Loudness and Dynamics

With the current “loudness” use in broadcasting and streaming services of ITU-R BS.1770-4, EBU R128 and other international standards, the audio engineer who is mixing or mastering should be aware of the metrics and requirements that will be imposed on any music files used in these delivery formats. Dynamics processing can change the loudness structure of music to help, or hinder, the final mix, and knowing how to monitor loudness while using a tool such as DynOne 3 can help streamline the mixing/mastering process. As a note though, we would like to point out that music is art and completely subjective. Therefore it is always recommended to mix and master with the intent to make it sound the best possible, given the artist’s intent, without loudness considerations. However, sometimes it’s useful to be aware of them and to know what will happen to your songs once they reach the platforms.

In addition to mastering for compact disk and vinyl release, which each have their own limits, today one must consider the current streaming and broadcast requirements. These delivery systems now use an “average” loudness target rather than an absolute maximum peak value like previous streaming systems and compact disks. The previous use of peak limits led to the “loudness war” that encouraged musicians and producers to demand ever louder mixes, attained using heavy compression, which resulted in music with average levels as high as -3.5 dBFS. Such compression creates distortion that may work for a few genres (Heavy Trash Deathcore Grunge Metal), but not for many since high levels of compression can make music feel constrained even when played loudly. The “loudness war” also caused problems listening to a varied selection of songs since the highly compressed ones sound louder than more dynamic mixes, requiring frequent adjustment of the volume control by the listener. 

In the new standards the average loudness is measured using LUFS which stands for Loudness Units relative to Full Scale. Using LUFS is an attempt to measure sound in a way that humans hear it. There are three defined LUFS measures, Momentary, Short Term and Integrated, each of which employs a two stage K-weighting filter that attempts to model perceived loudness of the human ear as a function of frequency. Momentary and Short Term measures are useful to estimate loudness in “real time”, with 400 msec and 3 second averaging times respectively. Integrated Loudness is used over an entire song and is measured using both the K-weighting filter and “level-gating” to prevent periods of silence or very quiet levels to skew the measurement low. DynOne 3 provides estimates of all three metrics enabling you to monitor loudness as you mix or master.

There has been a lot of discussion in recent years about “dynamic range” in music and the need to let music “breathe”. To audio engineers the term “dynamic range” is familiar for describing performance of power amps, preamps and microphones as the ratio of the maximum usable signal to the noise floor level. However, in regard to the new loudness standards, the term “dynamic range” is defined as the measure of the difference between the true peak signal level (in dB) and the loudness measured in LUFS. The new loudness standards include two metrics for this “dynamic range”, one measured over a short average time of three seconds (PSR – Peak to Short-term loudness Ratio) and one measured over an entire song (PLR – Peak to Loudness Ratio). Note that since dB and LU (loudness units) are both logarithmic scales, a difference in dB or in LU is a ratio in terms of linear signal levels, so this difference is correctly called a ratio. And since a 1 dB change in loudness is the same as a 1 LU change, we can justify using the difference between loudness measured in LUFS and the true peak in dBFS. The units for PSR and PLR have been debated, but either dB or LU is meaningful.

What is True Peak? With digital audio 0 dB is the highest possible sample peak and any audio signal trying to exceed that level is simply clipped. True Peak is a measurement that takes into account inter-sample peaks caused by digital-to-analog conversion and lossy formats like MP3, AAC, and Ogg Vorbis. A sample peak is the highest possible value of an audio data set, but a true peak level may exceed that value, typically by one or two dB, and up to 6 dB or more. The ITU-R BS.1770-4, EBU R128 and other international standards suggest limiting the estimated true peak of a file to as low as -2dBFS.

If you are making a mix of a loud genre specifically for CD release, you may want to use an average loudness as high as -9 or -8 LUFS and maximum peaks approaching 0 dBFS, but for any streaming service or broadcasting you should consider what the platform will do to your music. Basically, they will reduce the integrated loudness, if needed, to meet their target. Although various audio streaming and broadcast platforms use slightly different targets for Integrated Loudness, they fall in the range of -14 LUFS, so providing a track with an integrated loudness of -14 LUFS and true peaks no higher than about -1 LUFS will help create great sounding mixes that should play well on any platform. In addition, you may have some sections of a song play louder than -14 LUFS as long as other sections of the song are lower than -14 LUFS. Short term loudness levels as high as -9 LUFS for the loudest chorus may work if offset by quiet verses.

Note that True Peak is also considered by these music services and if a song when played at their chosen target still has true peak levels over -1 dBFS to -2 dBFS, the level may be turned down further or some form of limiting applied. If you have a mix that meets the target loudness and sounds great, but has a few true peaks exceeding -2 dBFS, you can apply a good limiter yourself to the final mix so the streaming services won’t do too much damage (they will still convert it to some lossy format, but that should not affect average loudness and dynamic range much).

What these standards and metrics mean can be best illustrated with some examples using DAW waveforms and ITU-R compliant plots.

The waveforms below illustrate how a streaming service would “turn down” a song that is a casualty of the loudness war – this song has an integrated loudness of -3.4 LUFS and a PLR of only 4.5 LU as measured using the ITU-R BS.1770-4 / EBU R128 standards. Even though its sample peak is 0 dBFS, it’s estimated true peak level actually exceeds +1 dBFS. On the left is the waveform viewed on DAW at the level it would have been played using the old peak loudness standard (allowing a maximum sample peak level of 0 dBFS) and on the right is the same song adjusted to -14 LUFS. It is no longer very loud.

