If you're burning CDs at home, you'll obtain the best quality if your audio data is at its highest possible average signal level. Most people yawn at this and reach for the normalisation button on their audio editor, but, as Craig Anderton explains, unless you're careful, you could be doing your material a disservice...
So, you want the highest average level possible on your home‑produced CD. Why? Because everyone else does. In fact, there are a whole class of tools — compressors, normalisers, level maximisers, you name it — designed to do just that. After all, loud is good, right? Given two identical sound sources, people invariably identify the one that's slightly louder as better (consumer alert: when buying speakers in a hi‑fi shop, check whether the brand the shop wants to push is slightly louder when compared against other speakers).
Frankly, I'm tired of recordings that use only the upper 6dB of a CD's dynamic range. Dynamics should be a part of music, and I always thought part of the beauty of digital recording was its wide dynamic range. Silly me! Apparently, the point of digital recording is to be as loud as humanly possible.
However, there is a way to put some serious average level on a CD without totally destroying the dynamics. It takes a little more work, but try this technique and see if it doesn't produce a result that's ultimately more satisfying than alternative methods.
So What's Wrong With Compression?
Electronic compression is useful, but comes with a price: breathing, pumping, increased noise, and transient mutilation. Add compression to a languid lead guitar track and it can sound very cool. But while a subtle amount of limiting can definitely help during the mastering process, over‑compression can drain the life out of the song. This is particularly the case with programme material, where compressing loud sections alters the softer elements in those sections as well. If you're looking for an ultra‑compressed sound, you're probably better off compressing each track individually and mixing them together, rather than compressing the final stereo master.
Multi‑band compression is much better, because it separates the signal into multiple frequency bands and compresses each one individually. This just about eliminates pumping and breathing because, for example, a heavy‑duty kick drum isn't going to affect the high frequencies in your mix. When compressing programme material, this is my technique of choice. Still, there's always a slight squashing that's hard to avoid, and is inappropriate for many types of material. And since any type of compression will bring up noise somewhat, it's worth remembering that multi‑band compressors have just that many more bands capable of adding noise.
Software‑based compressors (that usually run as plug‑ins for digital audio programs) are a completely different kettle of fish. Because they don't have to work in real‑time, they can analyse the signal and react to peaks instantaneously. This cleans up a lot of problems inherent in real‑time compression, but it's not a panacea. Pumping and other problems are still possible.
One reason why people like analogue tape so much is because it provides several benefits of compression without the side‑effects. Not only do signals gently saturate the tape, there's some added harmonic interest that helps overcome the apparent loss of high frequencies often noted with single‑band compressors. (This occurs because there is more energy in the low end than the high end of most programme material, so as the low end triggers compression to reduce the gain, this reduces the high end gain as well, even though it doesn't really need compression. Side‑chain processing can solve this problem, but that's another story for another time.) Despite this, tape has hiss, modulation noise, distortion, and all the other drawbacks that made people want digital in the first place. There must be a better way.
Why Normalisation Alone Isn't Enough
Normalisation is a digital signal processing function that's available in a lot of digital audio editing software (for example Sound Tools, Sound Forge, Alchemy, and so on). It scans through the programme material for the highest level, and if that level doesn't reach the maximum available dynamic range, the software boosts the overall signal so that the peak hits the highest level possible. For example, suppose you record a track of music and the highest peak registers at 6dB below the maximum available headroom. Normalisation brings the entire track up by 6dB. (Incidentally, most normalisation functions allow normalising to some percentage of the maximum available level; it needn't always be 100%.) There are some problems, though:
- Because normalisation boosts the entire signal, the noise floor comes up as well.
- Normalisation has nothing to do with a song's average level, only the peak level. Yet when balancing levels between tracks in the process of assembling a master tape, it is the average level that is usually most important. This is one reason why most mastering engineers recommend that you do not normalise each individual song.
- Excessive use of amplitude‑changing audio processes such as normalisation on linear, non‑floating‑point digital systems can cause so‑called 'round‑off errors' that, if allowed to accumulate, impart a 'fuzzy' quality to your sound. If you're going to normalise, it should be the very last process — don't normalise, then add EQ, then change the overall level, and then re‑normalise, for example.
Despite these potential problems for the unwary, in many cases normalisation can indeed help put the highest possible peak level onto CD. But this won't help very much if the average level of a track is relatively low, yet there are one or two major peaks that hit the maximum available dynamic range. Here's the answer.
The Level/Normalisation Connection
If you look at a typical two‑channel mix, you'll often see a few peaks that are considerably higher than the average signal level (see the highlighted section in Figure 1). When you normalise, these bump up against the maximum available headroom and essentially set a limit on how high the rest of the signal can be.
Bringing down the level of those few peaks prior to normalisation can increase the overall signal level a lot more. Here's how.
1. Identify the areas with the individual peaks, and work on one area at a time. Some software will even help locate peaks for you — a very polite thing to do, in my opinion, and it certainly makes the process go faster.
2. Highlight the peak that you want to cut down to size. The region containing the peak, which you're going to edit, should have its boundaries on zero crossings. In other words, the amplitude at each boundary should be 0.
3. Use the program's volume or scaling function to reduce the excessively loud peak, so that its level is in line with the other peaks. If the peak is only a half‑cycle wide, just process that peak. If it is a full cycle, scale both halves of the cycle simultaneously.
4. Perform the same process on other excessively loud peaks in the song.
5. Finally, normalise the entire song.
Wrapping Up
Another point to ponder: the idea of lowering peaks, rather than normalising the entire song, is a very 'analogue tape' sort of concept. When mixing down to analogue, if you hit the tape with lots of level, two things happen: the tape saturation simply 'absorbs' spikes like the ones shown in Figure 1, and the lower‑level signals seem pretty loud, because you're feeding the tape with a hot signal. Clipping the spike does cause distortion, but if it's only for a few milliseconds, your ear isn't going to notice it, especially since tape distortion is so 'soft'.
Granted, it takes more work to seek out and tame individual cycles using this method than it does to just set a compressor's in/out switch to 'in', but the results are worth it. Try it, and I think you'll agree.