Mastering has become a critical step in producing a CD or DVD — almost as important as tracking and mixing. Much of this final step involves tweaking various loudness levels, a process called dynamics control. Two of the most common types of dynamics controls are normalizing and maximizing. Here, we'll look at their similarities, their differences, and why they are important components of a mastering engineer's toolbox.
What Floats Your Boat?
Imagine that you control a harbor's water supply and that you can make the boats go up or down at will. Now imagine that all the boats are under a bridge and that you can raise the boats only until the tallest mast brushes the underside of the bridge. If you must raise the boats further, you can shorten the tallest masts, but if you raise them too high you risk causing a flood.
FIG. 1: The sine wave pictured below has two small amplitude peaks: one before normalization (top) and one after normalization (bottom).
The harbor scenario described above — all boats are raised until the tallest mast hits the bridge — is the basic concept behind normalizing and maximizing. A normalizing algorithm scans an audio file and finds the highest sample level. It calculates the percentage that that sample needs to be raised in order to hit 0 dB Full Scale (0 dBFS), and then it raises every sample by that percentage. Many normalizers allow you to choose what your maximum level should be, whether it's 0 dBFS or some percentage (with 0 dBFS being 100 percent of the maximum possible level). Fig. 1 shows a sine wave with two amplitude peaks: one before normalizing and one after.
Normalizing is useful when the lowest levels in an audio file are a bit too low. It's also fairly easy computationally, which is why just about any audio editor can do it. Normalizing can also be useful when transferring samples from a computer to a device such as a Kurzweil sampler, which adds a great deal of dynamic headroom. Normalizing in the Kurzweil allows the samples to play at optimum levels.
Normalizing is commonly used when mixes are assembled into an album. Often, different tracks have been recorded and mixed at different times and places. Each sounds fine in isolation, but when they're put together there can be overarching differences among tracks, such as overall EQ and volume levels. Normalizing allows you to set the volume context of your album, with carefully adjusted volume balances among the tracks.
Normalizing, however, works only if you have the headroom for it. The top panel in Fig. 2 shows a sine wave with a peak close to maximum. As the middle panel shows, normalizing accomplishes next to nothing. In real life, an occasional peak from a plucked string or snare drum can defeat normalizing. Adding insult to injury, that peak might not make much difference in volume, because the ear doesn't judge loudness by occasional peak levels; rather, it judges loudness by RMS levels. (RMS is a mathematical method of measuring the effective power of an AC signal. When an AC signal is an audio signal, its RMS value is closely tied to its perceived loudness. For sine waves, the RMS value is 0.707 times the peak amplitude level.) If those occasional peaks have little effect on the overall RMS level, they'll likely have little effect on peak loudness.
FIG. 2: Normalizing works best when there''s dynamic headroom in the file to support it. In the top figure, the large peak limits the effect of the normalization, as shown in the middle image. Maximization is more effective (bottom), but the price is a loss of dynamic range.
Maximizing is an attempt to inject your tracks with digital steroids, increasing the overall RMS level of a file without clipping it. The audio is first compressed, lowering the tallest peaks. With the added headroom, the track can be normalized, allowing all peaks to be raised. The bottom part of Fig. 2 shows a maximized sine wave.
Maximizing depends on the quality of the compression, and some software gives you lots of control over the process. The Normalize feature in Sony's Sound Forge, for example, has an option called Normalize Using Average RMS power and includes settings for the compressor's attack and release times. Sound Forge has the ability to apply varying amounts of normalization to different frequency ranges within a file. You can also keep extremely quiet and silent sections from being processed by setting the appropriate threshold in the Ignore Below option.
The Waves L1 Ultra-maximizer plug-in has a proprietary look-ahead technique that anticipates upcoming amplitude peaks in the file and trims them down before it's time for them to go into the DAC. It's like an ultracompressor, lowering peaks with zero attack and release times.
Loud … Louder …
While maximizing can be a beautiful thing, it's also possible to get carried away. And our ears can fool us: because of the way our brain processes sound, when the same material is played at different volume levels, the louder one often sounds more present (at least in the short term). That increased presence adds impact, especially when broadcast over the radio, which leads many producers to adopt an “ours goes to 11” approach to audio levels. (I even heard an audio producer at a seminar describe mastering as intended primarily to raise the album to the loudest possible level.)
Maximizing raises a file's RMS level, but it does so at the expense of dynamic range. Look again at the bottom of Fig. 2 and notice how squashed everything has become. Too much maximizing can result in music with little dynamic range and a distorted sound. A great deal more wisdom on this subject can be found on mastering engineer Bob Katz's Web site at http://digido.com.
It's also possible to be fooled into thinking that a DSP operation has improved things, when in truth all it's done is made things louder. To address that pitfall, TC Electronic's Finalizer Express Studio Mastering Processor has a matched-compare feature that adjusts the gain following processing to match the preprocessing volume level. Processed and unprocessed versions can be compared at the same apparent volume levels so that users aren't misled by a volume increase.
All in a Dither
Contrary to occasional appearances, however, maximizers do not perform miracles, and their results come at a price. Like any DSP operation, normalizing and maximizing inevitably exacerbate that Achilles heel of digital audio, quantization error (see “Square One: What's in a Word?” in the May 2005 issue of EM). Maximizing, like many processing operations, bumps up a file's word size while it does its work. Eventually, those words have to be pared back down, and maximizers make that trimming part of their standard procedure. That is one of the selling points of the Waves L1 Maximizer plug-in, whose proprietary Increased Digital Resolution (IDR) is optimized for trimming samples back down to the target word size (24-, 16-, or even 8 bit) following maximization. Ideally, maximization is a last-step procedure, particularly if its redithering involves noise shaping. It's meant to be a final coat of polish, not something that gets done over and over.
Effective normalizing and maximizing are subtle, fine arts that are learned over time. Used properly, they can add a touch of class to your album, giving it that final coat of polish and making every track sound as though it belongs with all the others.
Mark Ballora is trying to master teaching music technology at Penn State University.