Mastering involves a lot of usually subtle adjustments to a mix, but the most obvious of these—and the one that many people nowadays think of as synonymous with Mastering—is setting the final loudness of the audio file. This simple step has given rise to more discussion and hand-wringing than anything else that's done as part of the mastering process. Let's take a look at what's involved, and why current practices have been so controversial.
Fig 1 Various Brickwall Limiters / “Loudness Maximizers”.
Pump It Up
We've all seen and heard much commentary about the so-called “Loudness Wars”—the practice of making recordings as loud as possible (often in a misguided attempt to make those songs “stand out” against others on the radio or in a playlist). Many artists and engineers (especially the mastering engineers who are tasked with it) feel that overdoing this practice of “loudness maximizing” kills the dynamics of the music, sucking all the life out of the song. And yet, if a song is significantly lower in level than others, listeners may sometimes subconsciously react negatively to the song, blaming the lack of impact on the music rather than just the lower relative level.
So, how can an engineer crank up the level of the mix, and keep it competitive (in terms of level, at least), without sucking all the life out of the music? Well, a careful hand at the controls is required, as is a good understanding of the technical aspects of the process of pumping up the volume.
Average vs. Peak Level
The perceived level (volume) of a recording is based on the average level of the track, not the peak levels within that track. But the limit on how loud the level can be raised before (unacceptable) overload distortion sets in is determined by the peak levels in the audio. For those unfamiliar with these terms, here's a brief explanation.
When a note is played there are two sections to the wave. There's the initial attack of the note, called a transient—the thump of the hammer hitting a piano string, or the stick hitting a drum—and the rest of the note—the body of the note, the part that rings out and sustains or gradually decays (see Fig 2). This is called the average level.
Fig 2 Transient Peak vs Average level.
The initial attack Transient can be 10–20 dB louder than the main part of the note, but lasts for only 10–20 milliseconds—too quick for that higher level to be perceived consciously. So the listener's sense of the loudness of that note is based on the average level—the level of the body (the sustain portion) of the note. The transient level is instead perceived subconsciously as “punch” or “snap”—the thump that hits you in the chest from a live kick or snare drum, for example. When you go to turn up the level of a mix, you have to raise the average level of the wave to make the recording sound significantly louder. But as you raise the overall level, the transients will reach the point of maximum digital level (0 dBfs, the top of the digital level scale), and begin to clip, producing harsh popping or clicking distortion, and preventing any further increase in the perceived overall (average) level.
Normalizing—Not the Norm
Many DAWs and mastering plug-ins offer the option to automatically set the level of a mix as loud as it can go, based on raising the loudest transients in the wave up to, but not beyond, 0 dBfs. This option is called “Normalizing”. But using this to set the final levels of songs means that a recording with lots of strong transients, like, for example, a high-energy rock mix with a prominent drum part, cannot be made as loud as a song lacking in strong transients, like a softer folk ballad with a lightly-strummed acoustic guitar (see Fig 3).
Fig 3 A song with strong transients (bottom) resulting in a lower Average level than a musically quieter song (top) without such strong transients.
If that rock mix, with its Average level restricted by the presence of drum transients in the song, were to play after the ballad in a playlist, it's lower Average level would be a letdown—the listener's initial subconscious reaction might be “eh.. this song just doesn't grab me”.
Engineers realized that if you were to limit the level of just the Transients in a wave, you could then raise the Average level higher, resulting a much greater perceived loudness. That can be done with a Limiter, employing a very high ratio and a very fast attack and release, so only the transients would be affected. In the digital era, this gave rise to a specialized class of software limiters, known as Brickwall Limiters (aka Loudness Maximizers), which are designed to accomplish this transient reduction as transparently as possible. This is how mastering engineers bring up the Average level of recordings—they limit the level of transients, and then they can safely raise the Average level by that same amount, without digital overload/clipping (see Fig 4).
Fig 4 A typical Brickwall Limiter/Loudness Maximizer has 1 control which simultaneously lowers the limiting threshold & raises the Average level.
