A Japanese and Chinese instrument studio with a generative arranger. Every melodic instrument is synthesized in the browser — plucked strings modelled as vibrating delay lines, bamboo flutes as pitch-tracked breath noise, bells as stacks of inharmonic partials. The drums are played from recordings by default; untick Sampled drums, beside the ensemble picker, and they are synthesized too, from struck-membrane models.

Quick start

1. Hit Play (or Generate Piece) — it writes a full piece and plays it.
2. Change Key, Scale, Style, Tempo or Density, then Generate again.
3. Export it as MP3, WAV or MIDI.

Styles

• Sakura 桜 — Japanese folk: koto and shakuhachi, gentle, in-sen scale.
• Gagaku 雅楽 — imperial court music: very slow, a shō cluster holding under a hichiriki reed.
• Guzheng 古箏 — Chinese courtly: flowing right-hand runs with a dizi over the top.
• Taiko 太鼓 — drum ensemble: the percussion leads and the melody answers it.
• Zen 禅 — honkyoku meditation: a solo shakuhachi, a temple bowl and a great deal of silence.
• Silk Road 絲綢之路 — erhu and pipa over a walking bass; the caravan tread.
• Kabuki 歌舞伎 — theatrical shamisen, angular and dramatic, with wooden clappers.
• Neo-Tokyo 新東京 — fusion: a synthetic koto over a beat, dusty keys underneath.

Four action styles, built for games — fast, driving, and never letting the pulse go:

• Samurai 侍 (battle) — 140–164 BPM. A driving taiko pulse on every beat, an eighth-note biwa ostinato, syncopated pipa stabs and a shakuhachi over the top. In-sen.
• Shinobi 忍 (stealth) — 128–150 BPM. Dark iwato, a sub bass you feel, a muted koto ticking, off-beat shamisen stabs and a plucked shamisen lead. Tension rather than attack.
• Wuxia 武侠 (martial) — 136–160 BPM. The Chinese martial-arts film: the guzheng's running hand, galloping rim shots, the festival taiko pattern and a dizi lead. Shang mode.
• Cyber Edo 電脳 (boss) — 150–176 BPM. Everything synthetic and pushed forward, the battle pulse underneath and the hichiriki — the loudest reed in gagaku — carrying the tune.

Choosing a style sets its instrument and its ensemble together; the ensemble buttons still override it afterwards.

Each style also carries its own small mix trim behind the faders. A driven bass or a stabbed strum is several dB louder than a folk strum at the same fader, so without it the action styles would bury the melody and the temple styles' plucked parts would vanish. The trims hold every style to one measured balance: the taiko level with the melody, the koto under it, the bass and strums supporting, and the drone a bed. The faders keep meaning the same thing whatever is playing.

Solos

In about two action pieces out of three, the middle of the form is a Solo 独奏 instead of Ma. Where Ma strips the piece back to a flute over a drone, a solo does the opposite: the whole ensemble keeps playing while the lead goes to the top of its range. It is built the way 91 Rock builds a guitar solo, in this idiom's terms:

• Its own phrase. Written on fast lick rhythms, with the chaos engine let further off the leash than for the tune, so it wanders where the tune would not.
• A register arc across all eight bars: the lick and then the lick an octave up, a climb to the top for the last phrase, a peak in the middle, a fall, or staying put — with the odd octave leap on a weak note.
• Flourishes — the shakuhachi's rolled ascent, the dizi's burst of tonguing, the shamisen's cascade: a fast run up or down the scale into the next downbeat. Never two bars running, never more than one bar in three, and some solos have none at all.
• It runs through Ma's travelling harmony on the pentatonic, settling its long notes onto the centre, and sits a touch over the tune in the mix because it is the feature.

A solo is drawn on streams of its own: whether a piece has one, and what it plays, never changes a note of the tune, the harmony or the rest of the form.

