cfe77fd524
Completes the bank whose first half went in with the scene commits. The remaining work was all in what "covered" means. Two axis ranges were wrong, and being wrong made the coverage report lie in both directions. Loudness is a raw mean spectral magnitude, not a normalised 0..1, so every song read as 0.00 and the axis looked dead. Centroid is mapped to a log frequency axis, so a track of nothing but sub-bass and a 55Hz pad still measures 0.28 and pure hiss measures 0.93 — against a nominal [0,1] the bank would have reported a permanent 50% gap that no synth change could close. Both now carry the range the statistic can really take on, with the reason. Spread on every axis is not coverage of the space, so the tool also measures correlation between axes. It immediately found brightness and noisiness moving together at r=0.95: noise colour had been tied to brightness, one axis wearing two names, leaving the dark-and-noisy quadrant unreachable. Noise colour is now its own parameter — hiss over a sub-bass pad is an ordinary record. That got r to 0.94, and no further. Flatness is the geometric mean of the spectrum over its arithmetic mean across the whole band, so a signal only measures flat if it has energy everywhere, which is the same thing as measuring bright; band-limiting the noise to hold the centroid down empties the top and drops the flatness with it. It is a property of the analyser and real material does the same, so it is allowlisted with its reason. Anything not on that list still fails — the check is there to catch a bank that has collapsed, not to relitigate physics every run. The output is gitignored. It is 172MB of deterministic audio derived from a table that will keep changing: rebuild it, do not carry it. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
142 lines
7.8 KiB
JavaScript
142 lines
7.8 KiB
JavaScript
// The song bank: a set of tracks that between them exercise every input the
|
|
// look generator reads.
|
|
//
|
|
// Why this has to exist. Everything downstream of the audio is a function of
|
|
// five summary statistics and a section list, and until now every measurement
|
|
// in this project was taken on ONE synthetic track — which turned out to have
|
|
// two sections, both quiet, making half the scene library unreachable and the
|
|
// resulting numbers wrong. A single song cannot tell you whether the generator
|
|
// is varied; it can only tell you what it does with that song.
|
|
//
|
|
// The bank is built against what the generator actually consumes, not against
|
|
// what sounds like a reasonable spread of music:
|
|
//
|
|
// summary.bpm motion bias, animation rate, personality pace
|
|
// summary.meanCentroid director odds, line weight, softness, bloom, vignette
|
|
// summary.meanFlatness director odds, chroma, saturation, surface texture
|
|
// summary.dynamicRange feedback amount and decay, contrast
|
|
// summary.meanLoudness per-section energy, temperament intensity
|
|
// section kinds which families each director opens — the casting gate
|
|
//
|
|
// Coverage is a claim, so `tools/song-coverage.js` measures it rather than
|
|
// trusting this table. Each entry names the corner it is here to occupy; if a
|
|
// song stops occupying it, that is a bug in the synth, and the coverage tool is
|
|
// what catches it.
|
|
//
|
|
// These are test signals, not music. They exist to be measured.
|
|
|
|
import { FeatureTrack } from './FeatureTrack.js';
|
|
import { synthesizeSong } from './synth.js';
|
|
|
|
/**
|
|
* @typedef {object} SongSpec
|
|
* @property {string} name
|
|
* @property {string} covers the corner of the feature space this one holds
|
|
*/
|
|
export const SONGS = [
|
|
// --- slow, quiet, spacious -------------------------------------------
|
|
{ name: 'drone', bpm: 62, arrangement: 'ambient', brightness: 0.05, noise: 0.02, dynamics: 0.95,
|
|
covers: 'beatless · darkest · most dynamic' },
|
|
{ name: 'air', bpm: 76, arrangement: 'ambient', brightness: 0.90, noise: 0.25, noiseColour: 0.15, dynamics: 0.85,
|
|
covers: 'beatless but bright — brightness without a beat' },
|
|
{ name: 'ember', bpm: 68, arrangement: 'club', brightness: 0.20, noise: 0.05, dynamics: 0.55,
|
|
covers: 'slowest track with a detectable beat — anchors the low tempo end' },
|
|
{ name: 'elegy', bpm: 84, arrangement: 'ballad', brightness: 0.10, noise: 0.03, dynamics: 0.90,
|
|
covers: 'slow, dark, one build and one payoff' },
|
|
{ name: 'hymn', bpm: 92, arrangement: 'ballad', brightness: 0.80, noise: 0.20, dynamics: 0.70,
|
|
covers: 'slow and bright — separates tempo from brightness' },
|
|
{ name: 'still', bpm: 100, arrangement: 'sparse', brightness: 0.35, noise: 0.06, dynamics: 0.75,
|
|
covers: 'the degenerate two-section case, which must not break' },
|
|
|
|
// --- mid tempo, the bulk of the space ---------------------------------
|
|
