Quieter Recordings - One dB at a Time
By Anderton |
Sometimes Little Improvements Add Up To Big Improvements
By Craig Anderton
The whole is equal to the sum of its parts…as anyone who ever used analog tape will attest. Who can forget that feeling of hearing yet another contribution to the noise floor whenever you brought up a fader, as one more track of tape hiss worked its way to the output?
With digital recording, tape hiss isn’t an issue any more. But our standards are now more stringent, too. We expect 24-bit resolution, and noise floors that hit theoretical minimums. As a result, every little extra dB of noise, distortion, or coloration adds up, especially if you’re into using lots of tracks. A cheapo mic pre’s hiss might not make a big difference if it’s used only to capture a track of the lead singer in the punk band Snot Puppies of Doom, but if you’re using it to record twelve tracks of acoustic instruments, you will hear a difference.
I’ve often stated that all the matters in music is the emotional impact, but still, it’s even better when that emotional impact is married with pristine sound quality. So, let’s get out the "audio magnifying glass" (even though they don’t work for mixing, headphones are great when you need to really pay attention to details on a track), and clean up our tracks … one dB at a time.
PREVENTING THE NOISE PROBLEM
Even in today’s digital world, there’s hiss from converters, guitar amps, preamps, direct boxes, instrument outputs, and more. The individual contribution in one track may not be much, but when low level signals aren’t masked by noise, you’ll hear a much more "open" sound and improved soundstage. (And if you don’t think extremely low levels of noise make that much of a difference, consider dithering—it’s very low level, but has a significant effect on our perception of sound.)
The first way to reduce noise is prevention. Maybe it’s worth spending the bucks on a better mic pre if it’s going to shave a few dB off your noise figure. And what about your direct box? If it’s active, it might be time for an upgrade there as well. If it’s not active but transformer-based instead, then that’s an issue in itself as the transformer may pick up hum (first line of defense: re-orient it).
Here are some additional tips:
- Gain-staging (the process of setting levels as a signal travels from one stage to the next stage, so that one stage neither overloads the next stage, nor feeds it too little signal) is vital to minimizing noise, as you want to send the maximum level short of distortion to the next stage. But be careful. Personally, I’d rather lose a few dB of noise figure than experience distortion caused by an unintentional overload.
- Crackles can be even more problematic than hiss. Use contact cleaner on your patch cord plugs, jack contacts, and controls. Tiny crackles can be masked during the recording process by everything else that’s making noises, but may show up under scrutiny during playback. In a worst-case situation, the surfaces of dissimilar metals may have actually started to crystallize. Not only can that generate noise, but these crystals are all potential miniature crystal radios, which can turn RFI into audio that gets pumped audio into the connection. Not good.
- Make sure any unnecessary mixer channels are muted when you record. Every unmuted channel is another potential source of noise.
- Unless you have a high-end sound card like the Lynx line, avoid sending any analog signals into your computer. Use digital I/O and a separate, remote converter.
- Although most people use LCD monitors these days, if there's a CRT on while you’re recording, don’t forget that it’s pumping out a high frequency signal (around 15kHz). This can get into your mics. Turn it off while recording.
- When recording electric guitar, pickups are prone to picking up hum and other interference. Try various guitar positions until you find the one that generates the minimum amount of noise. If you have a Line 6 Variax, consider yourself fortunate —it won’t pick up hum due to using a piezo pickup.
No matter how hard you try, though, some noise is going to make it into your recorded tracks. That’s when it’s time to bring out the heavy artillery: noise removal, noise gating, and noise reduction.
DEALING WITH NOISE AFTER THE FACT
With a typical hard disk-based DAW, you have three main ways to get rid of constant noise (hiss and some types of hum): noise gating, noise removal, and noise reduction.
Noise gating is the crudest method of removing noise. As a refresher, a noise gate has a particular threshold level. Signals above this level pass through unimpeded to the gate out. Signals below this threshold (e.g., hiss, low level hum, etc.) cause the gate to switch off, so it doesn’t pass any audio and mutes the output.
Early noise gates were subject to a variety of problems, like "chattering" (i.e., as a signal decayed, its output level would criss-cross over the threshold, thus switching the gate on and off rapidly). Newer gates (Fig. 1) have controls that can specify attack time so that the gate ramps up instead of slamming on, decay time controls so the gate shuts off more smoothly, and a "look-ahead" function so you can set a bit of attack time yet not cut off initial transients.
