Mix Tools

FREQDIVER

Dynamic spectral unmasking — separation without level loss.

£55 £69Launch price

See the measurements

Formats: VST3 · AU · Standalone
Operating system: macOS · Universal
Version: v1.0.0

What it does

Make frequencies move out of the way.

FREQDIVER analyses a sidechain signal, builds a continuously morphing spectral priority mask, and applies that mask as frequency-dependent attenuation to the main signal. The instrumental steps aside only where and only when the vocal actually occupies the spectrum.

Per-bin spectral duckers conflate two different problems. Deciding what the mask should be is a perceptual estimation problem, solved on a 48-band ERB grid at control rate. Realising that mask on audio is a filter problem, solved by a swappable renderer. Separating them is why the same mask can be applied with zero latency or in linear phase without the two disagreeing about anything.

Running with no sidechain is legal and does nothing — the mask releases to zero and stays there. A frozen mask on a silent sidechain is the worst possible failure, because it sounds like the plugin broke the track.

Key features

Three concurrent STFT layers

Each responsible for the part of the spectrum where its window length is the right trade — roughly 85 ms below 300 Hz, 21 ms in the midrange, 5.3 ms above 3 kHz. Window lengths are specified in milliseconds and rounded per sample rate, so behaviour is identical at 44.1 kHz and 192 kHz.

Local prominence, not level

Band levels become level minus a broad spectral trend, which is what makes the plugin behave the same on a bright pop vocal and a dark spoken voice.

Two renderers, one mask

Zero Latency uses 40 Cam-spaced TPT bells plus two end shelves, driven by an interaction-compensated least-squares solve — minimum phase, no pre-echo, no latency. Linear Phase uses weighted overlap-add for mastering and parallel work.

Dive is contrast, not a second Amount

Low: broad, uniform separation. High: the mask collapses onto dominant spectral features. Mean depth is unchanged either way.

A drawable bias curve

Eight draggable points on the spectrum deciding how much each region may be ducked — plus Freeze and Learn, which capture the current or averaged sidechain shape and can write it straight into Bias.

Standard sidechain routing

Declared as a standard aux input bus, so Logic exposes it in its own sidechain menu and VST3 hosts see a second input bus. Mono or stereo main, with the sidechain mono, stereo, or disabled.

The interface

FREQDIVER with the advanced panel folded away, leaving the spectral display and the five main controls. — FREQDIVER · 1 of 1

How it works

The signal path, as the plugin’s own documentation states it.

sidechain ─► conditioner ─► 3× STFT layers ─► 48 ERB bands ─► mask pipeline ─┐
                                                                             │ G[b] dB
main ─────► lookahead delay ─────────────────► renderer ◄────────────────────┘
                                                  │
dry ──────► latency delay ────────────► mix ◄─────┘ ─► output ─► bypass

Audio

Audio demos for FREQDIVER are not published yet.

We would rather show you nothing than show you a demo that was not recorded through this plugin.

Measured

FreqDiverTests and Profile · 48 kHz

Test count: 81 tests, all ten audio-quality criteria green
pluginval: strictness level 10, clean
auval: clean
CPU — Normal, zero latency: ≈3 % of one core
CPU — linear phase: ≈1 %
Renderer accuracy at 3 dB of reduction: 0.28 dB
Renderer accuracy at 48 dB of reduction: 3.4 dB
End-to-end attack floor at 1 kHz — RMS detection: 29 ms

Figures are produced by this plugin’s own test suite and benchmark, on the hardware and sample rate named above. They are not estimates.

Controls

Core

Amount: Overall intensity. Tuned so 20 % is just audible, 35 % removes about 3.5 dB, 50 % is a typical mix move and 100 % is aggressive.
Dive: Contrast. Changes the mask's shape, not its mean depth.
Attack / Release: Per-band envelope, displayed as time-to-90 %.
Max Reduction: Ceiling, 0–48 dB.
Mix: Dry/wet, latency aligned.

Shaping

Range Low / High: Restrict processing to a frequency range.
Smoothness: Mask slope, 9 → 2 dB per Cam.
Gravity: Makes strong shadows reach further — signal-dependent, unlike Smoothness.
Sensitivity: Vertical offset before rectification.
Smart Amount: How much to ignore the sidechain's absolute level. 100 % is spectral shape only; 0 % means a louder sidechain ducks harder.
Bias curve: Eight draggable points deciding how much each region may be ducked.

Routing and monitoring

Stereo Link / Process Mode: Stereo, Mid, Side, Mid+Side, Left, Right.
Detection: Peak, RMS or Adaptive — a crest-factor blend.
Freeze / Learn: Capture the current or averaged sidechain shape.
Delta / SC Listen: Monitoring, in the top bar rather than behind Advanced.
Lookahead / Render / Quality: Latency-affecting, and therefore deliberately not automatable.

Technical details

Formats: VST3 · AU · Standalone
Operating system: macOS 10.13 or later
Architecture: Universal — Apple Silicon and Intel
Plugin code: Fdvr
Version: 1.0.0
CPU: ≈3 % of one core at 48 kHz zero-latency, ≈1 % in linear phase

Worth knowing

Attack cannot outrun the analysis. Measured end to end at 1 kHz, the floor is 29 ms (RMS), 55 ms (Adaptive) or 76 ms (Peak) however fast Attack is set.
Renderer accuracy is depth-dependent: 0.28 dB at 3 dB of reduction, 3.4 dB at 48 dB. This is a measured limit of the filter family, not a tuning issue.