Is normalizing audio a bad thing?

[killerteengohan]

I sometimes come across a DVD with very low volume to where the volume seems a bit low. Sometimes I have to turn my speakers all the way up to hear it as well as I would like and I can hear a constant hissing sound from the speakers being up too high. (It's easy to hear when theres no sound playing but their turned up all the way)

I have just been using Normalize Peaks to 90 (0-100 are the choices) when encoding audio to AAC and that seems to raise the volume enough to be acceptable and I don't need my speakers turned all the way up.

I hear normalizing is or can be damaging though and can cause clipping (Which I never noticed or don't know too much about, just what I read)
I haven't noticed anything bad audibly from doing this, but I also don't want to harm the audio too much.

Is what I've been doing a bad thing? It's been doing what I want and I never heard any audible downfalls to it.

[budwzr]

There's no volume standard for DVD's, so some are too loud and others are too quiet. Just raise the levels instead of normalizing.

P.S. I'm gonna get banged in the back of the head with a shovel here in a minute.

[newpball]

I have just been using Normalize Peaks to 90 (0-100 are the choices) when encoding audio to AAC and that seems to raise the volume enough to be acceptable and I don't need my speakers turned all the way up.

Nothing wrong normalizing your audio but obviously all parts need to be consistent. A whisper scene is obviously softer than an aircraft flying over.

I hear normalizing is or can be damaging though and can cause clipping.

Nonsense, normalizing does not cause clipping.

However if you normalize to 0 db you have no room for anything else on top of that.

Whether a track is perceived loud or not may or may not have anything to do with normalizing. A track with a high dynamic range may actually be perceived as soft even if normalized.

Normalization and compression are two different things.

[turk690]

However if you normalize to 0 db you have no room for anything else on top of that.

The peaks in an audio clip are the highest levels it has. Normalizing those peaks to 0dB (increasing the gain of the whole clip by a factor such that the peaks are 0dB) still make them the highest levels in that clip, so why is there need for room on top of that?

[newpball]

However if you normalize to 0 db you have no room for anything else on top of that.

The peaks in an audio clip are the highest levels it has. Normalizing those peaks to 0dB (increasing the gain of the whole clip by a factor such that the peaks are 0dB) still make them the highest levels in that clip, so why is there need for room on top of that?

True, if you do not add anything, don't equalize, don't apply any filters that would work.

Typically I would recommend -3db as the maximum in the final mix and -6db in any submix.

Basically normalizing does nothing intrinsically to the sound, it just changes the volume level.

[budwzr]

The "low volume" problem is because the DVD audio was Mastered way below "normal". Probably an older generation disc made before "Hot Mastering" became the norm.

[newpball]

The "low volume" problem is because the DVD audio was Mastered way below "normal". Probably an older generation disc made before "Hot Mastering" became the norm.

What exactly are you talking about?

[budwzr]

Modern audio/music is Mastered much louder in the last 10 years or so. http://en.wikipedia.org/wiki/Loudness_war

[newpball]

Modern audio/music is Mastered much louder in the last 10 years or so. http://en.wikipedia.org/wiki/Loudness_war

I am familiar with that but that is due to compression not due to normalizing.

Perhaps I am mistaken but it seems to me the OP is talking about normalizing not compression.

[budwzr]

Just explaining why older DVDs can peak at -20db or even lower. So there's room to raise the levels instead of Normalizing. That would preserve the original recording as intended by the movie maker.

So in a movie like Apocalypse Now or ScarFace, they jigger the volume by scene. Normalizing that type of movie would flatten the loud parts.

[hello_hello]

I hear normalizing is or can be damaging though and can cause clipping.

Nonsense, normalizing does not cause clipping.

However if you normalize to 0 db you have no room for anything else on top of that.

Technically it can, depending how it's done. If you normalise a wave file to 0dB and then encode it as MP3 or AAC etc, which would be a common menthod, after it's decoding and the waveform reconstructed the peak may exceed 0dB. That's one reason why some people recommend normalising to -3dB, although I'm not sure it's anything to fuss about too much. The newer methods of determining volume have "Peak", and "True Peak" options.

[pandy]

I hear normalizing is or can be damaging though and can cause clipping.

Nonsense, normalizing does not cause clipping.

However if you normalize to 0 db you have no room for anything else on top of that.

Technically it can, depending how it's done. If you normalise a wave file to 0dB and then encode it as MP3 or AAC etc, which would be a common menthod, after it's decoding and the waveform reconstructed the peak may exceed 0dB. That's one reason why some people recommend normalising to -3dB, although I'm not sure it's anything to fuss about too much. The newer methods of determining volume have "Peak", and "True Peak" options.

