Each partial in the sound drives the plate modes it is near. Sand sees the time-RMS of the total motion and settles where that is smallest.
136.1 Hz is the common Oṁ / Sa reference. Pitch comes from the McLeod method, not the FFT peak — on a chanted vowel the 3rd harmonic is usually the loudest partial, so a peak-picker reports the wrong note.
Mic is analysed only — it is never connected to the speakers.
No file loaded.
Record a mantra, then replay it into the plate.
Shift-click the plate to move the driver. A mode with a node at the driving point cannot be excited — this is what selects between the − and + members of each degenerate pair, exactly as the bowing point does on a real plate. Put the driver at the centre and most modes go silent.
Partials used = 1 drives the plate from the single loudest partial: its mode dominates, the RMS reaches zero on that mode's nodal set, and you get clean lines. Raising it gives the full multi-partial response, which is physically complete but washes the figure out — with several strong partials the RMS is zero only where all their nodal sets intersect.
Base f₀ sets how densely the modes fall across the chanted pitch range. Too high and only one mode lands there, so one figure sticks no matter what you chant; ~12–20 Hz puts a dozen modes in a voice's range so the figure moves with the note.
Friction is the threshold below which a grain is not thrown at all — it is what makes the nodal band a resting place, so you get lines rather than a scatter of dots. Grain size sets a floor on the residual jitter so lines reach a finite width.