Limiter processing is one of the hot topics on the internet about sound processing. Its close relation with mastering and loudness leveling makes it an unmissable tool for sound and music production.
In this first article, in a series of three, we will have a very basic look at limiters to help the less experienced to sort out what’s going on. In the next article we will go much deeper, so stay tuned!
If you are in a hurry, skip down to the TL;DR section of this article to get a summary, as well as a video explanation of the subject.
What is a True-Peak limiter, and are there different types of limiters?
The most common definition of a limiter, is to consider a limiter being a compressor with a ratio superior to 10:1. During this series of articles, we will assess that what we call a limiter is a True-Peak limiter, which has more constraints than the previously mentioned type of limiter.
The limiters we are interested in here, the true-peak limiters, are dynamic tools specifically designed to reduce audio crest, very much like a safety guard preventing any clipping from the digital to analog stage. Thanks to the peak reduction it is possible to use such a limiter to increase the loudness of the input signal.
Why the need for limiting?
The very last process of music production is to set the overall level, or loudness, of a mix. Limiting is the only safe way to amplify the loudness of a mix, but it comes at a cost.
A limiter will cut the crest of the signal to create some headroom to allow amplifying the rest of the signal. The cost of limiting is a loss of dynamic and an additional distortion.
The true-peak level is the actual level of the samples, or of the waveform if you prefer. The loudness is closer to our sound perception and smooth out quick sound variation because they do not matter that much in our perception of sound loudness.
Using a limiter also provides the guarantee to never clip the digital-to-analogue stage. This is a very important safeguard and explains why there is always a limiter at the end of the chain, even if it doesn’t do much.
Limiting will always come at the cost of more saturation added to the signal, but in a very much more transparent way than just cranking the output gain and clip the converters. Clipping the converter is considered as a technical error (even if some popular master does clips at the converting stage).
When is there a need for limiting ?
If you want to make a mix louder without clipping the digital-to-analogue converter way beyond the red, you will need a limiter. The limiter will reduce the peak and provide you with headroom to amplify the whole gain without clipping the output stage.
Limiting is also often needed to conform a mix to certain norms. For example, most music streaming platforms will refuse a mix with a true-peak level higher than -1 dBTP.
Is limiting mandatory ?
Limiting is mandatory in the sense that a mix should never exceed 0 dBFS. So using a limiter with a threshold at 0 dBFS will always prevent that from happening. Most of the time, the different target platforms (streaming, broadcast, etc.) ask for mixes that do not exceed -1 dBFS.
Increasing the loudness of a mix is never mandatory. Maybe we will create a debate here, but loudness in music production is very much an aesthetic decision. So, to continue with these controversial topics; a louder mix does not sound better than a quiet one. Actually, it sounds less dynamic and more distorted. Also, there are no norms in music diffusion. Each and every platform has its way to handle the loudness of submitted audio files.
For example, we often encounter the idea that a good deliverable should have a loudness of -14 LUFS-I with a true peak never exceeding -1 dBTP . This value comes mainly from the Spotify guidelines. But, it is not entirely exact, as Spotify offers different loudness targets for their customers. There is a loud (-11 LUFS-I), normal (-14 LUFS-I) and quiet mode (-19 LUFS-I). Apple has recently moved from -16 LUFS-I to -18 LUFS-I and before 2022, YouTube normalized loudness at -12 LUFS-I (now -14 LUFS-I). So which one should we choose? The common consensus is around -14 LUFS-I because it covers the biggest user base.
Then, what happens with a file that is above the target? If a file is submitted with a loudness target above the recommendation of the platform, the file volume will be simply dropped by the number of dB necessary to match the recommendation. So the process is transparent to what you’ve mixed and mastered.
If a file is lower than the target, most of the streaming services do nothing about it. The file will simply be quieter than the other one. YouTube used to be an exception before 2022, but Spotify in loud mode will limit the content to match the -11 LUFS-I target.
So how do we handle this mess? It seems there are three possible solutions.
- Follow the most common recommendation, aka Spotify (-14 LUFS at -1 dBTP)
- Align on the loudest one to make sure that no processing will affect your work (at the detriment of the dynamic range)
- … Don’t care about it?
Actually, the last point is the one defended by the author. Loudness and more importantly dynamic range is not only a technical thing, it is also an aesthetic choice. Some genres of music have built their aesthetic on very compressed and saturated sound, where others want to have all the accessible dynamics.
As a general guide, we will simply assume that it is a best practice to never exceed -1 dBTP. Also, it is preferred to have the loudest peak of a program, or a song, to always hit this -1 dBTP target.
Mastering limiters are designed to reduce the crest of a mix and allow it to increase its loudness. Their true-peak characteristic allows them to never let a sample cross the threshold.
Limiting should always be used to prevent a mix from clipping the digital-to-analogue converter. However, due to the many different loudness targets found in the streaming services, it is difficult, if not impossible, to “go-to” recommendation as for the loudness of a track. It seems to be more an aesthetic choice than a technical one, at least, in the music industry.
dB ? dBFS ? LUFS ? What is it all about ?
There are quite a few acronyms and concepts to explain around sound pressure level and how it is measured. Because sound is a mechanical wave, the primary way to measure the sound pressure level is to measure how the pressure evolves in a space.
First, the relation between sound pressure, and how we experience the sound level, is not linear. For example, when the sound pressure is doubled, we do not perceive a sound twice as loud. In fact, to have a sense for a sound being twice as loud, we need to multiply the pressure by ten. This is why we express the sound pressure level in decibels, which is a logarithmic scale that is much closer to our perception. When the sound pressure is doubled, there is a gain of +3 dB. When the pressure is multiplied by ten, there is a gain of +10 dB.
Depending on the field of interest, there are many different units built around the decibel scale. The one that is used in digital sound is the dBFS, or decibel full scale. In the digital domain, sound is represented by samples, whose amplitude can take absolute values between 0 and 1. The number of actual values that a sample can take between 0 and 1 is defined by the quantification (16 bits, 24 bits, etc.). But this is a linear scale, and thus, it does not correspond to our perception of sound. The dBFS solves this problem. A value of 1 in linear corresponds to 0 dBFS, a value of 0 in linear corresponds to -inf in dBFS (-96 dB at 16 bits, -144 at 24 bits, etc.).
Now that we have a scale that behaves closely to our perception, we need to find a way to measure sound loudness. Here, a peak measurement (the value of each actual sample in digital sound) is not a good candidate, because fast sound variation in volume does not matter that much in how we perceive loudness. Also, the frequency has a strong impact on how loud a sound seems. This is why engineers have proposed the loudness unit.
There are different time windows for the loudness measurement : momentary, short-term, long and integrated, which correspond to the following citation from the EBU Tech 3341:
1. The momentary loudness uses a sliding rectangular time window of length 0.4 s. The measurement is not gated.
2. The short-term loudness uses a sliding rectangular time window of length 3 s. The measurement is not gated. The update rate for ‘live meters’ shall be at least 10 Hz.
3. The integrated loudness uses gating as described in ITU-R BS.1770-4. The update rate for ‘live meters’ shall be at least 1 Hz.
In the music industry, it is the integrated value that is used as a reference for streaming services.
Next article in this series
Part 2 – Limiter Theory – Knowing your tools