Parameters
| Parameter | Range | Default |
|---|---|---|
| Duration | 10.0 – 100.0 | 30.0 |
| Sensitivity | 0.0 – 1.0 | 0.5 |
Duration — Length of the attack window kept around each detected onset, in milliseconds. Short values (10–25 ms) capture only the sharp click of an attack — useful for isolating drum transients to layer with a sub. Longer values (50–100 ms) keep more of the body, capturing the full “tk” of a snare or the pluck of a string before it sustains. Pairs with fade_out to control how cleanly the window closes.
Sensitivity — How easily an event is flagged as a transient. Low values only catch the loudest, sharpest attacks; high values pick up subtler events like soft hat ticks or slow string articulations. If you’re getting too many onsets, lower this; if percussive attacks are being missed, raise it. The detection runs on the percussive component (after HPSS pre-filtering), so it’s already biased toward attacks rather than tonal content.
Additional controls
Interval — Minimum time between consecutive detected transients, in milliseconds. Acts as a debounce — if two onsets are detected closer than this they’re collapsed into one. Useful for cleaning up false double-detections on cymbals and noisy drums. Set higher than your fastest expected event rate (e.g. 50 ms is safe for most beats; lower it for fast hi-hat work).
Fade Out — When on, fades the mask gently to zero across the duration window so the extracted transient ends smoothly. Off gives a hard cut at the window edge — sharper but more likely to introduce a click. Generally leave on; turn off only when you want the most surgical, precisely-shaped slice.
Invert — Swaps which output is which. With invert off (default), Output 1 carries the extracted transients and Output 2 the residual sustain. With invert on, Output 1 is the rejected transients (the sustain) and Output 2 is the kept transients. The UI labels update to show which is which. Most workflows use invert as a quick way to grab the sustain side without rewiring the graph.
About Transient Extractor
Every K2K extractor produces two outputs: extracted (the part you asked for) and rejected (everything else). The two sum back to the original — splitting transients from sustain doesn’t lose any information, so you can process each side independently and recombine them later. This is the K2K-specific take on transient processing: instead of one “shape transients” knob, you separate the two streams and treat them as independent signals.
Generated 2026-05-05 from K2K_Dev@96730bdc by scripts/gen_lexique.py. Edit _intros/ or _overrides/, not this file.