“Loudness Leveling” Denotes a Vague Description of Two Discrete Processes

Scores of audio producers in the Podcast Production space have adopted an inaccurate term when referring to basic Loudness Normalization: Loudness Leveling.

First – what is Loudness Normalization? Actually, it’s quite simple:

Audio is measured in it’s entirety. The existing Integrated (Program) Loudness is determined. A gain offset is applied relative to a spec. based or subjective Integrated Loudness target.

For example: if the source audio measures -20 LUFS, and the Loudness Target is -16 LUFS, +4 LU of gain will be applied.

As well, a True Peak Max. Ceiling is defined, which again may be spec. based or subjective. If the required Integrated Loudness gain offset results in overshoots – limiting is applied in order to maintain compliance.

It’s important to note that Loudness Normalization does not correct wide variations in audio levels. As well – it does not guarantee optimized intelligibility for spoken word. If an audio piece (e.g. multiple participant segment) contains inconsistencies as such, the Loudness Normalization gain offset will simply elevate (or reduce) the relative perceptual loudness of the audio. The original dynamic attributes will persist.

That’s it. There’s nothing more to it unless the Loudness Normalization tool features some sort of dynamics optimization process that may or may not be active.

For the record – the Loudness Module included in iZotope’s RX 7 Advanced Audio Editor applies basic Loudness Normalization (measurement, gain, and limiting). It does not apply optimization processing.

View this source clip waveform. There are two participants with noticeable level inconsistencies:

This is the same clip Loudness Normalized (to -19.0 LUFS). The perceptual loudness is higher. However the level inconsistencies persist:

Leveling is a process that addresses and corrects noted inconsistencies and level variations. It is accomplished by the use of gain riding plugins and/or specialty tools that rely on complex algorithms. One basic example is the use of an “RMS” Compressor featuring an optimal and often extended release time parameter.

This is a “leveled” version of the original source clip displayed above. The previously persistent level inconsistencies no longer exist.

Finally, this is the leveled audio, Loudness Normalized to -19.0 LUFS. The described processes were in fact discrete.

I hope I’ve made it clear that the term Loudness Leveling is not an accurate term to describe Loudness Normalization. The key is that Loudness Normalization is gain and limiting. It does not correct inconsistent level variations. You’ll need to implement discrete Leveling processes to address persistent inconsistencies.


Technorati Tags:

Leave a Comment