September 1, 2014

How to use Parallel Compression for unparalleled results in your audio mixes

Parallel Compression is one of the oldest tricks in the book of mixing sound.  It is used when processing individual instruments, drum and vocal buses, and on entire audio mixes.  Many mastering engineers also use parallel compression in the final stage of mastering in order to maintain dynamics while tightening up a mix for commercial playback and volume standards.  (Many mastering engineers also use multiband compression to fine-tune a final mix, which you can read about in this post.

The concept is simple: instead of applying compression to 100% of an incoming signal, you first split the incoming signal into two tracks.  One passes through to your output untouched, while the other first passes through a compressor before reaching the output.  These two signals are then recombined at the output stage at varying volume levels, depending on the blend the audio engineer finds optimal.  Latency adjustments may need to be made to ensure the recombination of signals does not lead to comb filtering or unnecessary noise cancelling.

parallel compression, paralell compressor, how to compress in parallel, what is parallel compression, audio mixing, audio compressor, side chain compressor, how to use a side chain
Parallel Compression blends raw and compressed signals into one

The idea behind parallel compression is that one can achieve the best of both worlds: the dynamics of the raw, unprocessed signal with the volume-equalizing, soft-noise-enhancement of compression.  The exact blend is in the eyes, or ears, of the beholder.  Personally, I usually like weighing more heavily on the processed, compressed signal (60-90%) and adding the raw signal to taste.  There is no right or wrong answer though, and it largely depends on the individual qualities of the instrument(s) or mix you are working on.

Before we get into the finer details of parallel compression, let's quickly review the basics of how an average compressor works.

How normal "downward" compression works

Normally when we talk about the use of compressors, we're talking about Downward Compression.  This refers to the reduction of volume of loud signals passing through the compressor.  The compression ratio used is the primary factor that determines just how much volume reduction occurs beyond the threshold.  Let's define these terms.

The compression Threshold specifies the level above which the compressor starts working (or more specifically, where gain reduction commences).  Signals falling below the threshold are not compressed or reduced in volume.  Some compressors that come with a dedicated threshold knob have what we call variable threshold settings, while other compressors (such as the popular Universal Audio 1176LN) have fixed thresholds that require additional input gain to overshoot the threshold.  These fixed-threshold compression units lack a dedicated threshold knob, but a fine balance can still be achieved by proper setting of the output gain knob in relation to the input gain settings.

how audio compressors work, audio compressor, how to use audio compressor, ua 1176, audio compression, mixing sound
How Audio Compressors Work to Reduce Gain

Input Gain allows an engineer to boost the incoming signal, making it easier for the signal to cross the specified threshold while increasing the volume of the quieter parts of the signal.  This is part of how one achieves a tighter dynamic range, but one must be careful not to overdo it, because low-level noise will also be raised.  This is where it pays to have good microphones, preamps, and a recording studio with proper acoustic treatment.  By minimizing extraneous noise at the source, more input gain can potentially be applied later during mixing and mastering without raising the level of offending frequencies to the point of being a nuisance.

The Ratio specifies just how much the compressor will clamp down on your signal once it crosses the threshold.  A higher ratio means more clamping down, and the signals over the threshold will be reduced more.  At a 1:1 ratio, no compression will occur at all.  At a 2:1 ratio, for every 2 dB a signal crosses over the threshold, the compressor will scale it down to just a 1 dB increase.  Similarly, at a 10:1 ratio, for every 10 dB a signal crosses over the threshold, it will only overshoot the threshold by 1 dB.

Hard Limiting occurs at a compression ratio of Infinity:1, where the compressor and/or limiter essentially "limits" the signal from crossing the threshold level, if ever so slightly.  Hence the word "limiter," or "hard limiting."  Many compressors can be used as limiters by simply cranking their ratios as high as you can, such as 500:1.

Attack defines how quickly gain reduction occurs and is usually set in the neighborhood of 0.010 ms (very fast) to 300 ms (slower).  Let's say you have your compressor attack set at 20 ms.  Many people mistakenly think this means nothing is happen during the period of 0 - 19.99 ms until reaching the 20 ms point, and believe that right at 20 ms the compressor suddenly starts attenuating the signal.  Rather, the compressor starts attenuating the signal as soon as it crosses the threshold (at ~0 ms), but does so over the course of a gradual curve that eventually arrives at the full specified attenuation amount (specified by the ratio) by the time the attack time is reached.

Attack times can be dialed in very specifically to preserve transients (which are necessarily different for different instruments) before noise attenuation occurs.  On drums, for example, an attack time might be set to preserve the initial "hit," but reach full attenuation by the time the instrument is "ringing out."  This strategy might be employed to retain the percussive nature of the toms in a busy mix while simultaneously minimizing/controlling the forthcoming boom to prevent the buildup of low frequency mud in an otherwise well-balanced mix.

Release controls how quickly gain reduction rises and is usually set between 10 ms - 2500 ms.  Slower release times lend themselves to more balanced, volume-equalized signals by not recovering too fast.  When using heavy compression, faster release times can produce "pumping" and "breathing" as a result of the  compression recovering too quickly.  When a well-attenuated signal rises back to full level quickly, it is often very perceptible.  While some engineers use this for effect, others strive to avoid it due to its noticeable and occasionally annoying nature.

Like with attack times, release times can be set strategically to control exactly how long a certain instrument or bus mix is "squashed" before returning to normal.  Depending on the exact compressor you are using, settings will need to be dialed in to taste due to different timing laws, which specify the math behind the gradual curve taking place in the attack and release stages.  Some are linear, which tend to be more noticeable, and others are exponential, which are more transparent and gradual.

compressor attack and release, attack and release, compressor attack, compressor release, attack settings, release settings, audio mix
Linear vs. Exponential Attack and Release Time Curves

 Now let's have one more look at Attack, Release, and time staging before moving on.

