Saramsha

No. 043  ·  29 October 2026

Every Row Is a Pitch

A spectrogram draws sound as a picture: time runs left to right, pitch runs bottom to top, and brightness is loudness. Draw directly onto one — or load a photo — and hear precisely what you drew played back as real tones.

time, left to right → ↑ higher pitch · brighter = louder

About this piece

The one picture a sound actually is

A microphone only ever records one number at a time — air pressure, bouncing up and down thousands of times a second — so a sound waveform drawn honestly is an unreadable scribble: too fast for the eye and, taken as a picture, carrying almost none of what the ear hears as pitch. A spectrogram is a different, far more useful drawing of the same sound: it slices the recording into short windows, runs each one through a Fourier transform to ask "how much of each frequency was present just then," and stacks the answers side by side. The result is a picture with three honest axes — time running one way, frequency running the other, and loudness as brightness — and it is the picture both speech scientists and audio engineers actually look at, because it shows what a waveform hides: which pitches are sounding, when, and how strongly.

This page runs that idea backwards. Instead of analysing a recording into a spectrogram, it lets you draw a spectrogram and synthesises the sound that picture describes. Underneath the canvas sit 48 silent oscillators, one per row, each tuned to a fixed pitch spaced evenly on a logarithmic scale — doubling in frequency, not adding a fixed number of hertz, because that is how the ear actually judges distance between two pitches. Pressing play simply tells each oscillator to trace your drawing's brightness along its own row as a loudness curve over time. Nothing is recorded or analysed; the picture is the instruction.

Why a diagonal line sounds like stairs

A true glissando is a pitch sliding continuously, with no steps in it at all. This page can't play one, and that limit is worth sitting with rather than hiding: there are only 48 rows, so only 48 pitches exist to choose from, and a line drawn diagonally across them is really a sequence of distinct pitches switched on and off in turn — which is exactly what you hear, a quick staircase rather than a slide. Real spectrograms made from real recordings have the same kind of graininess, just with far more rows (a few hundred to a few thousand) and far shorter time windows, which is why a spoken vowel's pitch looks like a smooth curve rather than a visible staircase on a proper one. More resolution hides the steps; it never removes them — a spectrogram, however fine, is still built from finite slices of time and frequency, never a perfectly continuous one.

Hiding a picture inside a sound

Because the trick runs in both directions, a handful of musicians have used it as a signature: Aphex Twin's Equation and Windowlicker, and Nine Inch Nails' My Violent Heart, all hide a drawn image in their closing seconds, audible as noise until you look at a spectrogram of the recording and find a face staring back. The Face preset here is a small, honest version of the same joke, run the opposite way: drawn first, so you hear it as sound before you ever see why it looks the way it does.

All synthesis happens locally with the Web Audio API — 48 sine oscillators, one gain envelope per row scheduled in a single call so the whole piece plays without clicks, summed through a limiter. Nothing you draw, photograph or play is recorded, analysed or sent anywhere.