Form — jo-ha-kyū (序破急)

Every piece is built on the organising principle of nearly every Japanese performing art. Jo 序 — a slow, unmeasured opening, a bell or a pair of clappers announcing it. Ha 破 — "the break", where the material develops. Ma 間 — the still centre, where the flute plays alone over a drone and the drums drop away. Then Ha again, Kyū 急 — the rapid finish — and Ketsu 結, a closing that thins back to nothing.

Scales

All the melodic material is pentatonic, and the five notes are not interchangeable. The Japanese in scales — hirajōshi, in-sen, iwato — carry a semitone above the tonic, which is where koto music gets its plaintive colour. The Chinese gong, shang and yu modes have no semitones at all, so they read open and ceremonial instead. Ryūkyū (Okinawa) is the outlier, with major thirds and no second or sixth.

Harmony

This music is heterophonic — one melody, ornamented by everyone at once — so there is no chord progression to write. What there is instead is a tone centre that shifts: long stretches on the tonic, then a move that works like a change of scenery rather than a cadence. Four rules keep it in the idiom, and all four are enforced rather than hoped for:

• The tonic never leaves. The shō holds a pedal underneath the whole piece, so a move to the fourth or fifth above it reads as colour rather than as a change of key — and the music can travel without ever needing a cadence to get home. Ma is the exception: the drums are gone, the flute is alone, and the bed follows the journey, which is what makes that section feel like the one place the ground actually moves.
• Centres move by fourth or fifth. Never by a semitone or a tritone — those are the two things a drone will not forgive. Where a mode has no perfect fifth above a degree (in-sen has a tritone there, and iwato has no fifth at all), the move takes the fourth instead.
• Voicings are chosen by interval, not by counting degrees. Stacking adjacent degrees of a pentatonic does not give fourths and fifths — on in-sen it gives a root, a flat second and a fourth, and on gong a root, a second and a major third. Each voice names the interval it wants and takes whichever scale degree sits exactly there. Across all twelve modes and every degree, no voicing contains a tritone.
• The shō is allowed the cluster. Gagaku's aitake chords genuinely do contain a second, and that second grinding against a held fifth is the sound of the instrument — so the drone gets it, and nothing else does.

How it opens and closes

Three shapes at each end, drawn per seed: the percussion alone calling the piece to order, one voice leading with the ensemble under it, or that voice entirely alone so the first taiko stroke lands on Ha I. Which voice is its own decision — the shō holding a cluster is how a gagaku piece actually begins, a solo shakuhachi is how a honkyoku does, and a koto figure or a single rolled chord are how the folk repertoire does. The close is drawn the same way and independently, so the way a piece opens tells you nothing about how it ends.

When the percussion steps aside, one stroke still sounds: the bell, clappers or gong that call the piece to order, answered at the very end by the same voice. That stroke is punctuation rather than accompaniment — a piece opening on a solo flute still opens with the bell before it, and a form that ends the way it began is the oldest way there is of telling a listener it is over.

Both choices are made on their own seeded streams, and the ornaments are hashed from each note's own identity rather than drawn in sequence, so changing how a piece begins cannot alter a note of what follows. Two pieces from one seed with different openings are the same piece, opened differently.

The shape of a piece

Repeating one four-bar cycle for a whole piece is not stillness, it is inertia — the harmony ends where it began without ever having been anywhere. Long traditional forms do move: each dan of a koto danmono sits in its own relationship to the tonic. So each section reads the piece's cycle differently:

• Jo and Ketsu hold the tonic — a piece opens and closes at home.
• Ha I is a question: it starts home and ends held on the partner centre, so it has not resolved when Ha II arrives on the tonic with the melody.
• Ha II plays the cycle as written.
• Ma travels.
• Ha III transposes the cycle onto the fifth — the same music seen from somewhere else.
• Kyū runs eight bars instead of four, built as a statement and a varied answer: the home cycle, two bars on a second partner centre, then the home cycle's close. It is the only place outside Ma that uses three centres, which is what makes the climax sound like more than the loudest repeat of Ha II.

