Volume vs Gain?

I have been wondering for months now how peaks / levels in individual tracks combine to make composite peaks / levels across multiple tracks. I still don’t fully understand it. Can someone explain it to me? I’m curious about the science / theory side of things as well as the hands on “do ______ to achieve better volume in your songs” advice.

I know that it definitely IS important to keep individual tracks under clipping. Most instruments sound crackly and just plain bad when they exceed 0 dB. So I have all of them set to a max of -.1 dB, usually actually more like -3 or -6 (I understand this is standard practice if you were ever to send stems to be mastered). And up until very recently I always did my best to avoid clipping on the aggregate / track level i.e. a song consisting of multiple tracks. But lately I’ve been experimenting with just saying “f%&^ it” and just ignoring the clip meter in Audacity when mixing a full song. I’m going well over clipping in multiple spots, which I know is bad in theory, but it doesn’t SOUND bad! It doesn’t sound bad in the way that a single track over clipping sounds bad. It actually sounds good.

Not sure if it matters, but I think I’m mainly getting the clipping on percussion / transients, things like kicks and snares mainly. Are the peaks of quick transients so brief that the “crackle” of clipping is inaudible? Is it there, but I’m just not hearing it because it’s masked by the transient? If so, then is clipping really an issue?

So does this relate to the difference between volume and gain? I can get a track up to -.1 dB (maximum without clipping), but then I can turn the gain up a number of decibels above zero? And it doesn’t necessarily start to sound like sh17? Huh?

I also read somewhere that sometimes during mastering engineers will use analog equipment (instead of digital) to get the signal to go above 0 without clipping, hence boosting the overall loudness of the track. At least I think that’s what I read.

Can someone help me make sense of this? Use all the jargon and science talk you want, because I want to understand this fully.

Thanks!

Outside of Audacity, in the regular 16-bit/24-bit world, exceeding 0 causes the digital system to run out of numbers. The file just stops following the performance during those brief instances when the blue waves try to smash their way to heaven.

Audacity internally works at 32-bit floating which doesn’t do that. So you can apply a filter or effect that causes the blue waves to go way over clipping and you can recover by just reducing the volume. If you tried to export a WAV file while the show was too loud, the export would die a grizzly death.

You may like the louder than normal show inside Audacity because loud always sounds good. That’s the loudness war thing. My music is louder therefore better than yours.

We recommend -6 for sound peaks during live performing to give the sound a fighting chance over background noise and still give a little “up” room for artistic expression. That’s doable for most people paying attention to the Audacity meters, particularly if you undock them and make them really big.

Koz
LiveRecordingLevels-650.jpg

That is the golden rule :slight_smile:

but like most rules, there is an exception…
While you are working in Audacity, assuming that the tracks are all in (the default) “32 bit float” format, then “over 0 dB” does not automatically mean you’ve blown it (luck for us).

“Normal” audio files have an absolute limit of 0 dB and it is impossible to exceed that limit. Attempting to do so just clips the tops/bottoms of the peaks at 0 dB, and once they are gone they are gone - repairing clipping is (almost) impossible (if the clipping is so slight that it is not audible, then it may be repairable, but once it gets to the stage that it is audible then the damage is permanent and irreversible).

“32 bit float” format is “special” in that it can handle audio over 0 dB. Of course, if you try playing audio that is over 0 dB then it will still sound horrible and distorted, but in 32 bit float format you are able to recover from the problem by “amplifying” the audio to a lower level (back down below 0 dB) and then it will be OK again (hooray).

The magic words are “32 bit float”. Most audio formats are “integer” format, and they just can’t go above 0 dB without damage, so if you export a file that is over 0 dB, then the exported file will be damaged.


The “Gain” control on the left end of each audio track controls how loud the track plays. It does not change the actual audio data (which in on your hard drive), it just changes the playback level. It is therefore safe to move that up and down as often as you like and it will not affect the audio data that is on disk.

When playing, audio will sound distorted if the level goes much over 0 dB. There are two ways to deal with that - either use an effect (such as “Amplify” or “Normalize” to reduce the volume (these effects change the actual audio data), or adjust the track Gain control. Use whichever method you prefer. The playback level can be seen (in dB) on the playback meters. Generally you would want to keep the level a little below 0 dB to keep it sounding nice.
IT IS ESSENTIAL THAT THE PLAYBACK LEVEL DOES NOT GO ABOVE 0 dB WHEN YOU ARE READY TO EXPORT THE FINISHED RECORDING.

If you have more than one track in the project so that they play at the same time, the playback level will be higher than either of the tracks individually. “Mixing” tracks “adds” the sounds together, and so makes the overall level increase.