{ name: 'dusk', bpm: 118, arrangement: 'classic', brightness: 0.02, noise: 0.01, dynamics: 0.60,
|
|
covers: 'darkest and most tonal with a full arrangement' },
|
|
{ name: 'centre', bpm: 124, arrangement: 'classic', brightness: 0.50, noise: 0.20, dynamics: 0.55,
|
|
covers: 'the middle of every axis — the null hypothesis' },
|
|
{ name: 'glare', bpm: 128, arrangement: 'classic', brightness: 0.95, noise: 0.55, dynamics: 0.50,
|
|
covers: 'bright and noisy together' },
|
|
{ name: 'slab', bpm: 126, arrangement: 'club', brightness: 0.45, noise: 0.25, dynamics: 0.05,
|
|
covers: 'limitered — lowest dynamic range, longest sections' },
|
|
{ name: 'chrome', bpm: 132, arrangement: 'club', brightness: 0.85, noise: 0.06, noiseColour: 0.1, dynamics: 0.35,
|
|
covers: 'bright and tonal at club tempo' },
|
|
{ name: 'murk', bpm: 88, arrangement: 'classic', brightness: 0.12, noise: 0.85, noiseColour: 0.95, dynamics: 0.70,
|
|
covers: 'dark tonal content under bright hiss — the quadrant that breaks the brightness/noise correlation' },
|
|
|
|
// --- fast, dense, broken ----------------------------------------------
|
|
{ name: 'lattice', bpm: 140, arrangement: 'breaks', brightness: 0.60, noise: 0.35, dynamics: 0.45,
|
|
covers: 'most sections — makes rosters actually rotate' },
|
|
{ name: 'grit', bpm: 150, arrangement: 'breaks', brightness: 0.15, noise: 0.90, noiseColour: 0.85, dynamics: 0.40,
|
|
covers: "noisiest and dark — the corrupt director's home ground" },
|
|
{ name: 'plate', bpm: 138, arrangement: 'club', brightness: 0.25, noise: 0.70, noiseColour: 0.9, dynamics: 0.10,
|
|
covers: 'noisy AND compressed — two extremes at once' },
|
|
{ name: 'runner', bpm: 174, arrangement: 'breaks', brightness: 0.70, noise: 0.40, dynamics: 0.50,
|
|
covers: 'fastest with a broken arrangement' },
|
|
{ name: 'flare', bpm: 168, arrangement: 'classic', brightness: 0.95, noise: 0.85, dynamics: 0.25,
|
|
covers: 'brightest and noisiest — the far corner' },
|
|
];
|
|
|
|
const DURATION = 120;
|
|
|
|
// Analysis is CPU-bound seconds per track and every caller wants the same
|
|
// tracks, so the bank is built once and held.
|
|
const cache = new Map();
|
|
|
|
/** One song from the bank, analysed. */
|
|
export function song(name, { fps = 60 } = {}) {
|
|
if (cache.has(name)) return cache.get(name);
|
|
const spec = SONGS.find((s) => s.name === name);
|
|
if (!spec) throw new Error(`no song named "${name}" — have: ${SONGS.map((s) => s.name).join(', ')}`);
|
|
|
|
const entry = {
|
|
...spec,
|
|
track: FeatureTrack.fromAudioBuffer(synthesizeSong({
|
|
bpm: spec.bpm,
|
|
duration: DURATION,
|
|
arrangement: spec.arrangement,
|
|
brightness: spec.brightness,
|
|
noise: spec.noise,
|
|
noiseColour: spec.noiseColour ?? null,
|
|
dynamics: spec.dynamics,
|
|
seed: 1 + SONGS.indexOf(spec),
|
|
}), { fps }),
|
|
};
|
|
cache.set(name, entry);
|
|
return entry;
|
|
}
|
|
|
|
/**
|
|
* The whole bank, analysed.
|
|
*
|
|
* `count` takes an evenly spaced subset rather than the first n, so a cheap run
|
|
* still spans the space instead of testing five slow ambient tracks.
|
|
*/
|
|
export function songBank({ count = null, fps = 60 } = {}) {
|
|
const specs = count && count < SONGS.length
|
|
? Array.from({ length: count }, (_, i) => SONGS[Math.round(i * (SONGS.length - 1) / (count - 1))])
|
|
: SONGS;
|
|
return specs.map((s) => song(s.name, { fps }));
|
|
}
|
|
|
|
/** Feature axes the bank claims to span, and where each is read. */
|
|
export const AXES = [
|
|
{ key: 'bpm', label: 'tempo', of: (t) => t.summary.bpm, range: [60, 180],
|
|
reads: 'motion bias · animation rate · personality pace' },
|
|
// Not [0,1]: centroid is mapped to a LOG frequency axis, so a track made of
|
|
// nothing but sub-bass and a 55Hz pad still measures ~0.28 and pure hiss
|
|
// measures ~0.9. The range here is what the statistic can really take on for
|
|
// music, and calling it [0,1] would report a permanent 50% gap that no
|
|
// synth change could close.
|
|
{ key: 'centroid', label: 'brightness', of: (t) => t.summary.meanCentroid, range: [0.25, 0.95],
|
|
reads: 'director odds · line weight · softness · bloom · vignette' },
|
|
{ key: 'flatness', label: 'noisiness', of: (t) => t.summary.meanFlatness, range: [0, 1],
|
|
reads: 'director odds · chroma · saturation · surface texture' },
|
|
{ key: 'dynamics', label: 'dynamic range', of: (t) => t.summary.dynamicRange, range: [0, 1],
|
|
reads: 'feedback amount and decay · contrast' },
|
|
// Raw mean spectral magnitude, not a normalised 0..1 — biasFor divides
|
|
// section energy by it, so its scale is whatever the analyser produces.
|
|
{ key: 'loudness', label: 'loudness', of: (t) => t.summary.meanLoudness, range: [0, 0.012],
|
|
reads: 'section energy · temperament intensity' },
|
|
{ key: 'sections', label: 'section count', of: (t) => t.sections.length, range: [2, 8],
|
|
reads: 'roster rotation · how often the image is allowed to change' },
|
|
];
|