compressor, normalize, clipping? for the "not-so-tech-savy"!

Post your questions & get help from friendly LibriVoxers
schrm
Posts: 698
Joined: February 10th, 2018, 11:02 am
Location: Austria

Post by schrm » May 21st, 2018, 1:46 pm

hi all,

as i understand it, normalizing or using the compressor is not recommended, because clipping happens.
so, what can i do with a file, which has really high volume-peaks?
you can see it here: hundred percent peaks!
Image

i have to confess it:
i tried to normalize it and the compressor, both show similar results..
and my speakers thanked me.
so, what is clipping? does that mean, it really cuts the peaks, instead of amplifying most peaks?
is there a difference between compressor and normalize?
which is worse to use, which is just not that good?
or did i misunderstand some warning about these two?

sorry to say it: i dont understand or know much in that area...
but maybe, someone more tech savy can teach me about another tool or these both?
on a sort of low level for the not so tech savy?
:-)
Last edited by schrm on June 6th, 2018, 1:17 am, edited 1 time in total.
schrm
reader/12275


deutsch
wunderwelten viewtopic.php?f=60&t=70793
sammlung kurzer prosa viewtopic.php?f=60&t=71431

english
violet viewtopic.php?f=2&t=70567

moniaqua
Posts: 1092
Joined: April 11th, 2013, 4:48 am
Location: Somewhere in the south

Post by moniaqua » May 21st, 2018, 2:18 pm

Well, I am not that much in audio technics, too, but I know a bit :)

To normalize is kind of like to amplify, only that you amplify to a certain level (a norm). You can do this two ways:

a) based on peaks, so that the loudest peak reaches the level of norm.
b) based on general loudness, so that loudness reaches a certain level.

With a), if you have high peaks, you might not be able to reach the desired volume for LibriVox (I am not quite sure how this one is measured, but it must be some kind of average of the wave form).

With b), clipping can occur if you have high peaks.

Clipping is cutting of on high peaks, so that the soundwave looks - cut :) Look here, wonderful music but sadly clipping badly: https://archive.org/details/MastiamDardoMano_201701

If you have peaks that prevent you from amplifying to the 89 dB, you can try compressor. This one, as the name says, compresses the dynamic range. That's to say, high peaks become softer, low regions louder. This way it should be easier to get higher volume without clipping, but of course you can overdo it resulting in a not nice to listen to voice.

Hier ist ein ganz guter Artikel, nicht erschrecken, er ist sehr umfangreich, aber er hat Hörbeispiele für die einzelnen Einstellungen dabei:
https://www.bonedo.de/artikel/einzelansicht/die-10-groessten-fehler-bei-der-kompressor-einstellung.html

schrm
Posts: 698
Joined: February 10th, 2018, 11:02 am
Location: Austria

Post by schrm » May 21st, 2018, 2:31 pm

should've asked you first :lol:
thank you!
moniaqua wrote:
May 21st, 2018, 2:18 pm
Well, I am not that much in audio technics, too, but I know a bit :)

To normalize is kind of like to amplify, only that you amplify to a certain level (a norm). You can do this two ways:

a) based on peaks, so that the loudest peak reaches the level of norm.
b) based on general loudness, so that loudness reaches a certain level.

With a), if you have high peaks, you might not be able to reach the desired volume for LibriVox (I am not quite sure how this one is measured, but it must be some kind of average of the wave form).

With b), clipping can occur if you have high peaks.

Clipping is cutting of on high peaks, so that the soundwave looks - cut :) Look here, wonderful music but sadly clipping badly: https://archive.org/details/MastiamDardoMano_201701

If you have peaks that prevent you from amplifying to the 89 dB, you can try compressor. This one, as the name says, compresses the dynamic range. That's to say, high peaks become softer, low regions louder. This way it should be easier to get higher volume without clipping, but of course you can overdo it resulting in a not nice to listen to voice.

