Waveform lab, round 2: A, B, C and F refined

Same Liquid Light background and layout; only the waveform changes. The audio is analysed into six bands with auto-gain, an envelope follower (fast attack, slow release) and an onset gate per band (spectral flux over an adaptive threshold). Round 2 runs every moving value through a spring or a wave simulation instead of mapping levels straight to pixels, and turns hits into motion that travels rather than flashes. The background is now a slow, soft morph that runs at full speed while playing and eases down to a crawl when paused. Round 1 versions are at the end for comparison. Press Play, then switch while it runs; keys 1 to 8 switch too.

What the research said, and how the analysis works

The pattern in the admired work: the best pieces give each frequency range its own visual job instead of having everything react to one loudness number. Emberflux runs three lines for bass, mids and highs, rippled by the waveform, with beat-triggered rings. On X, the waveform pieces people save most are 3D waveform landscapes and ribbons (@techartist_), Apple/Siri-style strands (@davidhaz, SiriWave), cubes that snap on and off on bass hits (@RedefineFX), and scope displays (Sungalizer). A Grok sweep of X found them; I did not check each post myself. Ridgelines come from Unknown Pleasures.

Why the old bars felt awkward: selectr fed raw FFT levels straight to the bars, so the bars twitched with every frame of noise, and a loud mix kept them near the top all the time. Motion-design practice (this guide, and the Bello onset tutorial) says: smooth with a fast attack and slow release, gate out the noise floor, calibrate to the mix, keep the output range narrow.

What this lab does with the audio: