Sparklelizer

Jazz builds the sound from around sixteen layers of the same vocal: the lead dead centre, with doubles panned dead left and dead right. On the side layers she puts a tempo-synced stereo delay whose two sides run different note divisions - say an eighth on the left and a half note on the right - and feeds them back into each other. A repeat born on the left re-emerges on the right a half note later, then returns to the left an eighth after that. The doubles bounce across the field while the lead stays put.

The chain is fixed, because the setup being modelled is fixed. Input seeds the two lines, each line is filtered before it plays so every pass thins and darkens, an LFO moves the read position so old repeats swirl more than new ones, then the wet path gets width, a little bloom and an optional duck from the dry input before it meets the untouched lead. There is no repatching, because there is nothing here that was ever patched.

The scope is the product as much as the audio is. It is parameter-side: the repeat train is computed from the same expansion the audio runs on, not measured from an analyser, so it is correct the instant the page paints and it shows you where the repeats will land before you have played a note. Time runs along a beat grid, height is stereo position after Width, brightness and size carry level, and a faint arc links each repeat to the one it came from so the bounce is drawn as a path.

The delay is not the end of the chain, because it was never the end of the chain being modelled. After dry and wet are summed there is a reverb, then a saturator, then a compressor. Reverb and saturation are a switch and one amount each; the compressor is a switch with Threshold, Ratio and Makeup, so you can set where it starts working, how hard it holds what crosses that point, and how much level goes back on afterwards. The reverb is a real room rather than the short bloom that sits on the wet path, the saturator gives the repeats an edge, and the compressor is last on purpose: its makeup gain brings the delay feedback up in volume, which is how the repeats stay audible under the lead instead of falling away between phrases. All three are off by default, and Off skips the stage rather than running it at unity, so a patch that does not use them renders exactly as it did before they existed.

That is the part a stock stereo delay does not do. A stock stereo delay gives you a time and a level per side. This one cross-feeds two different divisions so the repeats travel rather than alternate, mixes them with a real Dry/Wet pair and a duck from the dry vocal instead of a per-side output level, filters inside the feedback loop rather than after it, draws the repeat train on the beat grid before you play a note, and carries the reverb, saturation and compression that would otherwise be three more plugins behind it.

The page can run the same compiled ScriptC kernel as the native formats through a persistent AudioWorklet. Its real-Chromium promotion gate compares the browser render with the restricted-TypeScript oracle, including all 25 controls and the Reverb → Saturation → Compressor chain. The Web Audio implementation remains available as a reference preview, but it is not used as evidence for browser/plugin parity.

Tempo can follow the host on both surfaces. Select Host and each changed website transport or DAW tempo is written through the ordinary `sync.bpm` parameter path; both delay times, the scope and saved parameter state therefore see the same value. Manual leaves the Tempo knob in charge, and unchanged host values are suppressed rather than generating automation traffic every block.