Hier ist ein ganz guter Artikel, nicht erschrecken, er ist sehr umfangreich, aber er hat Hörbeispiele für die einzelnen Einstellungen dabei:
https://www.bonedo.de/artikel/einzelansicht/die-10-groessten-fehler-bei-der-kompressor-einstellung.html
schrm
reader/12275


deutsch
wunderwelten viewtopic.php?f=60&t=70793
sammlung kurzer prosa viewtopic.php?f=60&t=71431

english
violet viewtopic.php?f=2&t=70567

Peter Why
Posts: 4253
Joined: November 24th, 2005, 3:54 am
Location: Chigwell (North-East London, U.K.)

Post by Peter Why » May 21st, 2018, 10:37 pm

I get a lot of variation in amplitude when I record, too. I've worked out compressor and normalise/deamplify settings which help to clear it, though. The compressor option "compress based on peaks" does the work: compressor threshold -14, noise floor -40, ratio 5:1, attack time 0.2, release time 1, and both check boxes ticked. THEN normalise or deamplify by 5dB. I'll then export the file as an mp3 for Checker and make any needed volume adjustment (never more than 2dB is needed).

You can noise clean before or after doing this.

Try it for yourself and see if it works on your files.

Peter
"Anyone who believes that exponential growth can go on forever in a finite world is either a madman or an economist." Kenneth Boulding, 1973

schrm
Posts: 698
Joined: February 10th, 2018, 11:02 am
Location: Austria

Post by schrm » May 22nd, 2018, 1:53 am

thank you peter!

now, you gave me some basic settings, which i can relate to..
lets see..

thank you very much!
Peter Why wrote:
May 21st, 2018, 10:37 pm
I get a lot of variation in amplitude when I record, too. I've worked out compressor and normalise/deamplify settings which help to clear it, though. The compressor option "compress based on peaks" does the work: compressor threshold -14, noise floor -40, ratio 5:1, attack time 0.2, release time 1, and both check boxes ticked. THEN normalise or deamplify by 5dB. I'll then export the file as an mp3 for Checker and make any needed volume adjustment (never more than 2dB is needed).

You can noise clean before or after doing this.

Try it for yourself and see if it works on your files.

Peter
schrm
reader/12275


deutsch
wunderwelten viewtopic.php?f=60&t=70793
sammlung kurzer prosa viewtopic.php?f=60&t=71431

english
violet viewtopic.php?f=2&t=70567

Penumbra
Posts: 942
Joined: May 10th, 2016, 6:16 pm

Post by Penumbra » May 22nd, 2018, 1:41 pm

This thread has prompted me to make a few comments about Audacity effects. Here's hoping :) someone finds them helpful:

1) To the best of my knowledge, the standard audacity normalize and compressor effects cannot cause clipping. Other compressors and normalizers, including those that can be added to Audacity, might very well do so.

2) The audacity amplify effect can result in clipping if you check the box that says allow clipping. Without that box checked, no clipping will result.

3) If your system produces a significant DC offset, then it is a good idea to remove it early in your workflow by using the audacity normalize effect with the "remove DC offset" box checked. Leaving the DC offset in makes editing a pain and can make it difficult to amplify to 89 db.

4) The amplify and normalize effects differ in how they treat multiple tracks. But when you are working with just a single track, there is no difference in how they amplify.

5) Noise reduction will decrease the computed loudness of the track, although usually not by very much.
Tom Penn
If you are looking for an easy, fun read, LV has only 5 of Gutenberg's 40+ books for children and YA by Roy J Snell.

schrm
Posts: 698
Joined: February 10th, 2018, 11:02 am
Location: Austria

Post by schrm » May 23rd, 2018, 2:24 am

Penumbra wrote:
May 22nd, 2018, 1:41 pm
This thread has prompted me to make a few comments about Audacity effects. Here's hoping :) someone finds them helpful:

1) To the best of my knowledge, the standard audacity normalize and compressor effects cannot cause clipping. Other compressors and normalizers, including those that can be added to Audacity, might very well do so.