Clipping may occur in analog domain (assumption is that normalization is last step of signal procession - just before requantization) and as such normalizing to: 1/SQRT(2)= -3.103dBFS is recommended (in practical aspect this is loss of 0.5b i.e. half bit - beyond perception with properly dithered signal - especially with noise shaped dither)

[hello_hello]

I sometimes come across a DVD with very low volume to where the volume seems a bit low. Sometimes I have to turn my speakers all the way up to hear it as well as I would like and I can hear a constant hissing sound from the speakers being up too high. (It's easy to hear when theres no sound playing but their turned up all the way)

Don't forget some AC3 has dialogue normalisation. The idea is to adjust the entire volume up or down so dialogue from video to video it's always the same level and if there isn't any data in the AC3 audio the player is supposed to assume a certain volume. Invariably all that results in the volume being reduced to a certain extent and if it's 5.1ch audio being downmixed to stereo that might possibly make it sound even quieter, depending on how it's being downmixed.

Some of it might depend on whether there's dialogue normalisation and whether your player/decoder pays any attention to it. If not, that's not the problem, but you might find if you simply re-encode the audio to another format it'll sound louder even without peak normalisation as the "dialogue normalisation' is usually ignored when re-encoding.

When re-encoding stereo as stereo or 5.1ch as 5.1ch I usually don't use peak normalisation as I prefer to leave the volume as is it and I don't have trouble with the audio being too quiet. Peak normalisation often makes the audio a bit louder, but not always by the same amount. If you re-encode 5.1ch audio while down-mixing to stereo you usually need to "peak normalise" to prevent clipping when the channels are combined as stereo.

[Hoser Rob]

i don't have a problem with normalizing video but I never do it while encoding. Why not just do it in your playing software?

[pandy]

i don't have a problem with normalizing video but I never do it while encoding. Why not just do it in your playing software?

Normalizing should be done before encoding to fully use bitdepth.
Normalization is something else than perceived dynamics/loudness/dynamic compression.

[hello_hello]

i don't have a problem with normalizing video but I never do it while encoding. Why not just do it in your playing software?

Most player software normalises by increasing the level, then dropping it accordingly when there's a loud peak, then maybe slowly increasing it again (if it's configured to regain volume). It's generally a fairly horrible way to do it, in my opinion. It can't normalising in the same way peak normalisation does when encoding because it doesn't perform 2 passes.

[newpball]

Most player software normalises by increasing the level, then dropping it accordingly when there's a loud peak, then maybe slowly increasing it again ...

You are mistaken, what you describe is not normalizing but compression.

[budwzr]

Yah, that's a "look ahead limiter". The limiter squashes the peak via compression. The full sound is still there, but doesn't get clipped.

[Cornucopia]

A well-designed limiter still "clips", but usually in a smarter way (no single hard threshold, but a range). Basically, a special purpose compressor. And the full sound isn't still there in the same way - that's why the volume wars of the last 2 decades pissed so many people off: it ultimately "pinches" the sound too much and causes fatigue.

I think the OP needs to clarify: does he want all his programs to have a similar peak level (normalization), or does he want all his programs to have overall similarly perceived loudness?
The latter can be achieved by either un-normalizing (moving the level up or down to where the long-term/overall average hovers in the same area) or through compression (modifying the dynamics so all programs fit into a prescribed range), or a little of both.
BTW, AC3's dialnorm is, kind of like replay gain, a kind of UN-normalization (contrary to it's name). It just focuses on matching intended levels of the dialogue (as that is it's stated priority). And, like RpG, it is a metadata/flag-driven playback feature - it doesn't bake in the level like standard normalization does.

What the OP didn't mention in post#1, but I can guess is likely true he experienced and just doesn't understand why, is that a normalization of a file to a much higher level is going to present as much background hiss as the playback volume adjustment method. That's because neither normalization nor replay-gain nor dialnorm change the program's internal dynamics (range/spread/balance) and that includes its accompanying noise (which always exists to some degree). Raise the signal, raise the noise (whether saved to intermediate file or not).

The OP needs to learn about how to properly optimize the entire gain structure chain, and then decide what his programmatic volume priorities are, because you cannot have programs with different material and maintain BOTH their peak and their rms levels and homogenize them without also destroying some of their dynamics. "You can't have your cake and eat it, too." Something's going to have to give.

Scott

[hello_hello]

Most player software normalises by increasing the level, then dropping it accordingly when there's a loud peak, then maybe slowly increasing it again ...

You are mistaken, what you describe is not normalizing but compression.

