Vellum changelog

• v1.0.1windowsAugust 10, 2026

  A maintenance release. Nothing here changes how Vellum sounds. Everything was found while chasing a report of audio stalling in Ableton Live.

  Fixed

  • •Much lower CPU use while the plugin window is open. The moving trace over the main display was forcing everything underneath it — all the curves, the grid and the labels — to be redrawn thirty times a second during playback. It now redraws only itself. Most noticeable if you keep the window open while a track plays.
  • •A rare source of crackle on the thread that prepares Vellum's reverb tails. The main audio path was already protected against it; this one wasn't.

  Please read — a file we should not have written

  Earlier versions of Vellum wrote a small diagnostic file into your Documents folder. On most Windows machines Documents is redirected into OneDrive, which means that file was being uploaded to cloud storage without you choosing to put it there.

  That diagnostic has been removed from the plugin entirely. Everything Vellum saves now lives in its own application-data folder, alongside your preset library, which was always stored correctly.

  If you used an earlier version, you may still have a leftover folder named Vellum inside your Documents folder. It is no longer used and is safe to delete.

  We have also added a check that prevents any future version of Vellum from writing to a folder Windows syncs to cloud storage.

• v1.0.0windowsAugust 8, 2026

  First public release. Everything below describes the shipping product rather than a diff — 0.1.0 was development only and was never tagged, released or downloadable.

  The plugin

  • •Five controls, and no bad positions. MODE (Room / Plate / Hall / Air), SIZE, TONE, DUCK, MIX. Each is a macro over several internal parameters moving along a curated curve, so every setting is usable. Diffusion, modulation, the tail high-pass and predelay were each considered as front-panel controls and deliberately left off.
  • •Reverb engine — 8-line Householder feedback delay network, 8–16 Schroeder allpass diffusers, a nested allpass inside every delay line, T60 derived from the target decay and the real loop round-trip rather than dialled in, per-line damping, and a stereo return. JUCE-free and header-only, so the plugin, the offline voicing tool and every release gate run byte-identical DSP.
  • •DUCK — a side-chain off the input that pulls the finished wet signal down while you sing and lets it back up in the gaps. It does not shorten or damp the reverb; the tail continues behind the vocal untouched. Normalised by a slow running level, so its depth does not change with input gain.
  • •Advanced panel, nine controls, collapsed by default. Seven are offsets from the voicing with neutral defaults; DIFFUSE is an absolute 0–100% smoothness macro; DRY and WET are absolute output trims, which exist because MIX is a crossfade and therefore couples level to balance (measured: MIX 0.10 → 0.60 costs 7.04 dB of dry).
  • •Impulse display — not an illustration. The engine is driven with an impulse and the result measured: three per-band energy decay curves, a live T60 marker, and a warped time axis giving the first 250 ms 30% of the width, because that quarter-second is what distinguishes one reverb from another. The predelay marker is draggable.
  • •64 factory presets across five categories — Lead Vocal 16, Backing & Doubles 10, Spoken & VO 8, Instrument 14, Creative 16. Names are checkable, not decorative: every release run drives the real engine with each preset and prints its measured decay, from Booth Tone at 0.58 s to Endless Air at 5.64 s. Shipped both compiled into the plugin (so they cannot go missing) and as real .vellumpreset files beside it (so the library can be copied, backed up or shared).
  • •Preset library overlay — filter by category, search by name, save, delete, and move whole banks in and out.
  • •In-app SIGNAL FLOW overlay — every stage of the real signal path in the order samples pass through it, each explaining itself and naming the controls that reach it.
  • •User manual — 26 pages, HTML and PDF. Its preset table and every decay figure in it are generated from a measured gate run; the build refuses to produce a manual without one.
  • •VST3 and a Standalone application, Windows 64-bit. Mono → mono, mono → stereo and stereo → stereo. No added latency. Around 1.2% of one core at 48 kHz, and decay time holds within 0.2% across 44.1, 48 and 96 kHz.

  Notable fixes during development

  These were real defects found and closed before shipping, recorded because each cost something to find:

  • •Pitch-shifting when SIZE or TONE moved during playback. The engine glided delay read pointers, which is a Doppler shift by construction. Replaced with two stationary taps and a crossfade; a regression test now proves it, and caught the old behaviour at 0.774× — about four semitones.
  • •A "tinny", brittle character on real vocals that survived backing off both TONE and DIFFUSE. That survival was the clue: the early reflection cluster carried 3–4× the late tail's proportional high-frequency energy, because a diffuser is an allpass and cannot remove brightness at any setting. Fixed with a dedicated TONE-scaled damping stage on the early cluster.
  • •DIFFUSE was inverted relative to its label, and then, once corrected, barely audible — the diffuser only ever shaped the early signal, 14% of the wet at HALL. It is now a macro that both adds allpass stages and fades the early cluster out, so 100% is genuinely a smooth wash.
  • •Sample-rate dependence in the damping coefficients: T60 spread across rates fell from 5.2% to 0.2%.
  • •The stereo SEND control was completely inert (0.001 dB) because it low-passed the signed L−R difference, which averages to zero on any zero-mean signal. Rebuilt on magnitude envelopes: 9.65 dB.
  • •Saving a preset could lock the standalone up. The save dialog was a top-level desktop window in JUCE's global modal state; it is now an ordinary overlay inside the editor, so the class of failure no longer exists.
  • •The impulse display showed a hard staircase at short decays — real structure, individual early reflections resolving separately, against a smoothing kernel three times narrower than the steps it was meant to remove.