Loudness Normalization: Part 2 – The Standards
If you bounce your mix in Logic (or any other DAW), you have to know about “Loudness Normalization” – SERIOUSLY. No matter whether your mix will be played on the radio or Spotify, if you upload it to your YouTube channel, or just listen to it in your iTunes Library, your mix will be affected and you better know how. In this three-part series, I will explain all the details. Part 2: The Standards.
Part 1: What’s the Problem?
What is Loudness and why will it cause major problems with your mixes if you don’t understand some underlying principles (that I explain in part 1):
Part 2: Loudness Normalization Standards
Never ever use the Peak Normalization on your bounced mix, and instead, embrace the new global standard of Loudness Normalization (that I explain in part 2).
Part 3: Logic’s Loudness Meter
Instructions on how to use Logic’s Loudness Meter and other tools to avoid surprises due to Loudness Normalization when you later hear your songs on iTunes, Spotify, Youtube, etc (as I explain part 3).
Subjective Loudness Measurement
A Broken System
As I already mentioned in part 1, we are facing a few challenges when listening to songs or any audio program on broadcasts or streaming services, and even our own iTunes Playlists:
- The audio material often has different perceived loudness that requires to manual adjust the playback volume throughout the program.
- Standard Level Meters only measure electrical signals and won’t reflect the characteristic of our ears based on perceived loudness.
- Peak Normalization to 0dBFS is only a technical standard that doesn’t relate how we perceive loudness.
- The Loudness Wars make matters even worse, because it results in sound degradation or in other words, shitty sounding mixes when overly compressed.
How to fix the System
The first step in fixing the problem is to come up with a method for Subjective Loudness Measurement. This is nothing new and in the past there were different proposals and implementations trying to tackle that problem:
- VU Meter and RMS Meter: This is the traditional method of reading the average signal levels which happens to be close to perceived loudness.
- ReplayGain: This was a standard proposed by David Robinson in 2001 that is still implemented in some audio players. https://en.wikipedia.org/wiki/ReplayGain
- K-System: Proposed by renowned mixing and mastering engineer Bob Katz, this is a studio monitoring calibration system combined with special level metering to achieve a consistent dynamic range http://www.aes.org/technical/documentDownloads.cfm?docID=65
- ITU-R BS.1770: This is the title of the now global standard for the so-called “Loudness Normalization”. It is implemented by broadcast systems in many countries and adopted in some variations by most of the online streaming services.
The New Concept – “Loudness Normalization”
The major achievement of the new Loudness Normalization standard is that brilliant engineers found a method of how to measure Perceived Loudness with special “Loudness Meters” and establishing a “Target Level” that the audio program is referenced to. That means, a shift from Peak Normalization to Loudness Normalization, so all audio is normalized (referenced) to that Target Level and, therefore, establishing a consistent Perceived Loudness for the listeners throughout different audio material.
Peak Normalization vs Loudness Normalization
Before getting into the details about Loudness Normalization, here are three simple diagrams (inspired by Florian Camerer’s entertaining AES presentation from 2011), that demonstrate the concept of Loudness Normalization compared to Peak Normalization.
➊ Peak Normalization
The different items in the diagram represent various audio files with different dynamics. With Peak Normalization all those files are moved up to the red line to have the same 0dBFS Peak Level (red dot). The green dots represent the individual loudness of the various files based on their different dynamics and, therefore, the different perceived loudness that varies when playing the audio material in sequence. That means you have to adjust the playback volume.
➋ Peak Normalization + Compression
To compensate for the loudness variation, you could compress the audio files (as many radio stations do) to achieve a more consistent loudness. However, this would affect the quality of the individual audio files, especially if the compression is automated based on a single generic setting.
➌ Loudness Normalization
This is now how it’s done, the new standard based on three steps:
- Program Loudness: Each audio file (song, commercial, TV Show, etc.) is analyzed to determine its individual loudness, the so-called Program Loudness (green dot).
- Target Level: The broadcaster, streaming service, and even your iTunes app sets a fixed Target Level (green line) that the Program Loudness level (green dot) of each individual audio file is aligned to, the procedure called “Loudness Normalization”.