Section lengths are drawn per seed too: the statement runs three phrases or four, Ma is brief or long, and in four pieces out of ten the climax runs a whole cycle longer.

Question and answer

Inside each four-bar phrase the melody's middle statement pauses on the fifth — a phrase that stops there plainly hasn't finished — and the last statement comes down onto the tonic, approached from the note before it rather than dropped to the bottom of the range. Each motif is also written several times over and the most singable take is kept: mostly by scale step, one leap filled in by a step back, strong beats on the centre, one peak. A semitone costs as a melodic step unless it leans onto the tonic, which is the in-scale gesture that gives koto music its colour.

The harmony, the section lengths and the melody are each drawn on a stream of their own, so they belong to the seed alone: Density changes how busy the accompaniment is without rewriting the piece underneath it.

In the second half of Kyū the ensemble converges on the melody: the koto abandons its figure and plays the tune itself, a register below, stripped to the notes that carry it. That is what heterophony sounds like when a piece means it, and it is the loudest structural gesture the idiom has.

Instruments

• Plucked strings — koto, guzheng, shamisen, biwa, pipa. A burst of noise fired into a delay line that feeds back through a damping filter: the loop keeps the frequencies that fit a whole number of times into its length, which is what a real string does. The three knobs that separate these instruments are physical ones — where the string is plucked, how hard it's damped, and how bright the plectrum is.
• Blown — shakuhachi, dizi, xiao, hichiriki. Mostly air: noise band-passed around the note and tracking it, with the pitch arriving from underneath as the air column settles.
• Bowed — erhu and zhonghu. No fingerboard, so erhu playing lives on slides and a wide, singing vibrato; both are turned up.
• Struck metal — rin bowls and bells, built from inharmonic partials that decay at different rates. That inharmonicity is what a filtered sawtooth can never fake.

Ornament

Almost everything that makes these instruments sound like themselves happens in the first and last tenth of a note. The koto's oshide — the string pressed behind the bridge so the pitch rises after the attack. The shakuhachi's meri, the pitch dropped by tilting the flute. Grace notes leaning into a beat, and slides between notes close enough that a player's finger would travel rather than jump. Each style decides how freely the arranger reaches for them.

Taiko ensemble

Unlike a four-on-the-floor kick, the drum pattern is not fixed — which beats the big drum takes is the whole character of a piece, so it's drawn per seed. The ō-daiko booms, the shime-daiko rides tight on top, the ka is the same stick on the drum's wooden rim, and the tsuzumi bends upward after the strike because the player squeezes its cords. Around them sit woodblocks, clappers, suzu bells, a bowl and a gong.

Sampled drums

The Sampled drums switch replaces eight of those voices — ō-daiko, taiko, shime, ka, woodblock, clappers, gong and bowl — with recordings of the real thing. It is a per-voice swap rather than a mode: the tsuzumi, suzu and shaker have no recording good enough, so they carry on being synthesized alongside the samples.

Every Freesound recording here is CC0 — public domain — which is why those particular recordings were chosen rather than better-sounding ones under other terms. The ō-daiko, gong and clappers are the exception: they were supplied by the app's author rather than taken from Freesound. CC0 asks for no attribution and attaches no share-alike, so audio you export can keep its own CC0 dedication; a CC BY-SA sample set would quietly bind share-alike terms to everything made here. The recordings are credited below and in samples/CREDITS.md because the people who made them deserve it, not because the licence demands it.

The recordings — about 170 KB — load as the page opens. Each one is level-matched to the synthesized voice it replaces, drum by drum, so the switch changes the sound and not the volume. The match is made in the full mix rather than drum by drum in isolation: the recordings sit in the same registers as the koto, the melody and the bass, where they are partly covered, while the synthesized drums boom and click in the gaps — so the recordings are matched on how clearly they carry over the ensemble, not on how loud they are alone. Velocity does more than set gain: a softly struck drum is duller as well as quieter, so quiet hits are rolled off as well as turned down, and the two taiko recordings alternate so a run of hits doesn't machine-gun.

