By Craig Anderton
We all know the usual litany of ways to improve your sound: "Buy a better mic," "buy a better preamp," "buy a better..." Yeah, you get the idea. (My personal favorite is "write a better song," but that's beyond the scope of this article.)
However, there are a lot of simple fixes you to do to improve your sound - sometimes dramatically - that involve little or no money. And even if your ship has come in, these tips are still well worth following if you want to bring your recorded sound to a higher level.
I'm not entirely convinced this will result in an audible difference, but some people swear they can hear an improvement with sample rates above 44.1kHz. The question is whether the improvement is worth the extra storage space (and computer horsepower) - and if so, which sample rate to use. Few think that 176.4 or 192kHz is worth the effort not just because of issues like reduced track count, but also because there may be other issues, such as plug-ins not being designed to work at those rates.
While the high sample rate buzzword is 96kHz, sample rate converting down to the 44.1kHz of a Red Book CD requires some fancy math - your 96kHz signal gets divided by 2.176870748299319727891156462585. Sure, today's sample rate converters should be able to handle the calculations without roundoff errors, but you might get better results by recording at 88.2kHz (Fig. 1).
Fig. 1: MOTU's Digital Performer is one of many programs that's happy to work with an 88.2kHz sample rate.
88.2kHz provides virtually all the practical benefits of going with 96kHz, and some people think material recorded at 88.2kHz sounds better than material recorded at 96kHz by the time it ends up on a CD. In any event, give 88.2kHz a shot; if it sounds better to your ears, go for it.
We mean "24" as in bit resolution for recording. If you're still recording with 16 bits, flick that bit resolution switch in your host program now. Yes, your files will take up more space. But storage continues to get less expensive, and computers are fast enough to process this extra data without giving too much of a hit to your track count and ability to use plug-ins. 24 bits provides more headroom while recording, better dynamic range, and more "footroom" as well. Besides, with most converters, to obtain an "honest" 16 bits of resolution you need at least 20-bit converters anyway.
Oxidized contacts can definitely affect sound quality in a couple of ways. One is that resistance can build up, which is equivalent to putting a resistor in series with your cable. A more insidious problem is when crystallization builds up, and those little diodes act like little crystal radios, ready to detect RF and inject it into your system as low-level hash.
The more patch points and mechanical switches you can get rid of, the better. For those that remain, use contact cleaner like Caig's DeoxIT to keep your connections clean. This can make a big difference, especially if you have a lot of patching in your studio..
If you're piling on the tracks, doing complex mixes with a ton of automation, and want reverb tails to decay into nothingness, higher internal resolution can make a difference. For example, clicking on a check box within Cakewalk Sonar switches its audio engine over to 64-bit precision (Fig. 2).
Fig. 2: Here's the switch that turns on Cakewalk Sonar's 64-bit Double Precision audio engine.
As with high sample rates, higher audio engine resolutions are controversial as some people say there's no significant difference. Regardless, it's worth a try - this doesn't really stress out your computer, even if you're using a 32-bit operating system. If it sounds better...use it.
Note that this is not the same as the bit resolution used for recording; it's the resolution used when calculating levels, EQ, and other processes within host DAW software.
At the very least, resonances can be annoying. But even worse, you may mistake them as part of the sound coming out of your speakers. Bundle cables, caulk gaps, tighten down screws (and of course, turn off the snares on snare drums that aren't in use!).
Most digital meters, while more accurate than their inertia-ridden analog counterparts, measure the instantaneous level of the samples that make up the signal - not the actual signal level that results from interpolating those samples. So, it's entirely possible that the actual level is several dB higher than what the meter indicates, which means your signals could easily be going into clip-land occasionally without your knowing it. (Note that Fig. 3 shows an easy solution to seeing those clips: SSL offers a downloadable meter plug-in for Mac or Windows called X-ISM that indicates inter-sample clipping.)
Fig. 3: SSL's X-ISM plug-in for measuring inter-sample distortion. It works and it's free - what's not to like?
Granted, some will say "Use your ears; if you don't hear it, who cares?" But while you may not hear distortion on a single track, add together a bunch of mildly clipped tracks, and something may sound "wrong" - even if you can't identify the exact cause of the problem.
So, give your peaks a little breathing room and treat -6 as max. Besides, with 24-bit resolution, you're just throwing away an extra bit if you're recording 6dB lower. This won't make any significant difference in sound quality.
High clock speeds and dual core processors have pretty much put an end to the days of underpowered CPUs, but their legacy continues in many plug-ins that offer "high-quality" and "low-quality" options, with the latter placing less stress on your CPU. But you shelled out for that shiny new computer specifically to stress out your CPU, so seek out those "quality" switches (Fig. 4) and turn them all up to the max quality possible.
Fig. 4: In Propellerheads' Reason, disable the Low BW option in SubTractor and Dr. Rex. But with the NN-XT and NN-19, enable High Quality Interpolation.
This has been mentioned numerous times over the years, but just in case you missed it, use a sharp low-cut filter to roll off all unneeded bass frequencies (Fig. 5).
Fig. 5: Sonar's Quad EQ has a highpass filter whose slope can do up to 48dB/octave.
For example if the lowest fundamental in a track is 100Hz, start rolling off below that. Getting rid of unnecessary lows can help open up the sound of a mix.
They're great for keeping spit from your lead singer out of the mic during live performance, but they affect the high frequency response and just plain don't sound that good. If you don't mic real closely, your mic has a low frequency rolloff switch, and your singer doesn't get out of control, you may not even need a windscreen - try it. But if you do, get one of those round mesh models (Fig. 6) instead of using a foam "mic condom."
Fig. 6: Mesh pop filters cost more, but they'll protect your mic while preserving its sound quality.