Before exporting a multi-track project, take care that the peak level does not hit 0 dB.
There are two ways to do that:

  1. Play the entire project and check the playback meters.
  2. Mix the tracks down to a single track (select all the tracks, then “Tracks menu > Mix and Render”), then use the Amplify or Normalize effects to set the peak level.
    I generally use the second method and Normalize to -1 dB for WAV files, then Export, then Undo the “Mix and Render”, then save the project. I like to save the project with the tracks separate in case I want to go back to the project and change anything.

See, this is what I’m trying to wrap my head around. I AM exporting wavs that are well beyond clipping. Red lines everywhere. But when I listen to it, it doesn’t sound bad! No $h1tty clipping sounds, no crackling. It all sounds really good, and way louder than what I’ve been able to achieve before. The RMS is sooooo much bigger than it ever has been on any of my previous tracks. So what gives?

I understand that if you ever gave stems to someone to mix and master, they’d probably just send you packing if you gave them files with clipping. But if I’m just self-producing songs, and the “corrupted” wav just gets turned into a final product, an mp3, and they both sound fine even with the clipping, then what’s really the problem with clipping? That’s what I’m struggling to understand.

I have no intention of winning the loudness wars. But it does suck when your tracks are like 1/3 as loud as anything else you listen to that was professionally mixed and mastered. How the hell do they do that?

Yup, I get that. Because clipped audio in a single file generally sounds like crap. It’s my “produced” final product tracks that are well above clipping (but still sound good) that I don’t understand.


See, but this doesn’t. That’s what I don’t get. As I was asking before, is this because the peaks are basically all percussion? I know when I try to boost something with long, continuous waves, like basically any melodic part, it DOES sound like crap. Basslines, synth parts, whatever – crap. But not so with the combined peaks (mainly percussion) in these multi-part tracks.

Not sure if this tells you anything, but when I go into the wav and “amplify” by -.1 dB, all the red lines go away. I can see that the tops of the peaks (basically all of them coincide with kick and/or snare hits, and I think basically ONLY on kick and snare) are squared off. But it doesn’t actually make anything sound bad!

So, yeah, I’m still confused. I’m breaking the rules, but it doesn’t sound bad, and I’m getting better loudness. What gives? Am I just sooooooooooo good at mixing that the rules don’t apply to me? :wink:

At the risk of offending all my drummer friends, they’re basically in the business of producing intense harmonic distortion with a beat.

Most distortion sounds like harsh, peaky, intense, crunchy sounds mixed in, for example, with a mellow, gentle musical performance. If you added crunchy, sharp overtones to a cymbal crash, who would know? You could probably mis-manage a snare drum sound very badly and only the drummer would know that it’s not what it sounded like when he played it.

Drums are stunningly hard to mic because they’re all momentary transients and edges. It’s not like you can play a drum for ten seconds like the oboe player does for her “A” so the orchestra can to tune to it. Traditional US “VU” sound meters don’t even see drums giving the edge to the BBC PPM, and even that meter doesn’t measure them all. The first time you realize you have damage is after you find out the digital recording system ran out of numbers and whacked off the tops and bottoms of the sound waves.

It’s hard to wave off a newbie when they insist they’re going to mic the drums with one microphone. Probably not. Drums have to be carefully recorded so one drum doesn’t blow holes in the recording while the others are loafing. And etc.

I’m not shocked if you made a muffled drum recording much better and brighter and crisper after you overloaded it. In general, whatever makes the performance work is good, but be very clear you are making an accidental recording. Down the road is the performer who wants the drums just exactly the way they played them. Now you have to actually mic and record them. Now what?

When you Export a sound file from Audacity, it converts from its robust, hard-to-overload internal format to a format that does overload: WAV, MP3, etc.

Koz

Let us hear a bit of that.
Select 2 or 3 seconds of good sounding red lines everywhere audio, then “File > Export Selected Audio”. In the Export dialog screen, select “other uncompressed formats”, then click the “Options” button and set the format to “WAV (Microsoft)” ans “32 bit float”. Check that the file size is under 2MB, and add it to your reply using the “Upload attachment” option (below the message composing box).

Listen, I get that you guys are giving me sound advice that you would give to anyone, especially a noob, and I want to let you know I understand what you’re telling me, and that I know it’s good advice, in general. I get that “avoid clipping” is a general maxim that has a lot of merit to it. I’m just trying to dig into the particulars, the reasons behind it all, and the potential exceptions.

