Loudness Normalizing



We have discussed setting up your audio calibration to adhere to the EBU R128 standard. That gives us plenty of dynamic range headroom, to accommodate the most dynamic music, and to allow Crescendo some room to correct your hearing.

But we need to know what input levels to set for listening to an album of music. Or else we need to let Apple do its thing with SoundCheck. And while SoundCheck is a step in the right direction, it is flawed…

I find that, whatever algorithm is being used by SoundCheck, it isn’t consistent in establishing a playback level. It tends to vary between 6-8 dB louder than -23 LUFS. If it were providing a consistent elevation, then we could just set our input fader level and be done.

And then, there are times when SoundCheck simply fails. That happens a lot when I’m sharing music across several computers in my local area network. Remote files are simply not paying attention to any SoundCheck information, nor my indicated Equalizer settings, and I get blasted from time to time.

So, what I do is totally forego SoundCheck. I can’t rely on it. Instead, I wrote myself a simple utility to scan all the tracks in an album, and report all the loudness vital statistics to me. Here is an example from scanning the sound track to TRON Legacy:

The scanner reports back on every track in the album, and highlights the loudest one – the one with the largest value for LU23. That is the measured  R128 loudness in units above -23 LUFS. And that’s how much I need to back off on the input fader. All the rest of the tracks will maintain the level differences designed by the mastering engineer, and I won’t ever get blasted.

The other information lists the True Peak Level (TPL), the Loudness Range (LRA), and the Peak Ratio (PR). Loudness range is an indication of how dynamic the track is. Peak ratio tells you about the crest factors found on the track. But the most important information is that LU23 level.

The scanner also shows me graphs of my statistics on each track. 

The graph on the left shows the time history across the track, and the right shows me its histogram. The orange curve on the left is the accumulating long term loudness level, while the green trace is the 3s moving average. The orange bars along the left and bottom show the 10-95 percentile range found in those short term loudness measurements.

I wish it could be an automatic thing, to have your loudness controlled, as with SoundCheck. But there doesn’t seem to be a properly working system yet.

There is also a legitimate question of how one should decide on the loudness of an album. The mastering engineer had something in mind by arranging the tracks at the levels on the recording. Should we normalize against the loudest track on the album? Or should we run a loudness analysis over the entire album to decide, based on its final long-term loudness measurement? There are good arguments for either approach. My tool allows me to work both ways too.

  • DM

Author: dbmcclain

Astrophysicist, spook, musician, Lisp aficionado, deaf guy

Leave a Reply