2) The audacity amplify effect can result in clipping if you check the box that says allow clipping. Without that box checked, no clipping will result.

3) If your system produces a significant DC offset, then it is a good idea to remove it early in your workflow by using the audacity normalize effect with the "remove DC offset" box checked. Leaving the DC offset in makes editing a pain and can make it difficult to amplify to 89 db.

4) The amplify and normalize effects differ in how they treat multiple tracks. But when you are working with just a single track, there is no difference in how they amplify.

5) Noise reduction will decrease the computed loudness of the track, although usually not by very much.
hi tom,

i saw some youtube videos, which more or less confused me - and now, you are the first on lv, hinting at these aspects.
so, we can deal with these tools more relaxed, judging the work by the results?

also, can you give us a value for the dc-offset?
is it more around 1, around 5, or around 20?

thank you for this statement!

cheers,
schrm
schrm
reader/12275


deutsch
wunderwelten viewtopic.php?f=60&t=70793
sammlung kurzer prosa viewtopic.php?f=60&t=71431

english
violet viewtopic.php?f=2&t=70567

annise
LibriVox Admin Team
Posts: 29891
Joined: April 3rd, 2008, 3:55 am
Location: Melbourne,Australia

Post by annise » May 23rd, 2018, 2:58 am

There's not set value for DC offset if it is present, I've seen quite large values, it depends on recording set ups. I've certainly seen values of 10 plus , though I usually just get rid of it, not measure it.
The main problem with it is when editing, if the offset is say 10, the recording wave plunges straight down from 10 to 0 and this drop is audible.

Anne

schrm
Posts: 698
Joined: February 10th, 2018, 11:02 am
Location: Austria

Post by schrm » May 23rd, 2018, 4:50 am

annise wrote:
May 23rd, 2018, 2:58 am
There's not set value for DC offset if it is present, I've seen quite large values, it depends on recording set ups. I've certainly seen values of 10 plus , though I usually just get rid of it, not measure it.
The main problem with it is when editing, if the offset is say 10, the recording wave plunges straight down from 10 to 0 and this drop is audible.

Anne
thank you anne!
i think, this happened to me already, even with an offset of just 0,5-2
i will take care of it in future!

cheers,
schrm
schrm
reader/12275


deutsch
wunderwelten viewtopic.php?f=60&t=70793
sammlung kurzer prosa viewtopic.php?f=60&t=71431

english
violet viewtopic.php?f=2&t=70567

iBeScotty
Posts: 315
Joined: December 3rd, 2016, 2:19 pm
Location: California

Post by iBeScotty » May 23rd, 2018, 8:14 am

For sure, recording here can be tricky because we are both the talent and the engineer, though obviously just getting a sound is not too hard. In my humble opinion, learning a little bit the fundamentals of sound and general concepts of manipulation greatly helps in the use of the tools in any software, be it Audacity or anything else, which can maximize the recording quality.

John Hess at Filmmaker IQ does such a good job explaining things and gives some good overviews on sound, post processing/mastering tools and recording, though geared for recording dialogue for video, much of it applicable to recording speech in general

Here is his post processing overview talking about equalization, compression, de-noise, etc.: https://youtu.be/r4791OLkRrs

Generally I use some mild equalization first, to compensate for room and equipment coloring of sound. As part of this, rolling off low frequencies below 100Hz seems to take care of any dc offset issue as well. John says he boosts 160Hz but I find that my small room already accentuates this frequency and have to notch between around 160-200Hz to get rid of some “mud.”

As far as compression, i usually use a little mild, straight compression without any auto make-up gain and then amplify in the expander to keep the noise floor down. John likes the multiband compressor (essentially combing comp and eq) but even after following a tutororial with recommending settings for voice I could not get the sound how I liked it as well as with eq and straight compression (so this may require more experience).