I'm not mistaken. That's exactly how the majority of software player volume normalisation works. It doesn't normalise the same way as peak normalising when encoding as it can't scan the entire audio file in advance to find the maximum level. So the volume is increased by a predetermined amount, and when that would cause a peak to exceed maximum, the amplification is automatically reduced. Another higher peak would reduce the volume further etc, therefore it's possible for the amplification to be much less by the end of a video than it was at the beginning. So.....

If there's a "regain volume" option and it's checked, after each time the volume is reduced it's slowly increased again until the maximum amplification is reached once more. It's hard to determine the exact speed because it'd require an audio track with a loud peak followed by a period of very low volume or silence to check, but in the case of ffdshow it's normalisation option displays the amplification in dB so you can watch it slowly increasing. While a peak decreases the volume in an instant, I'd estimate the regain volume option takes roughly 20 seconds to increase the amplification by about 6dB. You are of course welcome to call that compression, but I call it slowly increasing the volume again.
However you describe it, the regain volume setting falls under ffdshow's normalisation option, as it does for MPC-HC's normalisation.

[budwzr]

I knew somebody was gonna get whacked. Glad it was newpball, not me. hehehe.

[hello_hello]

I guess you could argue that by definition a compressor reduces the volume of the louder bits, so normalisation can't be compression as it amplifies, but that's probably just semantics.

One of Potplayer's inbuilt normalisation options is referred to as TomSteady, and given most of Potplayer seems to be borrowed from other software, I'd guess that's this TomSteady. http://winampheritage.com/plugin/tomsteady/36297

I did play with it at one stage but I can't remember what it was like, however as a rule I think that sort of gain control tends to work better than traditional "compression" because it increases the volume of the quiet bits rather than trying to squish the loud bits as a compressor does, so it doesn't requires a compression threshold that matches the input volume to still work effectively. It'd probably be more effective than the tradition "turn the volume up slowly" type normalisation because the end result would more closely resemble "compression".
I've been using ffdshow's Winamp filter to run the RockSteady plugin on playback for years.

[newpball]

I guess you could argue that by definition a compressor reduces the volume of the louder bits, so normalisation can't be compression as it amplifies, but that's probably just semantics.

Just semantics....

Normalization is not compression, you can twist it any way you want but it won't make it the same.

The argument that you can interchange those terms because some software players misuse those terms is just absurd. Science is not a democratic process.

Normalization makes the loudest peak 0 dB (or any given value) and maintains the same dynamic range while compression means that the peaks are lowered while lower values are lowered less or not at all which causes the overall dynamic range to be reduced.

[turk690]

Normalization is not compression.

True. Some videohelp.com personalities nonetheless like to conflate the two anyway.

[Cornucopia]

@hello_hello, regardless of the # of non-professional apps that might like to conflate the 2 ideas, what you have been describing can/should only be described as "compression".

Normalization is ONLY a universal (whole-file), monolithic gain change (usually pre-calculated to raise the peak's peak to an arbitrary amount of headroom).

Programming history is full of examples where an app/setting/feature is created that purports to operate on a known/common parameter, but the actual parameter is NOT coded in line with the underlying scientifc understanding of that feature, and this has resulted in non-linear & surprising (buggy?) operation and overall consumer confusion. That's what happens when programmers themselves don't fully learn learn & clearly absorb the topic's fundamentals or follow professional nomenclature. I'm sure you've encountered this; I know I have.

There are not many times where newpball and I are in agreement, but this seems (at this point) to be one of those.

Scott

[budwzr]

it ultimately "pinches" the sound too much and causes fatigue.

Yeah, I like that term. I think Rock probably suffers from too much compression, but not dance/club.

[hello_hello]

Normalization is not compression, you can twist it any way you want but it won't make it the same.

I never said it was. In fact I only referred to software player normalisation was normalisation. You called it compression.

It's a fairly standard thing. VLC, MPC-HC, MPC-BE, Potplayer, LAV, ffdshow, AC3 Filter.... they all have a "normalisation" option that works as I described (although not all have the "regain volume" option).

Normalization makes the loudest peak 0 dB (or any given value) and maintains the same dynamic range while compression means that the peaks are lowered while lower values are lowered less or not at all which causes the overall dynamic range to be reduced.

Whatever the target peak level might be, the type of normalisation you refer to requires an examination of the entire audio beforehand in order to determine the peak level. Software players don't have that luxury. They must do it "on the fly". That doesn't make them compressors. They increase the audio by a predetermined amount and if the peaks don't exceed 0dB the whole audio is increased by that amount from beginning to end. Often though, the volume increase will cause a peak to exceed 0dB at some point, at which stage the volume is reduced accordingly. That's the price you pay for normalising "on the fly" but it certainly doesn't mean the audio was compressed. The dynamic range of the audio following the volume reduction remains exactly as it was, just at a lower volume. What compressor reduces the volume to compress a peak and then continues to apply the same gain reduction to any remaining audio? Do you often configure a compressor with a two hour release time? No you wouldn't. Aside from the fact it wouldn't make sense, for the next two hours it wouldn't be compressing, it'd just be applying a volume reduction.

