You have a vocal take that is mostly excellent, and a frequency somewhere in the low mids that blooms on certain words. You sweep a narrow boost, find it at 340 Hz, flip the boost to a cut — and the ring goes away along with a good deal of the singer.
That trade is not your EQ being bad at its job. It is a mismatch between the shape of the problem and the shape of the tool. This article is about why, and what actually fits.
The short version. Room resonance is intermittent, so a permanent cut is the wrong answer. You want attenuation that appears when the peak appears and leaves when it does. ENZYME does that per frequency bin, per frame — 14-day full trial, no account needed.
A room has modes: frequencies at which the distance between two surfaces matches a half-wavelength, so energy at that frequency reinforces itself instead of dissipating. A small room has its strongest modes in the low mids, which is exactly where the body of a voice lives.
A mode does not ring on its own. It rings when something excites it. If your room has a mode at 340 Hz, it blooms on the notes whose fundamental or low harmonics land near 340 Hz, and it stays quiet on the rest. In a three-minute take that might be six or seven words.
So the problem has a shape: narrow in frequency, intermittent in time. Hold on to that, because every tool below either matches that shape or does not.
A notch filter is narrow in frequency and permanent in time. It matches one half of the problem. For the six or seven words where the room blooms, it is exactly right. For the rest of the take it is removing energy that was never the problem — every other note whose harmonics pass through 340 Hz on the way past.
That is the thinning you hear. You did not cut too much at 340 Hz; you cut for too long. And the deeper you make the notch to handle the worst word, the more the other three minutes pay for it.
A useful test. Set your notch to fix the worst word, then solo a section with no resonance and toggle the EQ. If that section changes audibly, the notch is doing work it should not be doing.
A dynamic EQ band fixes the time half. Its gain follows a detector, so it cuts when there is something to cut and gets out of the way when there is not. On a single, stable, well-located resonance this is a genuinely good answer, and if you own a dynamic EQ you should reach for it before anything else.
Two things it still cannot do:
Which is why the practical workflow with dynamic EQ is to sweep, find the loudest mode, place a band, and then repeat for the second one if you have the patience. It works. It is just a lot of manual search for something a detector can do continuously.
Instead of choosing a frequency, choose a rule: attenuate anything that sticks up above the spectrum immediately around it, wherever and whenever that happens.
Concretely, the signal is transformed into short overlapping frames. In each frame, for each frequency bin, a local baseline is estimated — a running mean, in the power domain, of the bins around that bin. The excess over that baseline is what counts as resonance. Attenuation is applied per bin in proportion to that excess.
The consequences are the interesting part:
This is what ENZYME is. The manual covers every control; the rest of this article is the part you need to get a result today.
Make Delta a habit, not a rescue. Set it by ear, then listen to the difference and ask whether you would miss any of it. That one question catches most over-processing before it reaches a mix.
On a bus or a doubled take, the ring is often centred while the sides are fine. Detecting in mid/side rather than left/right puts the cut on the component that has the problem. Keep the two channels linked to one gain decision on bus material so a cut cannot pull the image sideways; unlink only when the problem genuinely lives on one side, such as a mode that one overhead hears and the other does not.
Being clear about the boundary saves you an hour of trying:
ENZYME — dynamic resonance suppression per frequency bin, per frame. VST3, CLAP and a native AUv2 for Logic and GarageBand. 14-day full trial, no account and no email required.
Because the notch is always on. A room resonance is excited only by the notes whose harmonics land on the room mode, so it might be present in two seconds of a three-minute take. A static notch removes that frequency for the entire performance, including every other note whose harmonics pass through it. You solve two seconds of ring by thinning three minutes of vocal.
It is closer, but not the same. A dynamic EQ band moves its gain over time, which fixes the always-on problem. It still has a fixed centre frequency and a fixed width, so it only helps if the resonance sits exactly where you put the band. A resonance suppressor evaluates every frequency bin independently and cuts wherever a peak protrudes, so it follows a resonance that moves and catches the second and third ones you did not find by sweeping.
The usual method is to sweep a narrow boost until the ring jumps out, then cut there. It works, but it biases you towards the single loudest mode and it is slow. If you are using a per-bin suppressor you can skip the search entirely: set the detector to only accept sharp peaks, and let it find every protrusion on its own.
It should not, and if it does you are cutting too broadly. The test is to monitor the difference signal — listen to only what is being removed. You want to hear ring and almost nothing else. If you can hear words, pitch or body in that signal, the cut is reaching into the performance.
You can usually fix it. Resonance is additive energy at specific frequencies at specific moments, and that is a tractable thing to attenuate. What you cannot recover afterwards is a genuinely smeared transient or a comb-filtered take from a reflection arriving very close behind the direct sound — that is a room treatment and mic placement problem, not a plug-in problem.
Yes. The detector does not know or care what made the peak. The same approach handles boxy notes on an acoustic guitar, sympathetic ring on a snare, filter self-oscillation on a synth, and narrow sibilance peaks that a broadband de-esser would over-treat.