Two compressors in series, each working 2 to 3 dB, control a lead vocal more cleanly than one compressor working 8. That is the whole argument, and it is not a matter of taste. A single unit asked to do everything needs a ratio and a release fast enough to catch consonants, and those same settings then chew on the sustained notes it should have left alone. Split the job instead: a fast FET stage catches the peaks it was designed for, a slow optical stage rides the phrase, and neither is pushed into the range where it starts to sound like a compressor.
What does each compressor in the chain actually do?
The two stages are different machines solving different problems.
The FET stage is the peak catcher. The circuit this whole category copies is the 1176, whose attack time runs from 20 to 800 microseconds, fast enough to act on a single consonant. Set to a 4:1 ratio with the attack around 3 on the dial and the release fast, it takes 2 to 3 dB off the hardest syllables and does nothing the rest of the time. The meter should sit at zero between phrases and flick on the peaks.
The optical stage is the leveller. An LA-2A style unit built around the T4 optical cell has a release in two stages, roughly 60 milliseconds for the first half of the recovery and anywhere up to several seconds for the rest, and that behaviour is exactly why it sounds like a hand on a fader rather than a circuit. It cannot catch a consonant, and in this chain it never has to. It sees a signal whose peaks are already handled and rides the phrase by another 2 to 3 dB.
Run this way, total gain reduction across the chain is 4 to 6 dB, but no single stage ever works hard enough to pump, and that is the difference the listener hears.
Which order should they go in?
FET first, optical second. The reasoning is mechanical rather than aesthetic. The optical cell's response depends on how hard and how long it is hit, so a raw vocal with 6 dB transient spikes keeps startling it into its longer release phase, and the level starts to breathe. Put the FET in front and the leveller receives a signal with the spikes already shaved, so it does the one thing it is good at, slow level riding, without being provoked.
The reverse order, leveller first, is not wrong, it is a different tool: it turns the FET into an effect stage that you drive for tone. That is a colour decision for a record that wants audible compression. For a vocal that has to sit in front of a dense arrangement and stay intelligible, peaks first, level second.
How much gain reduction is too much?
Numbers I actually hold to. FET stage: 2 to 3 dB on peaks, never more than 5, ratio 4:1. Optical stage: 2 to 3 dB on the loudest phrases, needle mostly between 0 and 3. If either meter is living past 5 dB, the problem is upstream, and the fix is clip gain: level the phrases by hand in the DAW first, so the compressors process a vocal that is already within 6 dB of itself. Ten minutes of clip gain saves both compressors from working as repairmen, which is the job they are worst at.
The check is always the same and always cheap: bypass both units and match the loudness. If the compressed vocal sounds smaller than the bypassed one at the same level, the chain is overworking, and the first thing to back off is the FET release.
What changes when the vocal has to survive broadcast?
A contest or television performance adds a processing chain nobody in the studio controls. European broadcast runs to EBU R128 at -23 LUFS, and the transmission path applies its own limiting on top of whatever the mix already did. A vocal squeezed to death in the mix gets squeezed again on air and arrives flat. Working on contest versions for Eurovision entries taught me to leave the dynamics budget deliberately underspent: peaks controlled, level ridden gently, and 2 to 3 dB of natural movement left in the phrase for the broadcast chain to eat. The mix that sounds slightly underdone in the control room is the one that survives the air chain. The full list of records behind that observation is on the [credits page](/credits.html), and the roles are documented per release.
When is one compressor enough?
A spoken word or voiceover session with a controlled performer often needs only the optical stage, because there are no transient spikes worth catching. A screamed rock vocal sometimes needs only the FET, because the performance has no quiet phrases to ride up. And a vocal recorded by a singer with real microphone technique, working the distance on loud notes, can occasionally need nothing but clip gain. The two-stage chain is the default for a produced pop lead, where the arrangement is dense, the dynamic range of the raw take is 10 dB or more, and the vocal has to hold the front of the mix for three minutes without a single syllable disappearing.
What to do with this
Print the chain as a starting point on the next vocal you mix: clip gain first, FET at 4:1 taking 2 to 3 dB on peaks, optical riding 2 to 3 dB behind it, then loudness-matched bypass as the test. If the bypassed vocal wins, halve the gain reduction and test again. More on how I work is on the [about page](/about.html), and the [credits](/credits.html) list the records these defaults come from.