Anyway, you can argue about it all day for all I care. Virtually every software player has a normalise function that does some sort of "on the fly" normalisation and that's how they do it. They all refer to it as normalisation. Any "regain volume" function that might increase the volume slowly again after it's been reduced is really just a variation on the usual "on the fly" normalisation and has about as much in common with real compression as manually turning the volume up and down using the volume control on a remote.

@hello_hello, regardless of the # of non-professional apps that might like to conflate the 2 ideas, what you have been describing can/should only be described as "compression".

Normalization is ONLY a universal (whole-file), monolithic gain change (usually pre-calculated to raise the peak's peak to an arbitrary amount of headroom).

That's exactly what software player normalisation does. Obviously you can't pick a universal gain change based on peak level as it's unknown, but you can still specify a universal gain change. Specify a universal 12dB gain increase and if the highest peak is -15dB you've just normalised to -3dB. If that's not the definition of normalisation I don't know what is.
Unfortunately though, when the peak level is unknown in advance, players need to be able to adjust the normalisation "on the fly" to prevent clipping. If that same file had a -10dB peak at some point the universal gain reduction would be reduced by 2dB. Do you really need to go back to the beginning of the video and play it from scratch again with only a 10dB audio gain increase before you can distinguish a volume adjustment from compression?

If you were going to refer to the initial amplification as something other than normalisation because you believe the laws of physics prohibit any type of "on the fly" normalisation, then maybe referring to it as a volume boost instead might be appropriate, but whether it be up, down or a combination of the two, extrapolating some slow gain adjustments as the audio progresses into "compression" seems quite a stretch.

[Cornucopia]

No stretch. True normalization is All-for-1/1-for-All. Anything else, "on the fly" for example, is NOT normalization. And, Slow or Fast, if there is adjustment to such extent that the internal dynamics change, it IS actually only some variation of compression/expansion. Normalization is BY DEFINITION a uniform adjustment. You (and possibly those apps) are playing fast & loose with longstanding standardized nomenclature. I'd call that a stretch.

Note that even the wikipedia article on normalization defines what you've been describing as "loudness normalization with clipping reduction through dynamic range compression".

I'm actually quite surprised at your response. I don't see you accepting fudging of similar processes in the visual realm.
Audio-norm, Photo, Video-Brightness
Audio-compression, Photo, Video-AutoContrast, Video-AGC

Scott

[hello_hello]

My response was in reply to a post by someone saying they prefer to use software player normalisation, and aside from expressing my opinion that it's not the best way to normalise, I simply explained how software players go about it. When a software player has a "normalise" function and it's enabled, the volume is increased by a predefined amount to begin with but reduced as playback progresses if necessary to prevent clipping. When a player's normalisation function also includes some sort of "regain volume" option and it's enabled, if the volume is reduced to prevent clipping it'll be slowly increased again until the pre-defined level is reached.

That's not opinion. It's fact. But of course newpball came along and decided to tell me I was mistaken because in his opinion I was describing compression. Unfortunately though he was once again wrong, because my description of typical software player normalisation was accurate.

For whatever reason, I seem to be able to cope with the concept of 1 pass normalisation where the volume might be reduced to prevent clipping if necessary. If it includes some sort of "regain volume" option I can still cope with including it under "normalisation" because that's usually where it's found in a GUI and as far as compression goes it'd have to be the worse compression ever.

http://en.wikipedia.org/wiki/Audio_normalization
Depending on the dynamic range of the content and the target level, loudness normalization can result in peaks that exceed the recording medium's limits. Software offering such normalization typically provides the option of using dynamic range compression to prevent clipping when this happens.

http://en.wikipedia.org/wiki/Dynamic_range_compression
Audio compression reduces loud sounds which are above a certain threshold while quiet sounds remain unaffected.

Because software player normalisation reduces the volume if necessary quiet sounds can be affected, and because the "regain volume" function increases the volume quite slowly and as a result quiet passages can slowly change in volume, it's not really like compression in that respect. However you describe it though, it's how software player normalisation tends to work.

Edit: I just noticed MPC-BE refers to their normalisation as "Auto Volume Control" which is probably a better description given "on the fly" normalisation isn't normalisation in the traditional sense but it's not really compression in the traditional sense either.

[killerteengohan]

Ill just use a lower normalization number thats big enough to raise volume without raising it too high for proper playback.

Thanks!