Raven.
Reshape the spectrum.
512 independent frequency bands, each with its own compressor. Raven measures the magnitude of every FFT bin and moves it toward a target spectral curve — pulling loud bands down, lifting quiet ones up, or doing both at once.
A compressor that reshapes the spectrum,
not just the level.
512 independent frequency bands, each with its own compressor. Raven measures the magnitude of every FFT bin and moves it toward a target spectral curve — pulling loud bands down, lifting quiet ones up, or doing both at once. The result is a reshaped frequency balance rather than a compressed dynamics curve.
Raven runs a weighted overlap-add STFT and treats every one of its 512 frequency bins as an independent compressor — not band-averaged groups, not broadband RMS.
Bins above the target curve are pulled down; bins below it are lifted up — either direction alone, or both at once, scaled by a single Compression depth macro. The target itself is a shaped spectral curve (Flat, Sloped, or OTT) you can tilt and offset, so you're sculpting frequency balance directly rather than reacting to a single broadband threshold.
A compressor in every bin.
Raven runs a weighted overlap-add STFT (Hann window, 75% overlap) and applies independent attack/release ballistics and a gain calculation to every one of its 512 frequency bins per frame — not band-averaged groups, not broadband RMS. Each bin is a separate compressor.
The envelope follower runs separately on each FFT bin's magnitude. A cross-bin smoothing pass lets you trade surgical precision for a more cohesive feel — 0% means every bin is completely independent, 100% means neighbouring bins pull each other into alignment.
Gain is computed as (target/detected)^amount — a true compression slope on both sides of the threshold. At amount 0.5 you get an effective 2:1 ratio; at 1.0 the bin is flattened exactly to target. Unlike a multiply-law, there is no hidden ceiling that collapses dynamics at moderate settings.
Downward and upward compression run simultaneously in the same STFT frame. Bins above target are attenuated using the Downward amount; bins below target are lifted using the Upward amount. A low-noise gate prevents near-silence from being boosted.
Hann window applied at analysis and synthesis (WOLA), hop = N/4, overlap-add normalization. The pipeline introduces a latency of exactly one FFT window, reported to the host for automatic PDC compensation. At the default 1024-point FFT this is approximately 23 ms at 44.1 kHz.
The spectral shape you're aiming for.
The Target sets the threshold level in dB and the Curve sets its spectral shape. The compressor pushes every bin toward this target — not toward a flat line, but toward a shaped reference that matches the spectral balance you want.
Uniform threshold at the Target level across all bins. Every frequency band has equal weight. Use as a starting point, or when you want the same compression depth across the whole spectrum.
A heavier threshold in the lows, lighter in the highs. Models the natural spectral balance of well-mixed bass content. Downward mode here tames sub buildup while leaving the top-end detail alone.
Over-the-top style compression: steep downward in the highs and upward in the lows simultaneously. Pushes transient detail forward while thickening the sub — the multiband-style sound of heavy processing without the band-seam artefacts.
Rotates the chosen curve around 1 kHz by ±6 dB per octave. Positive tilt raises the high-frequency target, making downward compression bite harder in the lows; negative tilt does the reverse. Trim the spectral weighting without switching curves.
Two directions, one pass.
The familiar compressor direction. Bins that exceed the target are attenuated proportionally — the farther above the target, the more gain reduction. At Downward 0.5 you get an effective 2:1 ratio; at 1.0 the bin is clamped to the target. The GR meter tracks only downward reduction so upward lift never reads as "compression."
Bins below the target are boosted toward it. Where a traditional compressor leaves quiet bands alone, Raven can pull them up — evening out an uneven spectral balance, adding low-mid body to a thin bass, or making every harmonic contribute. The same ratio law applies: 0.5 ≈ 2:1 lift, 1.0 flattens to target.
A single depth macro (0–100%) that scales both Upward and Downward amounts simultaneously in the DSP without writing back to the individual dials. Pull it back to blend the effect. Push it to 100% for maximum effect, then use the individual ratio dials to bias toward one direction.
Shape the feel of the compression.
0–200 dial, mapped geometrically to 0–280 ms per-bin envelope time. Default 40 (≈10 ms). At 0, the envelope is instantaneous — Raven behaves like a spectral clipper that hard-limits each bin to the target. Above 0, transients pass through the threshold before the envelope catches up, giving headroom and punch.
0–200 dial, mapped to 0–1 second. Default 90 (≈60 ms). Longer release times keep gain reduction in place between note hits — useful for smoothing out a choppy bass line. Shorter times let each bin recover independently, which can cause pumping on busy material.
A bidirectional low-pass filter applied across the bin array after detection. At 0%, each bin's envelope is completely independent — maximum frequency resolution. At 100%, neighboring bins heavily influence each other, producing behaviour closer to a traditional multiband compressor. 25% (default) is a light cohesion pass.
Final level and resolution controls.
Wet-path output gain applied before the Mix blend. Compensates for level changes introduced by upward/downward imbalance. The 18 dB ceiling gives enough headroom for extreme upward-only work.
Dry/wet blend. At 100% the output is fully spectral-compressed. Pull back to introduce original unprocessed signal — useful for parallel compression workflows where you want the dry transients intact.
Measures the net wet-level delta per frame (downward cuts minus upward lifts) and slews a compensating gain over ~1.5 seconds. When On, changing Upward or Downward ratios auditions the spectral character rather than a volume shift. Both channels share one gain ramp so the stereo image stays stable.
Sets frequency resolution and latency together. 256 = 128 bands / ≈6 ms latency — minimal delay, coarser separation. 4096 = 2048 bands / ≈93 ms latency — maximum bin count, high frequency resolution, PDC compensated. Default 1024 (512 bands / ~23 ms) matches the original patch.
Common approaches.