Saramsha

No. 058  ·  13 November 2026

Always Rising, Never Higher

Eight tones glide upward together, each one swelling in and fading out exactly an octave apart. Press play — the scale climbs forever, and it never arrives anywhere. Drag the column yourself to hear why.

one voice, mid-climb fading in or out where the envelope peaks — 440 Hz, concert A

About this piece

Eight sirens, one illusion

In 1964 the psychologist Roger Shepard, at Bell Labs, wanted to study a strange fact about pitch: it has two separate components. One is height — plainly, a flute note is higher than a tuba note. The other is chroma, the note's name — a C is a C whether it is played low or high, and the twelve chromas sit on a circle, not a line, because C and the C an octave above feel like "the same note, higher up" rather than two unrelated pitches. Shepard built a tone to isolate chroma from height completely: stack copies of one note in every octave at once, shade them with a bell-shaped loudness envelope centred on one comfortable register, and the ear is handed a clear note name with no reliable sense of which octave it actually lives in. Raise every copy by a semitone, let the envelope stay put, and something has to give: the chroma plainly climbed, but there is no single pitch left behind to say it arrived anywhere higher. People reliably hear it as rising. Played in a loop of twelve such steps, it rises and rises and is back where it "started" the whole time.

Shepard's version moved in discrete semitone jumps. In 1969 the composer Jean-Claude Risset made it continuous — an unbroken glide instead of twelve steps — which is what this page plays. It has shown up since in some well-known places: Hans Zimmer's score for Dunkirk runs a version of it under nearly the entire film, a slow, wordless screw-tightening that never resolves because, mathematically, it cannot.

What is actually sounding

This page runs n sine-wave voices (eight by default), spaced evenly apart in phase around one repeating register. Voice j's position in that register at "time" t is phase = frac(t + j/n), which climbs from 0 to 1 and then wraps straight back to 0. Its frequency is 55 Hz · 2^(6·phase) — six octaves, from 55 Hz up to 3,520 Hz — and its loudness is sin(π · phase), a bell curve that is exactly zero at phase 0 and phase 1 and peaks, not by accident, right on concert A at 440 Hz in the middle. Slide every voice's phase up together and each one simply rides through that fixed bell: silent at the bottom, loudest in the middle, back to silent at the top — at which instant, already inaudible, it is reborn at the bottom. You never hear the seam, because there is nothing there to hear.

Σ sin²(π(t + j/n)) = n/2 — for any t, any n ≥ 2

That line is the whole trick's safety net, and it is a plain trigonometric identity, not a coincidence tuned for this page: however many voices are spaced evenly around the register, and whatever point in the climb you catch them at, the total loudness — the sum of their squared amplitudes — comes out to exactly half the voice count, dead flat. The panel on the right computes this live. Push it to 2 voices or 14, drag the column, change the speed: the number does not move. Nothing is swelling or fading overall; only the identity of which voice is currently doing the swelling changes.

A cousin worth knowing about

In 1986 the psychologist Diana Deutsch found a related effect with just two such tones a tritone apart (half an octave): people disagree, person to person but consistently within a person, about which one is "higher" — some always hear the first as higher, some always the second, something closer to a stable personal trait than a judgement call. It is called the tritone paradox, and it points at the same underlying fact this page is built on: once height is taken away, the ear falls back on some other, more arbitrary rule to decide "up."

Everything here runs on up to fourteen sine oscillators, always playing, gated by the envelope above — nothing is recorded, uploaded or sent anywhere. Drag up or down on the column to scrub it by hand; the sound follows your finger exactly. ?t= sets the starting position (0–1), ?n= the voice count, ?rate= seconds per octave, and ?dir=-1 starts it falling.