) but I think it illustrates a flaw in apply dither without considering the context.

Uf, just wanted to ask if the two channels of stereo track are normalized together or separately,

As you know, Audacity treats stereo tracks as 2 mono tracks (side by side). Each (mono) channel is normalised independently, just like any other mono track. This can be quite useful if you have a recording that has the pan out of balance, providing a simple way to correct it.

[jan.kolar]

Now I zeroed part of track (Ctrl+L, 1.3.5.) and normalized whole of it to see zero changed to some almost zero. Whats that?

If you are not working with a 32 bit track format, the Normalize effect will apply dither to the the output. Didn't you recently have a conversation on the forum about how Audacity applies dither to silence?

1. The point is that it did not look like the ususal dither.
1a. There were spikes of the same height. In dither noise in exprort I usually see about three different heights. (Well might be the 'fast' dither now?)
1b. The spikes were only of positive polarity.
OK, I took another look today. (Inded I had 16bit track.) To be continued few lines below...

2. I have bad memory. What I would say I rember is this: Perhaps I had conversation on Forum about dither
and then notified Gale on -devel about my observation on non-dithering silence.

I think it would be better if Audacity did not apply dither to long periods of silence (even though it is inaudible).

Just said it: My observation is that it (1.3.5.) does not dither silence (prehaps even short periods.)

... so we can continue.
A picture (normalized normalized silence inside longer low level track): Please care only about the upper track.
The left part suffered one more normalization. You see there Large dither and Small dither combined.
The small dither is one unit height, sometimes up and sometimes down (nevertheless, always from zero, which I decided to ignore right now.)
The large dither is one Unit height, too (however the Unit is larger that the previous unit, because amplified).
However it does not consist of spikes of both polarities. Instead it is like ---.--.----.--.---..-----
two levels, where the upper is much prevalent.

--

dither_unbalanced.jpg (114.46 KiB) Viewed 3441 times

This might be caused by interferrence between "dither only nonzero" and DC-offset removal
(a bit of silenced-out in a track probably gets DC-biased when the whole track is DC-removed.)
So I feel there is something a bit unheatlhy. I do not say it is important or not, but once dither is considered
important, it should work as expected, not as it happens to happen.

It makes sense for Normalize to apply dither, since there may be slight (very slight) clicking as an audio signal fades out due to quantising errors,

This does not need dither: copy & cut (in the same format), add/mixing (with multiplier 2), special amplifications (integer multiplier), silencing.
This needs dither: Generate sine and others, general amplifications (including 1/2), normalize, almost any effect.
This needs dither but deserves special treatment (at least at 8 or 10 bits): fade in/out.

I did not thing about it deeply, but I think this should be the order : 16source->float->manipulation(in float)->dither(in float)->16bit convert. That is, dither in float before conversion.
[If anybody votes for dither after conversion, I say that might be fast but not clever, it could not shift quantisation noise to other frequencies, but it can only mask it. Then dither would best come later, at the export time.]

but I think after a few samples of silence, dither should not be used.

How many samples uses the 1.3.5 to decide? If too few, gets wrong.

Uf, just wanted to ask if the two channels of stereo track are normalized together or separately,

As you know, Audacity treats stereo tracks as 2 mono tracks (side by side). Each (mono) channel is normalised independently, just like any other mono track. This can be quite useful if you have a recording that has the pan out of balance, providing a simple way to correct it.

I confirm it does what you say. (I would split the tracks to get this. And I would cry for RMS normalization.).
But it can destroy carrefully balanced recording. Orchestra - does not it have drums to one side, for example?
And I would guess live stereo piano recording would seem to be about half a dB unbalanced, at least.
OK, it is easier to write <<for each track: normalize>> than almost four time longer <<for each track: if mono: normalize else: get peaks of both and amplify both and skip normalizing right track again>>.

[steve]

1. The point is that it did not look like the ususal dither.
1a. There were spikes of the same height. In dither noise in exprort I usually see about three different heights. (Well might be the 'fast' dither now?)
1b. The spikes were only of positive polarity.....

Those pictures are indeed strange and quite different from what I get. Without knowing your exact method I am unable to reproduce what you get.

This is what I get. The first track is using shaped dither on silence and normalised twice. The second track is a low amplitude sine wave that is first amplified to a very low level and then normalised (also shaped dither). The third track is a small amplitude sine wave, amplified to a very low level then amplified, but this time using rectangle dither.

Screenshot.png (63.46 KiB) Viewed 3425 times

The introduced noise looks to me very much like dither noise.
It is also interesting to note that using rectangle dither does not introduce noise to silence.

How many samples uses the 1.3.5 to decide?

I've no idea I'll leave that to the experts to work out. I'm just indicating that the current method of shaped dither produces unexpected and undesirable results in certain cases.

And I would cry for RMS normalization

I think it would be quite easy to write a Nyquist plug-in to do this, although using actual RMS may not be exactly what you mean. RMS normalisation gives a better approximation to "Equal loudness" normalisation than peak level normalisation (and I am betting that this is what you really want), but it is not perfect. Equal loudness normalization should follow the response curve of the human ear, which is not easy to do because it varies from one individual to another - however RMS normalisation is probably closer than peak level normalisation.

But it can destroy carrefully balanced recording. Orchestra - does not it have drums to one side, for example?

In this case you would use "Amplify" rather than "Normalize". Since you are talking here about a stereo mix rather than individual instruments on individual tracks, "Amplify" would achieve the desired result whereas "Normalize" would not (If you need to correct DC offset, then this can be done with the "Normalize" effect without changing the levels).

The important point here is to know the difference between what "Normalize" and "Amplify" do in Audacity. Fortunately, having both of these effects at our disposal, just about every case is covered.

[demctaggart]

Back to normalizing.

I use the Anwida DX Reverb Lite plugin.

When I apply the reverb, I lose 6-18db average.

So, I end up having to amplify like crazy to get back to 0db.

I read about normalizing in this thread and decided to try that as well.

I had to normalize at -12db to account for all the loss and to end up with
a normal size waveform/sample.

Here's the problem, all that is very good except for one major issue.

The line that is supposed to be flat in the middle of the track at 0 is huge.
I think the noise floor is getting amplified with everything else and even the
noise removal tool cannot get rid of the loud hiss without ruining the recording.

Question: is there a way to do this without the above mentioned problem?
Or is it just the equipment generating the noise and it is more pronounced
after normalization/amplification?

In other words I want to know how to keep the flat line, even when normalizing/amplifying.