Here, maybe I can put this question in a different way to help you understand what I’m really getting at. Is there a certain amount of a very specific type of clipping that might be considered acceptable? Why do so many DAWs (also Audacity) have VU/clip meters that are sensitive to the NUMBER of samples that are clipped? Six consecutive clipped samples seems to be a common practice. What does six samples come out to in milliseconds? Is it longer than the typical peak of a transient? If clipping should be avoided at all cost, then what reason could there be for having a VU/clip meter set to anything EXCEPT “show me any and all clipping, even single samples?” Do you see what I’m saying? I assume the reason is that at least some people are not worried about clipping in very transient sounds i.e. drums. If there is some other explanation for this, please dish.

From Wikipedia: “Quasi-PPMs have a short integration time in order to register peaks longer than a few milliseconds in duration. In the original context of AM radio broadcasting in the 1930s, overloads due to shorter peaks were considered unimportant on the grounds that the human ear could not detect distortion due to momentary clipping. Ignoring momentary clipping made it possible to increase average modulation levels. In modern digital audio practice, where quality standards are hopefully much higher than AM radio in the 1930s, clipping of even short peaks is usually regarded as something to avoid.”

So, I get that it’s totally normal, standard industry practice to avoid clipping in all its forms. But I’m just questioning whether I personally need to be so concerned with it in my self-produced tracks if it’s only happening to the transients, and a listener can’t even tell. I guess I’m thinking of it in more of a “Proof’s in the pudding” kind of way. If it sounds fine – it’s fine. The reason for the “avoid clipping” rule is to prevent unpleasant sounds, right? I export as wav, back it off by -.1 dB so the file itself has nothing out of range (but still lopped off peaks, of course), it sounds fine, and RMS is much higher than if I was obsessively avoiding clipping. It gives me more freedom to mix it how I want.

Just to clarify – the song in question is not a recording. It is comprised of wav drum samples (all acoustic drums, studio quality recordings) and wav files generated from a softsynth for the melodic parts. None of the sounds are “muffled.” They’re all high-quality samples/wavs arranged into parts by yours truly. No one individual sound is going above 0 dB. It’s just the composite track (and the exported wav) that is going over 0 dB, and only in the transients.

Can you give a little more explanation / context to this? Why do they not measure it? If these work for U.S. professionals, then doesn’t that mean they prefer them to be that way for some reason?

Yup, the exported wav has hacked off peaks that seem to be affecting only the transients/percussion, not any of the melodic stuff, hence no annoying crackly distortion of extended / melodic sounds.

I’ll have to get to that when I’m at home. But if you’re hearing unpleasant distortion, your ears are a lot better than mine!

Is there more than one option for wav? I always just select the first one that pops up – I assume that’s the 32 bit float, right? I hope it is.

The usual “WAV” option is 16 bit. To give it its full name: “Microsoft PCM 16 bit integer WAV”

If you select “other uncopressed files” as the file format, then there are lots of other options, including WAV 32 bit float.
As discussed previously, 32 bit float has the magical property that peaks over 0 dB can survive, thus we will be able to see precisely how much clipping is actually occurring.

I think this is what I use already, because when I open up an exported wav with clipping, the red lines are there.

Also, how many seconds of a song is 2 MB on average?

Here are some online sources that I found that confirm my suspicions that intentional clipping of transients is sometimes used to achive better loudness in a mix. They’re going about it in a different way, of course, but I think I’m able to “get away with it” because I’m already controlling the dynamic range so tightly through careful editing / construction of dynamic range, and the way things are mixed, only transients are affected. Same end goal, different way to get there.

http://onlinemixingmaster.com/secret-mastering-techniques-they-dont-want-you-to-know-1/

http://www.soundonsound.com/sos/mar12/articles/loudness.htm#Top

Well, here it is. You’ll see a huge blotch of clipping on the combined kick/tabla hit at about 1 second, and then a number of smaller areas of clipping in tons of spots. Personally, I can’t hear any distortion or crackling.

I’m not sure how you have managed it, but you have defeated the magic of 32 bit float format. The peaks are entirely “chopped off” at 0 dB.
I presume that at some point in the process the audio has been converted to an integer format.

This is a close-up of your audio sample:


As you can see, the peaks are banging into the 0 dB limit.
To see if the peaks have survived, we can amplify the level down a little:

No, the peaks are gone and lost forever.

The clipping is only occurring on the loudest peaks, and there is so much buzzy high frequency stuff going on in the music that it ‘masks’ the distortion.

The reason for the “avoid clipping” rule is to prevent unpleasant sounds, right?

Not exactly. The object of the recording medium is to present an accurate copy of the performance. When you clip it stops doing that. The drum did something in real life that didn’t make it into the show. You can intentionally make file system defects part of the performance if you want, but it’s not good practice.

The ANSI C16.5 American VU meter (attached) was designed in the dawn of recorded history. Widely considered completely terrible, it was the same terrible everywhere, and so became valuable. When you sent a show down the radio network from Chicago with certain meter movements, I can confirm that it got to Los Angeles with the same meter movements, and thus was probably the same show at the same quality.