We'll assume you've already placed your near-fields on stands, and made sure there aren't reflective surfaces between the speakers and your ears (e.g., desktops, mixing consoles, etc.). But you still may have problems because of sound coupling from the speakers to the stands, which then causes other surfaces to vibrate. Although you can buy decoupling pads, the cheapest solution is to gather together some of those thick, neoprene promotional mouse pads you never use anyway and put them between the speakers and stands. However, a far better and more effective solution is the Primacoustic Recoil Stabilizer line (Fig. 7).
Fig. 7: Primacoustic's Recoil Stabilizer delivers excellent decoupling at a reasonable price - it may seem like snake oil, but unlike something like high sample rates you really can hear a difference.
If there was a lot of coupling going on, decoupling the speakers will result in a more focused, tighter sound, with much more defined bass and clearer highs.
Not all EQ plug-ins sound the same. Run some signal sources through several different EQs at extreme, but identical, settings; for example, try boosting treble while processing crash cymbals, and determine which EQ gives the "sweetest" sound. Try boosting upper mids with vocals to find out which vocals get "harsher" and which ones simply get more present. Also experiment with cutting extreme amounts of mids to find out how various EQs hold up.
In a real acoustic space, reverb consists of millions of reflections. No matter how hard a reverb algorithm tries, it can only approximate that degree of complexity. Even convolution reverbs, while very realistic, cannot duplicate the sound of a real acoustic space - only simulate it.
One quick fix is to run two reverbs in parallel. For example, if you have a really good hall sound, run it in parallel with a plate sound (Fig. 8). Each reverb will tend to "fill in the cracks" in the other one's sound, producing a more complex and satisfying reverb effect.
Fig. 8: Running two reverbs in parallel or even in series can yield a much smoother, richer sound than relying on a single reverb. Here, IK's CSR Hall and CSR Plate are inserted in parallel FX buses within Steinberg's Cubase.
Also try combining reverbs in series; a lot depends on the types of reverbs you're using. At some point while you're experimenting, you'll likely find a perfect combination of the two. Save both presets, because you'll likely want to use them again.
It stands to reason that a $300 box isn't going to include a $1,000 D/A converter on board, but many effects do include a digital out - and with quality conversion, you can hear what a device really sounds like.
Of course, if you have a suitable digital audio interface, you can feed the effect's digital out directly to your computer. But sometimes, getting a little analog mojo into the signal chain - especially if it's high-quality analog mojo - can add a character to the sound you won't obtain by going digital-to-digital.
One common trick is to lower latency while recording, then kick it up to a higher value when mixing. This causes less stress on the CPU, thus allowing more plug-ins and virtual instruments to run, as well as providing the bandwidth to handle complex automation and other tasks.
However, some engineers swear that using more latency than you really need doesn't help the sound, because it causes buffer timing issues that have the same kind of effect as using a loose clock signal: Smearing of the sound, and narrowing of the soundstage. I don't know of any hard proof about this, but there's enough anecdotal evidence floating around that this concept deserves a closer look. Meanwhile, it's probably a good idea to use no more buffering than is really needed, even if you're mixing.
Different dithering algorithms are subtly different. But when you're dealing with something that's happening around the noise floor, it's hard to quantify exactly what's happening.
To judge how dithering affects the sound, record something acoustic with a long decay, like a decaying piano chord with the sustain pedal up. Next, cut just the end of each track (say, where the signal dips below -65dB or so), turn the volume way up while being very careful to make sure no high-level noises can get into the mix, then apply various types of dithering with different noise levels and noise shaping. Decide which one works best for you - assuming you actually need any, as with today's high resolution recording and computing options, dithering may do nothing more than add a layer of noise you don't really need.
Taken on a track-by-track basis, you may not hear any hiss in a project. But add together a bunch of tracks with low-level hiss, and it's not so low-level any more.
Fortunately, today's noise reduction algorithms (such as the noise reduction tools in Sony Sound Forge and iZotope RX3) do a superb job of minimizing noise while maintaining transparency of sound. The less noise they need to get rid of, the better the sound quality. So if you're using noise reduction to reduce noise that sits around, say, -65dB or so, you can bring the noise down to -80dB with virtually no audible degradation.
The key to good noise reduction is to take a "fingerprint" of only the noise. Often you can find this at the head of a track, or during silences in the middle. Subtract this from your audio using a noise reduction tool that supports this type of operation, and do this for all your tracks; you may be startled by the kind of clarity this imparts to the final mix - it's like removing a scrim in a theater production.
These last two options aren't low cost, but they're worth mentioning anyway if you're more concerned about pushing performance than pinching pennies.
I always knew that having properly conditioned power was good for your equipment, but never really believed it made an actual sonic difference until I reviewed the Equi=Tech balanced power system for EQ magazine several years ago. Taking residual noise measurements with and without the Equi=Tech revealed about a few dB less noise with the Equi=Tech in use. It's a relatively costly way to shave a couple dB, but every little bit helps - and good power filtering/conditioning helps promote happier, longer-lived gear anyway.
Digital Distortion: From Harsh to Creamy
Not all distortion is bad - just ask a guitarist. However, some digitally-generated distortion can have a harsh quality, even when it's not supposed to (such as amp simulation software). One "magic bullet" I've found for smoothing out sounds like power chords is the Declick and Decrackler processor in iZotope's RX3 (Fig. 9; Adobe Audition's Click and Pop remover also works). Seriously.
Fig. 9: Smooth out intentional digital distortion with iZotope's RX Declick and Decrackle processing.
Depending on how heavily you apply it, it smoothes out the spiky stuff. However, don't assume that more is better - sometimes a light amount of decrackling is all you need. RX2 has a handy "Output Clicks Only" button so you can hear only what's being removed.
Craig Anderton is Editor Emeritus of Harmony Central. 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.