Setting up the stage…
🎚 Source and equalizer
The science

What the bars are. The tones are synthesised sample by sample at 44.1 kHz (square, saw and triangle as exact band-limited Fourier series, no aliasing), passed through three RBJ biquad filters (low shelf 200 Hz, peaking 1 kHz with Q = 0.9, high shelf 4 kHz), and the last 4096 samples are multiplied by a Hann window and transformed with a radix-2 FFT. Each bin gives a sine amplitude Ak = 2|Xk| / Σw, shown in dBFS: 20 log10 Ak. The 48 bars are log-spaced bands from 30 Hz to 16 kHz (the loudest bin in each; bands narrower than one 10.8 Hz bin are interpolated), 24 LED cells span 72 dB, white caps are peak holds.

Spectrum of waveforms. A square wave has only odd harmonics with amplitudes 4/(πn), a sawtooth every harmonic with 2/(πn), a triangle odd harmonics falling like 1/n²; the bars show exactly that. A beat between f and f + 4 Hz is two lines the 10.8 Hz bin width cannot separate, so you see one line whose height wobbles four times a second. White noise is flat per bin, so on a log axis it rises 3 dB per octave when each band takes its loudest bin.

The equalizer. The curve under the sliders is the exact magnitude |H(ejω)| of the three cascaded filters; the bars show the measured effect, and the sound button plays the very same samples, so you hear what you see.

Honest simplifications. Window leakage and the bin grid limit the resolution (10.8 Hz) and the first bands are interpolated; dBFS is digital, not sound pressure level; the browser's speakers add their own response; the ScriptProcessor audio path has some tens of milliseconds of latency, so bars and sound are not frame-locked. Stage, LED cells, speakers and light truss were modelled and textured in Blender; the audio and analysis run live in JavaScript.

Independent checks (run on load):

running…

Analyser

Strongest tone-
Nearest note-
Peak level-
Centroid-
FFT4096 pt, 10.8 Hz
Drag to orbit, scroll or pinch to zoom. Tap “Sound” to hear it.
Made with ๐ŸŸ  Blender ยท Three.js