Anybody who has ever tried to record a drum solo with one of these things found that because of the intentional sluggishness of the movement (I’m not making that up. It’s part of the ballistics specification) they almost always got really hot, sometimes damaged recordings. The oldbies learned the conversion. “Run drum solos around -10, or lower.”

Koz
Screen Shot 2015-02-25 at 18.37.21.png

I’m not totally sure what that means, but the original thing I uploaded was indeed exported as a 16 bit wav (thanks for clueing me in to the difference). Check out this current one, which was exported as 32 bit. I think it sounds exactly the same to my ears.

Yup, and because of the way it’s mixed, those peaks are all percussion. None of the synthy melodic stuff is even close to 0 dB, and it’s not being affected

I don’t think that’s actually it, at least not entirely. Check out the drums only attachment. I’m picking up some distortion in the big combined kick/tabla hit (1.2 seconds), but not in anything else, even though you can see clipping abounds. And I think I’m only getting distortion in the big combined kick/tabla hit because it is so long, so many continuous clipped samples. So I probably do need to dial that back.

So, yeah, I think it really is the transient / number of clips thing. When it’s only very few, it’s not actually audible.

That sounds like great advice for actually recording drums, but I’m not recording these drum parts. But I do record things sometimes, including some percussion, and I am mindful to avoid clipping when I do. Because if you clip when you record, you have something that you can’t control, a performance that is lost etc. What I’m doing, as a stage of mixing/mastering, that’s not happening, because nothing is irrevocably lost.

My reading of that Sound-on-Sound article was not that it was advocating digital clipping, rather that it was warning of the dangers, but saying that you may sometimes get away with a bit of clipping on some types of sounds (which is pretty much what I’ve been saying :wink:).

Generally “soft clipping” is preferred over “hard” digital clipping. “Soft clipping” is an effect that truncates peaks (like hard clipping) but without the abrupt corners. In effect it has squashed down the tops of the peaks, retaining some of the original waveform shape, albeit with squashed tops/bottoms. The benefit of this is that it has the same “maximizing” characteristics, but less aliased harmonics. In other words, it is a less bad sounding distortion :smiley:

If you want to try “soft clipping”, try these plugins:
http://wiki.audacityteam.org/wiki/Nyquist_Effect_Plug-ins#Tape_Saturation_Limiter
http://wiki.audacityteam.org/wiki/Nyquist_Effect_Plug-ins#Broadcast_Limiter_II
http://wiki.audacityteam.org/wiki/Nyquist_Effect_Plug-ins#Broadcast_Limiter_III

Yes, the Sound on Sound article is definitely more conservative/cautious in its stance. But if you read some of the sources in the other link, you can find engineers going into detail about how they actually go about it as a deliberate choice, what equipment they use etc. Some of them are more like “hey, I’m just doing this because that’s what the client wants!” but others are more willing to defend it as being a viable alternative to compressing/limiting the sh1t out of everything. I share that same goal (not compressing/limiting everything), so that’s why I’m interested in finding alternative approaches. I want my music to be louder (mainly because it sounds better to me that way, but also because people don’t take it seriously if it can’t compete), but I don’t want to completely squash the dynamic range in pursuit of that goal, which I think a lot of people do with overuse of compression. I work hard to build in dynamics in what I do, so it absolutely kills me to think about compressing that. Intentional use of transient clipping (soft or hard) is one more tool in that fight.

Yup, and I am most definitely going to look into those methods now that I have a better understanding of all this stuff. But for the time being, I am not afraid to allow hard clipping if I can’t hear it. I think this is a case where you can “break the rules” if you know what you’re doing, and you’re mindful of how far is too far.

Excellent! I saw those while perusing the plugins a few weeks ago, but I honestly had very little idea what they were, so I skipped past them.

Thanks for hashing this out with me, guys

Sure, but I’ve heard all sorts of abominations holding high the flag of “mastering” :wink:

Leaving aside to one moment what clipping does to the sound and to your ears, there is another consideration which is what it does to the equipment. Clipping is the single most common cause of tweeters being destroyed. Hard clipping creates vast amounts of high frequency content, which may be beyond the audible range (particularly for people that have hearing damage caused by overexposure to loud noise - which tends to reduce high frequency sensitivity first). The amounts of high frequency energy can exceed “normal” levels by a very large factor, causing the electrical current through the tweeter coils to be many more times greater than their design specification. Current causes heat, and tweeters can literally be “fried” by clipped audio.
https://www.google.co.uk/search?q=clipping+tweeters

I don’t take any of it as the final word. Jut another approach / perspective to consider.

How does soft clipping fare in this way?