These specialized limiters work well—recordings are louder than ever—but there are some caveats.
No matter how well-designed a brickwall limiter may be, there are still limits to how much transient-shearing you can get away with before the laws of physics catch up to you, and the sound of the recording begins to suffer. You can reduce transient levels by as much as around 6 dB or so without really noticing any difference—the beats may not have quite the same level of impact, but the mix will sound pretty much unchanged, except that the average level can now be up to 6 dB higher, a significant increase in loudness.
But current practices often try for even louder average levels, with even more aggressive transient reduction, and that's where you have to be careful. As you start to reduce transient levels beyond approximately 6 dB, it can begin to have a noticeable effect on the mix—drums (which are usually the first to be affected) may turn “mushy” (making it sound like a less aggressive, more musically lackluster performance), and the drum part can recede into the mix. If the drum transients were what was making the beat stick out, then that all-important rhythmic contribution might be compromised—I've heard songs where the kick drum all but disappeared in parts of the song from over-aggressive limiting, robbing the tune of the important back-and-forth of kick and snare that's so important to the groove.
If you keep lowering the limiting threshold, eventually you'll be left with no transient impacts at all. This lack of rhythmic punch can make the musical arrangement seem like just a big wash of sound, inducing listener fatigue more quickly than anyone would want.
Fig 5 A wave (of a mix) with its transients intact (top); after brickwall limiting has reduced transient levels and increased Average level (bottom).
Better Levels Through Science
So how can you avoid these issues, yet still bring the average level up to a point where your mixes will stand up against others? Well, the first thing to do is to make sure you're always aware of just how much punch and snap you may be sacrificing for loudness. Our ears tend to adjust to changes we make, so it's important to have the original mix ready for immediate A–B comparison (on another track), so you can always hear how far you've gotten away from it. Since you're making the average level louder, you'll need to temporarily lower the (channel fader) level of the Mastered version for the A–B comparison by the amount you've raised it in the Limiter/Maximizer. Then when you A–B the original mix against the maximized version, any potentially unacceptable losses in punch and clarity will be immediately apparent.
Another thing that some engineers do is to apply limiting in stages. I would sometimes find that, after applying Brickwall Limiting to a mix, the kick and snare were reduced too much in level, compared with the balance in the original mix, even though the limited sound of those drums was fine. When I had access to the individual mix tracks (or even just drum stems), I reduced the drum transients a little in the mix itself, re-balancing the slightly more compressed drums at that stage to compensate. Then when the Mastering Limiter was applied, there wasn't as much of a change, and the drum balance was better preserved.
Since you're pushing the level right up to the absolute digital limit of 0 dBfs, you'll want to insure that all playback devices can handle it. You should set the Output Ceiling control (the other control in a Limiter/Maximizer) to slightly less than 0 dBfs—around -0.3 dBfs is a fairly standard choice, though if you're not as worried about absolute maximum loudness, -1 dBfs would be even safer, and might help avoid a potential D-A reproduction artifact called “inter-sample peaks” (many Limiters/Maximizers have two options for metering, one that shows actual digital output levels, and another that anticipates this converter artifact).
Fig 6 Limiter/Maximizers, set for 7 dB of transient reduction/average level gain, with an output ceiling of -0.3 dBfs.
Even though it's still common practice to shoot for loud masters, the desire to be louder than everything else out there seems to be abating somewhat. While some artists/clients may still reject a mastering job for not being loud enough, more people seem to be willing to forego that last 2-3 dB of loudness—that extra level that often comes at the expense of overall mix quality. Mastering to a level a couple of dB lower that the very loudest songs out there can often be the best way to insure not only a reasonable relative listening level, but a good, punchy, still-dynamic recording as well, which ultimately will have a more positive effect on the listener than sheer volume ever could.
So, armed with a little knowledge and a careful hand on the controls, you can achieve good, loud recordings without taking away from the music—and that's what we're all really after!