Resonance on a real harsh vocal: what the tone tests missed

A lab note. Measured 14 November 2026 on the version of Resonance in development at the time.

We tuned Resonance on test signals: one tone over noise, at many levels and frequencies. On those, its cuts landed where they should. Then we ran it on a genuinely harsh real vocal, and it didn't.

The OneMove Resonance panel on iPad: the live spectrum with the cut curve drawn in green, the Depth knob, the band editor and Key Link
Resonance, the plug-in in this note, on its current version: the green curve on the spectrum is the cut.

The test

  • Vocal: a dry, bright, shouted chorus, 30 seconds at 48 kHz ("Stayin' at Home – Vocals" by mindmapthat, CC BY 3.0).
  • Method: each render is compared with the dry vocal, band by band: how much energy each band lost over the whole excerpt. Black-box: audio in, audio out.

What we found

render200–500 Hz1–2 kHz2–3 kHz3–4 kHz7–10 kHzspread
Resonance, default, Depth 5−0.9−1.0−1.0−1.1−1.10.3 dB
Resonance, default, Depth 10−2.4−2.4−2.4−2.4−2.40.6 dB
Resonance, Harsh Vocal preset, Depth 10−2.6−3.8−3.4−3.6−3.71.2 dB
Energy each render took out of the harsh vocal, by band 30 s shouted chorus, compared with the dry vocal; a selective cut would dip deeper at 2–4 kHz and 7–10 kHz
-4-3-2-10200–500 Hz1–2 kHz2–3 kHz3–4 kHz7–10 kHz
  • Default, Depth 5
  • Default, Depth 10
  • Harsh Vocal preset, Depth 10
The values (dB)
200–500 Hz1–2 kHz2–3 kHz3–4 kHz7–10 kHz
Default, Depth 5-0.9-1.0-1.0-1.1-1.1
Default, Depth 10-2.4-2.4-2.4-2.4-2.4
Harsh Vocal preset, Depth 10-2.6-3.8-3.4-3.6-3.7

"Spread" is the difference between the most-cut and the least-cut band: how selective the cut is. On this vocal the harshness is typically at the 2–4 kHz edge and the 7–10 kHz sizzle, so a de-harshing cut should be deepest there. Resonance cut about the same amount everywhere (a spread of 0.3–1.2 dB), which is a gentle level drop rather than de-harshing.

Why

On a single tone, there is one peak to find. On a sung note, every harmonic stands above its surroundings, from the bass to the top. Resonance measured each peak against an average taken over a narrow window, so it found "resonances" everywhere and cut them all a little.

A sketch of a sung note's spectrum: every harmonic is a peak above a narrow local average drawn in red
Why the old detector spread its cut. Illustration.

What we changed

  • The test set: real vocals are now part of every fit, with held-out singers that are never fitted on, and a pass bar per measure. The same check now runs on every Vocal Console module.
  • Resonance itself was rebuilt for version 1.1 to judge each peak against wider references that a voice's harmonics don't stand out from, so its cut now gathers where a vocal is harsh instead of spreading evenly. The cut is the same at 0 dB and −20 dB input to within 0.05 dB. Measured on the 1.1 build of 14 November 2026. You can hear it on real vocals in the listening comparison.
Resonance's spectrum display: the cut it makes, drawn in green over the voice.
Resonance's spectrum display: the cut it makes, drawn in green over the voice.

Measured on 14 November 2026.