- Peak Level: An absolute Peak Level (red line) is still maintained to prevent clipping above 0dBFS.
As you can see, each audio material is played back with its original dynamics at a playback level that ensures that every song has a similar loudness. By setting the Target Level pretty low (-16LUFS … -24LUFS), the system has enough headroom before clipping to allow the playback of dynamic audio material and ends the urge to produce hyper-compressed mixes.
Basic Loudness Normalization Procedure
Here is the simplified procedure how Loudness Normalization works in action. This is the crucial part to understand, because it shows you how you should mix your song, and more importantly, how not to mix your song to avoid being “punished” and played back at a lower level than other songs.
I explain the the unit “LUFS” (Loudness Unit, referenced to Full Scale) in a moment, but for now, think of it as of the regular dB scale (1 LU = 1 dB, 0 LUFS = 0 dBFS).
#1 – Target Level
A system (broadcaster, streaming service, personal music player, etc.) that uses Loudness Normalization sets its Target Level (-14LUFS … -23LUFS). This is the Loudness reference that all the audio material played on that system is aligned to. Remember, “the good thing about standards is that there are so many of them”. This is also true for the Target Level. European broadcast systems use a Target Level of -23LUFS, America uses -24LUFS, and most audio streaming services use a higher Target Level close to the AES recommended -16LUFS.
#2 – Measuring individual Program Loudness
Any audio material (songs, commercials, long-form shows, etc.) that is played on the system is first analyzed to determine their individual Program Loudness (different procedure for live broadcasts). That means, for example, the signal level of a song is measured from the beginning to the end with a standardized Loudness Meter, to find the integrated “Program Loudness” for that specific song. For example, -10LUFS.
#3 – Apply Gain Offset
When a song is played on a Loudness Normalized system, it looks at the Program Loudness for that particular song and raises or lowers the level to adjust the Program Loudness of that song to the Target Level of the system.
A few thing to be aware of:
- The song itself is not processed, only a gain offset is applied if required.
- If the Target Level of the system (i.e. iTunes) is -16LUFS, then a song with a Program Loudness of -18LUFS is raised by 2dB and song a with a Program Loudness of -14LUFS is lowered by 2dB.
- Only a song with the Program Loudness of -16LUFS plays back without any applied level offset.
- A highly compressed song, resulting in a high Program Loudness of -4LUFS, will be played back on that system 12dB lower and ends up having the same perceived loudness as a track with a healthy dynamic. The hyper-compressed track just might sound worse due to its distortion and lack of transients and nuances.
#4 – Prevent Clipping
The system still checks the absolute level of a song to avoid any clipping. For example, an individual song with a low Program Loudness will not be raised to the Target Level and only as much so the Peak Level of that song would not go over 0dBFS of the system. See the Peak to Loudness Ration (PLR) below.
#5 – Compress (optional)
The Loudness Normalization standard also defines a dynamic range value of a song (“LU Range”). A broadcast system has the option to use this information to compress the audio material if it is necessary to comply to a specific dynamic of the transmitting channel or program.
Although international standards are in place to set the rules for how to implement and use Loudness Normalization, there are a lot of variations published in additional or competing standard proposals. In case those acronyms pop up here and there, here is a quick overview. For the curious reader, I will list a few links to those documents and presentations at the end of this article.
The ITU (International Telecommunication Union) was the first international body in 2006 that came up with a tool to objectively measure the subjective Perceived Loudness of the human ear. It was published in the standard named ITU-R BS.1770.
The EBU (European Broadcasting Union) later formed its own research group to extended the ITU-R BS.1770 standard and published their recommendation as EBU R128. Further revisions of the ITU-R BS.1770 standard adopted some of the EBU R128 recommendations.
The AES (Audio Engineering Society) published their own recommendation for Loudness Normalization for streaming audio in the paper AES TD1004.1.15-10.
The ATSC (Advanced Television System Committee) in the US published their paper known as ATSC A/85, the “Recommended Practice: Techniques for Establishing and Maintaining Audio Loudness for Digital Television”.