While mixing or mastering you can use the meters of DynOne 3 for guidance, and there are tools, such as the Klangfreund Multimeter (https://www.klangfreund.com/), that can accurately measure all the defined metrics of the new loudness standards. Below is the Multimeter analysis of the entire “crushed” song. 

On the left the Multimeter shows some song metrics for the whole song plotted over time, and on the right are overall song metrics (note that we have omitted several metrics to make this example clear – the Multimeter can plot and summarize all of the new standards’ metrics). The orange line is a plot of short-term loudness (3 second moving average), the yellow line is Integrated Loudness at the plotted time, and the lower pattern is a graphical representation of PSR over time. Note how the PSR collapses to “nothing” by the end of the song as the short term loudness is increased for “effect”. In fact, the last minute of this song is so distorted that it sounds like it’s being played through a ripped speaker cone! Possibly it was the artist’s choice, but for listeners it pretty much destroys any impact of the song.

On the right we see the Integrated Loudness measure (I) of -3.4 LUFS (this is before normalization by a streaming service – this was measured directly from a CD). The maximum True Peak is +1.14 dB, which would contribute further to distortion on a streaming service (not that this song needs any more distortion!). We also see a number (+10.6 LU) in the yellow loudness bar above the Integrated Loudness readout which indicates this song is 10.6 LU too high for this streaming service (with an assumed target of -14 LUFS). This is the reduction used in the first screen-shot, approximately a 70% drop. The Multimeter also shows Loudness Range (LRA) of 5.0 LU which is a measure of the short term loudness variation throughout the song, from start to finish, and you can observe with the orange line that this is approximately the change from the very beginning of the song to the last few seconds. This song has small “wiggles” in its short term loudness, but no clear variation between choruses and verses. The overall Peak to Loudness Ratio (PLR) is 4.5 LU, an unfortunately low dynamic range.

Below is shown the Multimeter analysis of a more dynamic song that did not participate in the loudness war, even though it was recorded years after the “crushed” song. Again the Short-Term Loudness is plotted as an orange line, Integrated Loudness at the plotted time shown as a yellow line, and the lower pattern shows both greater PSR overall and greater variation in PSR over time than the crushed song. On the right we see Integrated Loudness (I) is -11.0 LUFS, Loudness Range (LRA) is 10.7 LU, and PLR is 11.3 LU, a very good dynamic range. Note how variation in the Short-Term Loudness between verse and chorus acts to keep the overall Integrated Loudness from rising too high. This is a good mixing technique, mentioned earlier, that can add variety to a song and keep overall loudness down so the streaming service won’t turn it down for you. In this case loudness would still be reduced by 3 LU to meet a -14 LUFS target, but the song would retain its PLR of 11.3 LU. 

How would these two songs compare when played on a streaming service where both are adjusted to the same average loudness? Below is a section of both songs at the same scale and same integrated loudness – the top pair of channels is the more dynamic song and the bottom pair is the loudness war victim. You can see (and if you were to listen, clearly hear) that the more dynamic song is actually louder than the crushed song at some times, and quieter at others, but even in its quieter moments its peaks are about the same level as the war victim’s peaks. The more dynamic song sounds much better, with more “movement” and clarity.

DynOne and RootOne Can Help Shape Dynamic Music

To produce a fine, dynamic and appropriately loud song you first need a good song, a good arrangement, and a good performance, recorded well! But even the best recorded performance can be ruined with an inappropriate mix. And a good performance can be enhanced with the proper mixing techniques.

Both DynOne and RootOne can help optimize a mix, even expanding dynamic range if needed. DynOne can lift up the low level detail of music without squashing the internal dynamics, which can improve clarity, and even help a song sound louder when it is not technically louder. RootOne can strengthen the lowest octaves to provide a solid bass foundation while also creating harmonics that can help a song play well on bass-limited speakers.

The plots below show both the Klangfreund Multimeter analyses and DAW waveforms of the same mix before and after processing with DynOne 3 on the master bus. As mentioned earlier, the LUFS measure attempts to model perceived loudness of the human ear, but there is more to psychoacoustical perception than simple loudness, even when filtered to approximate the human ear’s “frequency response”. Micro-dynamics and macro-dynamics can make music sound significantly louder without increasing the LUFS measurement!

Although you can’t hear the examples below, you can observe in the waveform plots that the lower levels of the corresponding tracks are a little louder in the lower plot (green waveform is thicker). The Multimeter plots show short-term loudness (orange line) is very similar before and after the DynOne processing, but the PSR is consistently higher throughout the song. In the summary measurements you can see the PLR is greater and the LRA slightly smaller in the “louder” version, even though Integrated LUFS are identical. You can also observe the maximum short-term loudness is slightly higher in the louder tracks. In addition there is a psychoacoustic effect not observable in these analyses – there are very short variations in loudness due to different filter bands acting at slightly different times and different levels. This provides “movement” that improves clarity and is perceived as a louder mix.

Although the new loudness standards are intended to make all music sound “equally loud”, DynOne 3 can make your music louder! 

Written by Dennis J Wilkins.