De-esser latency: what 5 ms of look-ahead actually buys you

"Look-ahead" is one of those features that sounds like something you'd always want more of. It lets a de-esser catch an ess before it peaks — magic, right? — so the instinct is to crank it. We'd gently push back on that instinct. Every millisecond of look-ahead is a millisecond of latency you pay everywhere, all session long. The real question isn't "how much can I get." It's "how little do I actually need."

What look-ahead is doing

An ess has a shape. It ramps up over a few milliseconds before it hits its loudest point. A de-esser with no look-ahead only reacts after the level crosses the threshold — so it's always a hair late, and the very front of the ess slips through before the gain reduction catches up. You hear it as a little spit on the front of the word.

Look-ahead fixes that by delaying the audio a touch and letting the detector see the onset coming. By the time the ess arrives, the reduction is already sitting there waiting for it. Clean catch instead of a clipped transient.

The part nobody mentions: look-ahead is latency

That delay isn't free, and this is where the "more is better" instinct falls over. Whatever you set as look-ahead becomes the plugin's reported latency — and that has knock-on effects:

• Mixing, your DAW compensates by nudging every other track back to match. Fine in principle, but it stacks up across a chain, and a sluggish round-trip makes comping and editing feel a half-step removed from your hands.
• Tracking or performing, there's nothing to compensate against. The latency just becomes the delay the singer hears in their own voice. A few milliseconds is livable. Past ~10 ms it's genuinely distracting, and nobody can perform through it.

So look-ahead trades a cleaner transient for a slower plugin. The craft is spending just enough to get ahead of the onset — and not a millisecond more.

How little you actually need

Here's the bit the big-number instinct gets wrong: a sibilant onset is fast. You don't need 15 or 20 ms to see it coming. A few milliseconds already puts you ahead of the transient. Everything past that is latency you're paying for nothing, because the detector saw what it needed long ago and is now just… waiting, making you wait too.

Which is why Sibilance ships two profiles instead of one heroic slider:

• Studio — around 5 ms. Enough to catch the onset cleanly, low enough that the round-trip stays comfortable while you mix.
• Live — around 1 ms. Basically no latency, for tracking and monitoring — trading a sliver of onset precision for a plugin you can actually sing through.

Five milliseconds catches the ess. Fifteen mostly just makes you wait for it.

One more latency trap: linear-phase

While we're here — some de-essers offer a linear-phase mode, and it sounds like the responsible choice. Perfect phase, no smearing of the relationships between frequencies. But linear-phase filtering adds real latency and pre-ringing — a faint smear that arrives before the transient it belongs to. For a tool whose whole job is chasing fast consonants, ringing ahead of the sound is exactly the wrong artifact.

So we went the other way, on purpose. Sibilance is minimum-phase: low latency, no pre-ringing, honest transients. The phase shift that introduces is inaudible on the narrow bands a de-esser works in — and you keep every bit of the speed.

The takeaway

Look-ahead is a tool, not a virtue. Use enough to get ahead of the onset — a few milliseconds — and stop. Mixing? A studio profile around 5 ms is plenty. Tracking? Drop to near-zero and keep the singer in the moment. Longer isn't more accurate. It's just later. If you want to see how this shakes out against the reference de-esser, we put the two side by side, latency and all.

Sibilance is the single-authority de-esser this comes from — a real free tier, and a 7-day Pro trial with no card.