Seeds

Every piece comes from a seed. Each Generate rolls a new one; close the lock beside it (or type a seed) to keep it — the same seed always rebuilds the identical piece, so you can tweak tempo or instruments against one you like.

Chaos Engine

The chaos engine does not write the melody a note at a time any more — that produced a line with a new shape every bar, which is precisely why none of it was memorable. It works a level up instead, choosing the two things that decide what a phrase sounds like: the rhythmic cell the motif is built on, and the contour it follows. Those are then fixed and repeated, so the phrase stays learnable while the attractors still decide what it is.

• ρ Lorenz — shapes the melody twice. One attractor reads a position along the cell list, from sparse and long-limbed to busy; another reads the contour, from falling to rising; and the same system then picks the notes inside that shape. Low ρ settles, ~28 is the butterfly, high roams.
• Chaos ↔ Taste — a real interpolation, not a switch. At 0 every motif gets the middling rhythm and the plain arch, and the learned policy picks every note. At 1 the attractors choose outright. In between, the choice slides between the two.
• Temperature — how freely the note-by-note choice explores inside the contour: low is greedy and repetitive, high wanders.
• r Logistic — drives the percussion ornaments, the koto figure and the bass, not the melody. Below 3.57 the orbit is periodic and bars repeat; toward 4.0 no two are alike.

Two things bound all of it, and they are what keep the tune singable however the sliders are set: every note is held within a scale degree of its contour target, and the motif's span is capped at an octave. The attractors decide the shape; the shape decides the notes.

Effects

The temple reverb is a long, dark hall — its impulse loses its top end as it decays, the way a wooden building does. The echo is tempo-synced with a filter in its feedback loop, so the repeats can be pushed without the low end turning to mud. Ensemble spreads three slowly-detuned copies across the stereo field: on a koto ensemble it stands in for the fact that no two instruments are ever quite in tune with each other, which is a feature of the sound rather than a flaw.

Exports

MP3 and WAV are rendered offline — faster than realtime, and sample-deterministic, so the file is exactly the mix you set up rather than a recording of playback. Whatever the drums are set to, sampled or synthesized, is what gets rendered. Either one asks first how you want it written out:

• Single file — the whole mix in one file, with your mixer levels and mutes applied.
• Separate instruments — one file per part (taiko, bass, koto, strums, drone, melody), named after the seed so a set stays together. Every part is rendered from the same event list with the same lead-in, so the stems line up sample-accurately in a DAW, and each carries only its own reverb and echo tails.

MIDI writes one track per part with the percussion on channel 10. General MIDI has real programs for most of this ensemble — koto, shamisen and shakuhachi are all in the standard set — so the file plays back recognisably rather than as five pianos.

Credits & licence

Created by: Wilfredo Bigol Jr

Licence

The music is yours. All generated and exported audio is released under the Creative Commons CC0 1.0 Universal Public Domain Dedication. You may copy, modify, distribute, and perform the work, even for commercial purposes, all without asking permission.

The engine is not. The Chaos Engine — the generative system behind this app — is the intellectual property of Wilfredo Bigol Jr. Reproducing, distributing or modifying it, in whole or in part, without permission is prohibited.

Sampled drums — CC0 recordings from Freesound, trimmed to a single strike and level-matched, plus the ō-daiko, gong and clappers supplied by the app's author:

• ō-daiko — supplied by Wilfredo Bigol Jr
• taiko — taiko.wav and taiko2.wav, Anuj_Shrestha1
• shime-daiko and ka — taiko tom 4 and taiko tom 5, Logicogonist
• woodblock — wood block.wav, calaudio
• gong — supplied by Wilfredo Bigol Jr
• bowl — Bell at Daitokuji temple, Kyoto, nahmandub
• hyoshigi (clappers) — supplied by Wilfredo Bigol Jr

MP3 encoding by LAME (lamejs 1.2.1), used under the LGPL-3.0 and kept as a separate, replaceable file.