Fact #1: LUFS and LKFS are reference units: R128 is a standard.
As of this writing, the current loudness standards are based on a document called the ITU-R BS.1770-4 which is a recommendation by the International Telecommunications Union on the implementation of a series of algorithms that measure perceived loudness and true peak levels.
The title of the paper is literally “Algorithms to measure audio program loudness and true-peak audio level”.
The EBU R128 is a document (among several others) outlining the European response to that recommendation.
The ATSC A/85 in the US, and the TR-B32 in Japan are similar documents/standards and are all in close compliance with ITU-R BS.1770-4, with minor differences.
So, EBU R128 is not equivalent to LUFS or LKFS. It’s like saying decibels are the same as the manual explaining them.
LUFS and LKFS are a new reference unit of loudness measurement, but they are not the standard itself.
FACT #2: LUFS and LKFS are terms which mean Loudness Units referenced to Digital Full Scale (dBFS) with K-weighting. (For more on K-weighting see Myth #4). As of 2016, LKFS and LUFS are exactly the same thing.
The Loudness Unit (LU) is equivalent to 1dB—that is, an increase (or decrease) of one LU is the same as raising or lowering by 1dB.
Here’s a bonus myth debunk: LU’s are NOT louder than dBs!
On an EBU R128 compliant loudness meter, a stereo -18dBFS sine tone at 1kHz measures -18 LUFS.
On an EBU R128 compliant loudness meter the scale can be absolute or relative meaning that on the meter itself you can set a specific target level to equal 0 LU or measure directly in LUFS.
On an “EBU Mode” loudness meter 0 LU = -23 LUFS (relative scale) or you can set it so that -23 dBFS/LUFS = -23 LUFS (absolute scale).
Here is a stereo -23 dBFS reference sine tone at 1kHz being measured by an EBU Mode meter on a relative scale reading -0.1 LU:
Here is the same stereo -23 dBFS reference sine tone at 1kHz being measured by an EBU Mode meter on an absolute scale reading -23.1 LUFS.
This is not as confusing as it first seems—on VU meters 0 can be calibrated to any desired reference level too, but typically 0VU is equal to +4dBu, which is equal to -20dBFS.
Fact #3: Well, yes and no. It is true that the documents described above are primarily outlining broadcast standards (that’s what the “BS” in ITU-R BS.1770-4 refers to) but there is growing evidence that YouTube, iTunes and other major online music streamers are implementing some kind of loudness averaging.
They are not necessarily adhering to any of the broadcast standards though, YouTube seems to be normalizing audio on some official videos to between -14 and -12 LUFS...
... and iTunes’ “soundcheck “ feature appears to be leveling audio to around -16 LUFS.
Radio has yet to get on board with the standards, but when it does, the need for any musician, bedroom producer, mix engineer or mastering engineer to maximize loudness via brickwall limiting, or mix to arbitrary peak levels will come to an end. (See Myth #5 and Conclusion.)
Fact #4: The loudness algorithms measure audio and adjust overall gain accordingly, they don’t process it.
Loudness Normalization uses EQ curves (designated K weighting) that closely resemble how the human ear perceives loudness, it then measures the average peak to trough difference of the entire “program material”, ignoring levels below a certain threshold, and then calculates a value called an integrated loudness level.
This Integrated level is then used to determine the overall loudness of the material and the levels of the whole program are turned up or down to comply with the various loudness standards mentioned above.
Furthermore, don’t confuse Loudness Normalization with Peak Normalization.
With Peak Normalization an audio file’s total gain is raised to specified amount (usually to 0 dBFS), but only based on the highest measured peak in the audio.
Here’s a plastic example; you have an audio file in which two characters are talking, and they are interrupted by loud gunshot. The gunshot nearly clips at -1 dBFS (i.e., the waveform nearly reaches 0dBFS) so peak normalizing the track will only raise the whole file by 1dB to 0dBFS, and leave the gain of the two characters talking perceptually unchanged — remember, raising gain by 1dB is barely perceptible to the human ear.
With loudness normalization the whole file is measured using the aforementioned set of algorithms and noise gates to determine the average or Integrated Loudness of the whole file. The algorithms and gates take into account the loud and quiet parts of the “program material” ignoring quiet parts (below a threshold of -70 LUFS as defined in the documents) and allowing for louder parts momentarily, and then spit out an Integrated Loudness value (I) after the whole file has been analyzed.
The Integrated Loudness level is the value that will determine the perceived loudness of the whole track.
When program material needs to be -23LUFs +/- 0.5 this is the value to check—it is important to understand that during playback parts of the material (music/dialog/EFX etc.) can be louder or softer than the target (I), but again, it’s the overall average loudness which is taken into account.
Referring to our plastic example above; as long as the gunshot peaks at or below a permitted momentary loudness level (EBU R128 specifies a maximum short term loudness (3 seconds or less) of +/- 5LU or -18LUFS) the whole file will be raised (or lowered) in volume “x” amount of LU so that the dialog (average loudness) sits at -23 LUFS while the gunshot (outside the average) has heaps of headroom to play into.
This will have more of an impact on the audience because the dialog is now audible and the gunshot has not been squashed down by a limiter or compression and thereby lessening the dynamics (and drama) between the two sonic elements.
The implications for more dynamics in music are also apparent.
Fact #5: Peak metering is rapidly becoming unnecessary, and essentially never gave us useful information to begin with. Intersample peaks are not correctly registered by peak-sample meters. For example, a traditional sample-peak meter that displays a max of -0.2 dB could read as high as +3 dB on a true-peak meter.
With the new Maximum True Peak Level of -1 dBTP, the previous PML (Permitted Maximum Level) -9 dBFS (as defined in ITU-R BS.645-2) is effectively obsolete and potentially replaces the previous music mixing standard for CD and online material of peaks no higher than -0.3 - 0.5 dBFS (once mastered).
As of this writing Logic X 10.2.2 has integrated True Peak measuring into all of its native meters and it is highly recommended to use true peak measurements from here on out.
As for RMS—RMS is much more useful for gauging the actual, longer term, levels of a given waveform, but RMS is only a measurement (or display) of signal voltage, so it doesn't really give us an idea of perceived loudness. Two music tracks measuring the same RMS values may not necessarily have the same perceived loudness because RMS does not take into account the psychoacoustic nature of apparent loudness as heard by the human ear, specifically that low, mid and high frequencies of the same level are not perceived as being the same loudness.
The Integrated loudness measurement specifically takes into account this aspect of human hearing perception of loudness and adjusts accordingly.
So what does all this mean for music?
In the EBU-R128 documentation it is explicitly suggested that no major changes to current mixing styles (as of 2016) are immediately necessary, but it is strongly recommended to consider the implications.
For music producers and engineers there are two choices:
When you mix/master your music track to the current standard of 16 bit 44.1kHz with peaks at between -0.3 and -0.5 dBFS with an average RMS of say -12dB to - 6dB (brickwall limited and loud) this track when measured with a EBU compliant meter will show levels way above -23 LUFs (and possibly true peaks upwards of +3dB) and thus will be turned down until it has an integrated loudness of -23 LUFS.
No compression, no further processing, just literally turned down.
What this means is that pushing for high RMS values and squashing out dynamic range will now actually work against your music when your “sausage” is played against music mixed to utilize the dynamic range afforded by the -23 LUFS mix headroom.
“Loud” over compressed and brickwall limited music - read: music with no dynamics - really cannot compete sonically with more dynamic material in the new standards.