Wow! Misconceptions about Loudness Normalizing



I just finished scanning for web documentation on the various loudness normalizing schemes in use today. While EBU R128 has become a de facto standard today, there are still lots of audio players out there using  something else. And that’s okay!

But people seem to have conflated understanding of loudness normalization when it comes to recording, mixing, and playback. So many confused opinions out there. There are a lot of reasons why you might want to set the 0 dBVU level much higher than -23 LUFS. That’s fine.

[ … if you are a broadcaster, then you are legally required to keep your system at -23 LUFS on transmission. ]

Maybe you really want to maximize your recording signal to noise ratio, and you only have 16 bits in your digital medium – like standard CD’s. So you intentionally record hot. Then there are DVD’s with 24 bits.

I would say that 24 bits is right on the edge of not needing to record hot, but that is debatable. Mixing engineers need some garbage bits on the bottom that they can feel free to play with, and not impair their recordings. 24 bits gives you some, but maybe not quite enough. And that’s also why a 16-bit medium, like a CD, may need to be recorded hot.

Understand that the record level on your medium is best set to maximize your signal compared to the noise floor. And the best way to do that, while maximizing your dynamic range in the recording is to use peak leveling, so you utilize every bit available. That doesn’t mean to over-compress and squash out all the dynamic range in the music, in trying to hug the 0 dBFS level. It just means that it’s okay to push your peaks up there.

But that has nothing to do with listening levels today. We have some pretty good systems for approximating human perception that can be used to automatically level out the loudness on playback, regardless of the levels used to record the music to the storage medium. How well these automated levelers work depends on how well they measure what we perceive in the music.

From the looks of it, today’s ReplayGain and SoundCheck are reasonably good. I saw a graph comparing the R128 levels against the ReplayGain levels on a collection of several thousand recordings. They line up pretty well along a unit slope. Only their zero points differ, and that’s fine. But what that means is that you can count on both R128 and ReplayGain to provide consistent leveling.

[ … right now I’m running the same kind of test against SoundCheck. Apart from implementation glitches, such as not performing properly across networks, the jury is still out for me. It seemed to vary much more than ReplayGain and R128. But impressions could be wrong, so let’s actually measure it…]

[ … where ReplayGain and R128 are inconsistent, sometimes by a lot, is in unusual sound recordings. If I normalize ReplayGain against R128 using a Pink noise calibration, then they depart significantly with music playback, and even more on a 1 kHz calibration tone.

But on most music, they seem to differ by about 5 dB, with ReplayGain claiming -18 LUFS as its target level for 0 dBVU, against -23 LUFS for R128. But how loud these sound depends entirely on how you set your Volume control. These LUFS readings only show up on the mixing console, but shouldn’t matter to your ears. ]

The level you choose to listen to depends on your environment. For example, when listening while driving in a car, you would be hard pressed to appreciate the full dynamic range in the music, so some degree of dynamic range reduction is frequently used for those occasions, and you probably set the playback level much higher too.

The important thing is that the auto-leveling system accurately respond to the music the same way we do, or at least with consistent difference. Once we have leveled music, it is up to us to decide just how loudly to play that music back. But we can depend on it not suddenly blasting us out of our seats.

So whether you want to use your own personal standard of -16 LUFS or -14 LUFS, or the R128 standard at -23 LUFS, that’s entirely up to you in your recording and mixing.

No one level is “better” than any other for this purpose, as long as it captures the whole dynamic range of your music without clipping. On playback they will sound the same, under auto-leveling control.

But since we cannot physically exceed 0 LUFS, you do allow for much more dynamic range when you set your reference level lower. A lower LUFS recording level gives you more headroom for forteforte passages, without causing your track to clip. You don’t lose playback volume! That’s up to you with your Volume control on the amplifier.

So we need to keep these concepts separate in our minds. How hot a recording is on the CD should have no bearing on your playback levels. The mixing engineers just decided that their material did best at those hot levels, making maximum use of the limited bit-range available to them. How loud it sounds is up to you and your faithful auto-leveling system.

Let’s hope that your auto-leveler is a good one.

  • DM


Author: dbmcclain

Astrophysicist, spook, musician, Lisp aficionado, deaf guy

Leave a Reply