There are other national standards like the TR-B32 for Japan or the OP-59 for Australia. The individual streaming services like iTunes, Spotify, SoundCloud, YouTube, Tidel, Pandora, etc. don’t commit to any of the international standards (at least not publicly), but most of them have implemented Loudness Normalization standards, each choosing their own Target Loudness Levels.
Keep in mind that the topic of Loudness Normalization is a “moving target”, because commercial streaming services can change their implementation without necessarily announcing it to their users.
Now lets get a little bit into the details of Loudness Normalization to learn the terminology and everything around it to better understand how that system works and understand how to use Loudness Meters when you mix in Logic or any other DAW.
Units: LU and LUFS
The ITU standard established a new unit, the LU “Loudness Unit”. This unit is similar to the dB units for signal levels, LU just indicates that it refers to the loudness level and not the signal level.
- Relative Loudness Unit [LU]: The difference of 1 LU is the same as 1 dB (raise by 1 LU means raise by 1 dB). “LU” can also be written as “LK”.
- Absolute Loudness Unit [LUFS]: LUFS, Loudness Unit, referred to full scale (which can be pronounced “loves”) is referenced to 0dBFS.
LUFS vs. LKFS
Please note that there is some inconsistency with the labeling. The European standard uses the unit “LUFS”, but ITU and the US uses the unit “LKFS” (Loudness, K-Weighted, referenced to digital Full Scale). Both units are the same. The letter K is just an unfortunate choice, originally established by the ITU. It has nothing to do with the “K-System” proposed by Bob Katz, instead, it is referring to the so-called “K-Weighing Filter” that the Loudness Meter uses to measure Perceived Loudness.
The Program Loudness, as we discussed already, describes how loud a section is, no matter if that section is a 20s commercial, a 5 minutes song, or a 2 hour feature film. It is measured from the beginning to end of a section, determining its integrated average loudness, using the unit LUFS (Loudness Unit referenced to digital full scale).
Target Level (Target Loudness)
The Target Level (or Target Loudness) is the reference level of a system (broadcast, streaming service, playback app, etc) each song, commercial, or TV-show will be “normalized” to based on their individual Program Loudness.
True Peak Meter
In addition to the Loudness Level, you still have to consider the Peak Levels of an audio recording. The ITU-R BS.1770 standard defines a new method on how to measure peak levels. Instead of measuring Quasi-Peak with conventional “Sample Peak Meters”, it requires the use of True Peak Meters to determine the peak levels of the program. This is important not to introduce distortion due to clipping (during conversion or transmission).
True Peak Meters use (at least 4x) oversampling to detect inter-sample peaks. The standard defines a maximum allowed peak of -1 dBTP (or -2 dBTP for data-compressed material). The reason for the headroom is that True Peak Meters based on oversampling are still not perfect and can produce errors.
Peak to Loudness Ratio (PLR)
The Peak to Loudness Ratio is the difference between the maximum True Peak Level of a program and its Program Loudness (its integrated loudness). This is basically the headroom between True Peak and Program Loudness.
This value is important in a situation where, for example, the Program Loudness of a song is -20 LUFS and the True Peak is -1dBTP. In that case the PLR is 19. If the Target Level of the streaming service, however, is -16LUFS, then you end up with three options:
- The song is clipping by 3 dB
- The song is turned down by 3dB
- The song is limited by 3 dB
Method to Measure Perceived Loudness
The most important step in Loudness Normalization was in 2006 when the ITU came up with a method on how to actually measure the subjective Perceived Loudness. At its core it uses a simple weighting filter, called the “K-Weighting Filter” that reflects the non-linear frequency response of the ear (see Equal-Loudness Contours).
Careful with the “K”
There are other types of filter curves (A-weighted, B-weighted, Z-weighted) mainly used for measuring sound pressure levels (dB SPL). Unfortunately, the ITU chose the letter “K” for this curve, and also uses it as the unit for the absolute loudness level “LKFS” (Loudness, K-Weighted, referenced to digital Full Scale). This can cause confusion with the “K-System” or “K-Metering”, the metering and monitoring concept proposed by Bob Katz. It is even more confusing, because the K-System tries to tackle the same problem with loudness but it is different from the Loudness Normalization based on ITU standard.