Here is his overview on the science of sound which may also be of help: https://youtu.be/ZbUTyMC8_X8

And finally, recording sound, though geared for film, good tips in recording levels and mic placement, etc.: https://youtu.be/S9cP1WHL0Zo

Hope this helps. Bottom line is play around, experiment, have fun and make something great!
Scotty

ProDigit
Posts: 10
Joined: May 12th, 2018, 5:57 pm

Post by ProDigit » June 5th, 2018, 8:46 pm

Clipping is cutting off peaks of data. In most softer cases, clipping won't do much harm, as any pc nowadays can clip up to 3dB without really losing quality.
They have something called 'soft clipping', where instead of the natural curve of the wave, the tip gets rounded off more VS totally cutting off the tip horizontally like with regular hard clipping.
Hard clipping is audible. Soft clipping isn't very audible.

Nevertheless, clipping is unwanted.

Normalize:
To maximize your audio volume without clipping the peaks.

Compression:
To increase low volume signals more than high volume signals.
For instance,
If you speak at 65dB volume, and your 'p' and 't' and 's' sounds (pop and sizzle) is at 80dB, using a compressor, you can increase the 65dB sound levels by 15dB to 80dB, while the peaks of 80dB might see only a 5dB increase to 85dB.
You will reduce dynamics, and get more consistent volume levels.
It's good to get a great signal to noise ratio.
If applied too much of compression, background noise will be amplified much more. Which is why before applying compression, you always should try to apply a noise gate (remove noise levels of eg: 25dB or less), and a mild noise filter (remove or reduce some of the recurring background noise).

Preferably, you'll want to generate a HOT signal (one where the signal peaks reach 0dB, meaning, right before they start clipping).
If your clip light flicks on occasionally, there's no harm done. If it flicks on regularly, best is to decrease the gain a bit.

schrm
Posts: 698
Joined: February 10th, 2018, 11:02 am
Location: Austria

Post by schrm » June 6th, 2018, 1:16 am

and here we have another perfect answer - thank you very much!
i will change the thread title: for others, not so tech savy, this thread may be helpful.
schrm
reader/12275


deutsch
wunderwelten viewtopic.php?f=60&t=70793
sammlung kurzer prosa viewtopic.php?f=60&t=71431

english
violet viewtopic.php?f=2&t=70567

lurcherlover
Posts: 555
Joined: November 10th, 2016, 3:54 am
Location: LONDON UK

Post by lurcherlover » June 11th, 2018, 1:00 pm

Recording narration is rather different to recording music where the dynamic range can be as much as 30dB plus. Narration is usually about 5-10dB dynamic range. It is to do with mic technique, in other words, not letting the voice produce extremes of loudness and quietness. If you need to shout, move well back from the mic and shout at 90 degrees. If you need to whisper, make it a loud whisper. This way you keep the dynamic range small and tight.

There is then no need usually to use compression and other ways, such as a limiter, to keep the waveform under control. It sounds much better and is less complicated.

schrm
Posts: 698
Joined: February 10th, 2018, 11:02 am
Location: Austria

Post by schrm » June 11th, 2018, 2:44 pm

lurcherlover wrote:
June 11th, 2018, 1:00 pm
It is to do with mic technique, in other words, not letting the voice produce extremes of loudness and quietness. If you need to shout, move well back from the mic and shout at 90 degrees. If you need to whisper, make it a loud whisper. This way you keep the dynamic range small and tight.
thank you for this advice!
i would have needed that in my solo the last two contributions - and will try it for my next recording :-)
schrm
reader/12275


deutsch
wunderwelten viewtopic.php?f=60&t=70793
sammlung kurzer prosa viewtopic.php?f=60&t=71431

english
violet viewtopic.php?f=2&t=70567

moniaqua
Posts: 1092
Joined: April 11th, 2013, 4:48 am
Location: Somewhere in the south

Post by moniaqua » June 11th, 2018, 11:13 pm

lurcherlover wrote:
June 11th, 2018, 1:00 pm
If you need to whisper, make it a loud whisper.
What do you mean with "loud whisper"?

Post Reply