Fig. 1: The Gate section of Cubase’s VST Dynamics module (the compressor is toward the right) includes all traditional functions, but also offers gating based on frequency so that only particular frequencies open the gate. This makes it useful as a special effect as well as for reducing noise. In this case, the kick is being isolated and gated.
Noise gates are effective with very low level signals and tracks with defined "blocks" of sound with noise inbetween, but the noise remains when signal is present—it’s just masked. (For more about noise gates, check out the article "Noise Gates Don't Have to Be Boring.")
Manual noise removal is the manual version of noise gating (Fig. 2). It’s a far more tedious process, but can lead to better results with "problem" material.
Fig. 2: The upper vocal track (shown in Cakewalk Sonar) has had the noise between phrases removed manually, with fades added; the lower track hasn't been processed yet.
With noise removal, you cut the quiet spaces between the audio you want to keep, adding fades as desired to fade in or out of the silence, thus making any transitions less noticeable. However, doing this for all the tracks in a tune can be pretty time-consuming; in most cases, noise gating will do an equally satisfactory job.
Noise reduction subtracts the noise from a track, rather than simply masking it. Because noise reduction is a complex process, you’ll usually need to use a stand-alone application like Adobe Audition (Fig. 3), Steinberg Wavelab, Sony Sound Forge, iZotope RX2, and various Waves plug-ins.
Fig. 3: Sound Forge's Noise Reduction tools have been around for years, but remain both effective and easy to use.
With stand-alone programs, you’ll likely have to export the track in your DAW as a separate audio file, process it in the noise reduction program, then import it back into your project. Also, you'll generally need a sample of the noise you’re trying to remove (called a "noise print," in the same sense as a fingerprint). It need only be a few hundred milliseconds, but should consist solely of the signal you’re trying to remove, and nothing else. Once you have this sample, the program can mathematically subtract it from the waveform, thus leaving a de-noised waveform.
However, some noise reduction algorithms don’t need a noise print; instead, they use filtering to remove high frequencies when only hiss is present. This is related to how a noise gate works, except that it’s a more evolved way to remove noise as (hopefully) only the frequencies containing noise are affected.
"Surgical" removal makes it possible to remove specific artifacts, like a finger squeak on a guitar string, or a cough in the middle of a live performance. The main way to do this is with a spectral view that shows not only amplitude and time, but also, frequency. This makes it easy to pick out something like a squeak or cough from the music, then remove it (Fig. 4).
Fig. 4: Adobe Audition's spectral view and "Spot Healing Brush Tool" makes it easy to remove extraneous sounds. Here, a cough has been isolated and selected for removal. Audition does elaborate background copying and crossfading to "fill in" the space caused by the removal.
While this all sounds good in theory—and 90\% of the time, it’s good in practice too—there are a few cautions.
- Noise reduction works best on signals that don’t have a lot of noise. Trying to take out large chunks of noise will inevitably remove some of the audio you want to keep.
- Use the minimum amount of noise reduction needed to achieve the desired result. 6 to 10dB is usually pretty safe. Larger values may work, but this may also add some artifacts to the audio. Let your ears be the judge; like distortion, I find audible artifacts more objectionable than a little bit of noise.
- You can sometimes save presets of particular noise prints, for example, of a preamp you always use. This lets you apply noise reduction to signals even if you can’t find a section with noise only.
- In some cases you may obtain better results by running the noise reduction twice with light noise removal rather than once with more extensive removal.
So is all this effort worth it? I think you’ll be pretty surprised when you hear what happens to a mix when the noise contributed by each track is gone. Granted, it’s not the biggest difference in the world, and we’re talking about something that happens at a very low level. But minimizing even low-level noise can lead to a major improvement to the final sound … like removing the dust from a fine piece of art.
Craig Anderton is Executive Editor of Electronic Musician magazine. He has played on, mixed, or produced over 20 major label releases (as well as mastered over a hundred tracks for various musicians), and written over a thousand articles for magazines like Guitar Player, Keyboard, Sound on Sound (UK), and Sound + Recording (Germany). He has also lectured on technology and the arts in 38 states, 10 countries, and three languages.