The Loudness Meter also uses a so-called Gating Method that improves the measurement of the Program Loudness for programs with high dynamic and sections of silence. It ignores the silent or very soft parts of the program so it doesn’t push the average level (Program Loudness) down and ends up with being too loud when normalized to Target Level.
The system actually uses two Gates
- The 10LU Gate follows the measured program and gates signals below 10LU.
- The 70LU Gate ignores unwanted noise and also functions as an auto-start for the loudness measure. You can start the measurement, but it only reads the signal when it detects a proper signal level.
EBU-compliant Loudness Meters
The “EBU R128” standard determines the features and functionality of a Loudness Meter. Because many of the manufacturers of plugins and audio devices were part of the group that drafted the standard, these EBU-compliant Loudness Meters are wildly available from inexpensive ones to the professional plugins from Izotope and NuGen Audio. Even Logic has a built-in EBU-compliant Loudness Meter plugin.
The concept of measuring the Loudness Level based on EBU R128 can be applied to all areas of audio production, production, broadcast, transmission, archiving, etc.
Here are the main parameters and controls found on Loudness Meter:
- Target Level: A marker or color code lets you set the Target Loudness Level that you want to use.
- Integrated Loudness (I or or ILK): This is the meter that measures the actual Program Loudness. Unlike conventional meters, this one requires that you actually start and stop the measurement to determine the integrated average loudness measured during that time window (from the beginning to the end of the song). For this “stopwatch” functionality, most Loudness Meters provide some sort of Start and Stop buttons, or use autostart.
- Momentary Loudness (M or MLK): This meter displays the current level with a response time of 400ms. Please note that this is not as fast as a Peak Meter that has around 10ms response time. Unlike the Integrated Meter, this meter has no Gate, so you can see the actual level no mater how low it is.
- Short-term Loudness (S or SLK) : This meter displays the current level with a very slow response time of 3s similar to an RMS Meter. It is a “rolling 3s time window” that continuously displays the integrated loudness of the last 3 seconds of the currently played program. It also uses no gate mechanism.
- Loudness Range: The Loudness Range (“LU Range” or “LRA”) is usually a numeric display that shows the dynamics of the measured material (i.e. your song). Please note that this is not the raw dynamic range of a song from the lowest to the highest level. Instead, it measures the average dynamic, the “Statistical Loudness distribution by excluding extreme levels” (this was developed by TC Electronic and made available to the EBU to incorporate in the open EBU standard). The Loudness Range value is an indicator if any dynamic treatment of that program is required under specific circumstances (distribution platform, playback devices, etc.). For example, a radio station, limiting the program to a specific dynamic range.
- True Peak: Some Loudness Meter also have a True Peak Meter or numeric readout so you don’t have to open a separate True Peak Meter Plugin to avoid any clipping.
Loudness Meter Scales
There are two scales for Loudness Meters.
- Absolute Loudness Scale (0 LUFS = 0 dBFS): The absolute scale reads like a standard Level Meter where the 0LUFS is the same as 0dBFS. You set the Target Level (i.e. -23 LUFS) with some type of marker or color code and that is your reference during the metering.
- Relative Loudness Scale (0 LU = -23 LUFS): Most engineers are used to a 0dB mark as a reference when they follow the meters during recording or broadcasting. This relative scale takes that into account by setting the Target Level (-23LUFS or any other Target Level) to 0LU on the scale so engineers don’t have to change their metering habits.
Advantage of the Loudness Meter
With the new Loudness Meter we now know three things about a program:
- What is the loudness (how loud it is?)
- What is the true peak level (highest absolute value)
- What is the dynamic of the program
Please note that a conventional Sample Peak Meter doesn’t provide any of that valuable information!
Conclusion Part 2
Equipped with the understanding of all the controls of a Loudness Meter and their functionality, I will discuss Logic’s Loudness Meter and other tools in the third part of this article, plus some tips and considerations on how to use Loudness Normalization.
Latest posts by Edgar Rothermich (see all)
- Loudness Normalization: Part 3 – Logic’s Loudness Meter - June 16, 2017
- Loudness Normalization: Part 2 – The Standards - June 14, 2017
- Loudness Normalization: Part 1 – What’s The Problem? - June 13, 2017