Attack and Release function independent of Threshold (huh?!)

Important Point: Many people mistakenly believe release only happens when a signal crosses back below the threshold.  However, this is not the case!  Attack and release both act on gain reduction independent of threshold, because they are part of the time stage of the compressor, which is fed by the scale stage (ratio).  It is the scale stage that is linked to threshold, and only after the scale stage is triggered to apply gain reduction (by a signal crossing the threshold) do the timing laws come into place -- by slowing down the speed with which that reduction takes place.

Example: The best way to illustrate this is with a signal that is already over the threshold, which rises even higher over the threshold, and then back to a lower point still over the threshold.  Say we have a sample signal that starts out 4 dB over the threshold and then jumps to 6 dB over the threshold, and finally back to +4 dB over again.  Imagine our ratio is set at 2:1.  Initially, our compressor will reduce the initial +4 dB to +2 dB, effectively subtracting 2 dB from the volume.  But when the signal spikes to +6 dB, that same 2:1 ratio will subtract a total 3 dB from the volume, leaving the original +6 dB signal at just +3 dB.  When the input signal drops back to +4 dB again, the compressor will momentarily still be applying a -3 dB reduction from processing the +6 dB signal, bringing our now +4 dB input signal to +1 dB.  But since we are at a 2:1 ratio, we know this +4 dB signal should really be at +2 dB, and it's the release time that specifies how long it takes for the signal to return to that level.

compressor attack release, audio compressor, compression attack, compression release, compression threshold, compression ratio
Attack and Release operate independently of Threshold

Remember, unless you are setting your compressor for hard limiting, your signal will often remain above the threshold as it fluctuates, and the release will still be affecting gain reduction as this happens.  

Back to Parallel Compression (and Side-Chaining)

Straight downward compression, where the entire signal is run through a compressor, can often lead to an unnatural sound, "compression artifacts," and other undesirables that we seek to minimize.  It can be frustrating trying to produce a rich, full, and even sound that includes the subtle nuances of one's playing without squashing a mix too hard, creating pumping and breathing, or otherwise losing the life or dynamics of the sound.  This is where parallel compression comes in handy, or even side-chaining at times (or both!).

Side-chaining is sometimes used to get around this, or generally to affect the compressor in a different manner, by feeding the compressor an external signal by which it is instructed to fire.  In other words, instead of the gain levels from the input signal controlling the compression, the compressor applies to the input signal based on a completely different signal's level as it crosses the threshold.

I've used this technique often to trigger a guitar bus compressor to lightly reduce the overall volume of the guitars in a mix whenever vocals are present, by feeding my vocal bus to the side-chain of my guitar bus compressor.  By setting up the compressor to apply light and smooth gain reduction of a mere 1.5 - 3 dB, space can be created for the vocals to come through more clearly without noticeably affecting the levels of the guitars.  This is much faster than automating volume fader levels on the guitar bus, and helps avoid inter-mix volume wars whereby you might be inclined to keep turning everything up louder until your mix is a complete mess.

Speaking of vocals, this is often the first place that I will use parallel compression.  Vocals have such a crazy dynamic range that they can easily stick out in the mix like a sore thumb, only to be completely lost and buried just seconds later in the same phrase.  Heavy compression with ample input gain is often needed to get the vocals to sit in a dynamic "pocket" where the quietest nuances can still be heard, while the loudest belt-it-out choruses do not overwhelm the mix.  In fact, many engineers I know rely on stacking multiple compressors together in series to deal with this problem, making strategic settings to threshold, attack, and release on each one to achieve this ideal balance.  I've seen guys like Matt Murphy use up to 4-5 compressors on a single vocal before!

When performing this kind of operation, it can be inevitable that some of the life and vibrance of the original signal will be lost.  Unless, of course, the original unprocessed signal is blended back into the mix!  The way I usually achieve this is by first getting my heavily compressed vocal to a point where it sits in the pocket I want it to sit in, then bringing up the fader of the raw track until I can hear it adding some dynamic life and transient character back to the peaks without being obstructive.  Mentally I think of it sort of like dropping an "anchor" with my compressed track, keeping things firmly rooted in place, then moving about that anchor with a more controlled sense of volume.  The compressed track ensures details are heard and the vocal never gets buried, and the raw track maintains the vibrance, life, and range of the singer.

Parallel Compression Strategies

There are many ways to approach parallel compression, and none of them is necessarily "correct."  One common method is to send the original track to a bus, and then use that bus input to feed aux input tracks (one processed, one not).  This way, any changes made to the original track, such as volume, EQ, etc., affect both tracks and not just one.  However, for stylistic purposes one could play around with duplicating the original track and processing each one differently for varying effects -- the possibilities are endless.

Similarly, many people will squash their compressed track pretty hard when parallel compressing because they can.  They know the raw track will preserve some of the life and transient nature, so they can get away with faster attack times and slower release times.  On the other hand, more subtle results can be achieved by using more moderate threshold, attack, and release settings on the track processed in parallel.

Parallel methods can even be applied to a mix bus, which we discuss in this article.  It's a bigger beast to tame, but with the right setup, livliness can be maintained while controlling the overall mix.  Note also that for those who really want to get scientific with it, parallel compression does not need to be limited to just two tracks.  One raw track could be blended with two or more compressed tracks, each processed differently before the input stage and with different compression settings.  Just realizing that weaving this type of web is often unnecessary, and will mean making changes to tons of settings, tracks, and plugins should any one of your variables need to change at some point.

Credits/References: Many thanks to Roey Izhaki (Mixing Audio),, Sound on Sound, Matt Murphy, and the many internet articles I've scoured for helping me acquire and utilize this knowledge.


Post a Comment