toolchain/timbre · core · Tangled

aesthetic.computer / core

Monorepo for Aesthetic.Computer aesthetic.computer

core toolchain timbre

2 files

at aesel-native-integration-plan

prompt.ac/@jeffrey · Timbre gets a geometry: AC had been addressing tone colour by catalogue number everywhere — a flat list you Tab through, a 0-127 digit buffer, a random seed — while pitch got hues, chord morphing, distance and a twelve-rung practice ladder, which is exactly the imbalance Wessel opens his 1979 paper by blaming on acoustic instruments ("very accurate control over pitch but little in the way of compositionally specifiable manipulation of timbre"), and which no longer has an excuse in a fleet where gm_synth exposes ~100 continuous parameters per program and Fluoddity exposes 80 floats. The rhythm shelves of that platter shipped long ago as pop/lib/necklace.mjs; shelf 10 had nothing. So: measure the axes Wessel used to INTERPRET his space — a Bark centroid over a Schroeder-spread excitation pattern for brightness, rise time and cross-band onset spread for bite — and hang the levers off them. Steps 2 and 3 of his method are deliberately skipped: no listeners were asked and no MDS was run, so every generated file says in its header that this is a proxy for a timbre space and not one. The axis was validated before anything was built on it (sine 2.72 < square 4.29 < sawtooth 5.30 < noise 18.86 Bark, with the C harness and its JS port agreeing to three decimals), and its stimulus-set relativity measured rather than waved at (brightness rank correlation against C4 is rho 0.92 at C3, 0.82 at C5). Five levers land: notepat's Tab cycle is now a derived brightness ramp rather than authoring order (sine, composite, triangle, square, whistle, sawtooth, harp — the hand-written list had square and sawtooth swapped); Menu Band's 8x16 picker is re-laid on the measured coordinates with brightest on top and bite increasing rightward per Figure 1, arrows walking the grid so "right" means "more bite", family colours left keyed to program number so the palette scatters and the scatter is the finding, and `defaults write ... instrumentGridLayout catalogue` restoring the old map; streamSplitRisk predicts melodic fission from brightness gap times van Noorden's rate term and reports rather than forbids, because the split is also a device; Fluoddity's genome becomes addressable through lerp, the Ehresman/Wessel parallelogram and a distance, with rule() so a bred instrument can be saved instead of being a one-way door, and a path harness that renders and measures the blend instead of assuming it is smooth (no cliff, worst step 2.2-3.7x mean; parallelogram exact); and p-centres pull each note earlier by its own perceptual attack lag in kidlisp's timeline, mean-zero so tempo never drifts and a single-timbre line comes out bit-identical. · 1mo ago

Timbre gets a geometry: AC had been addressing tone colour by catalogue number everywhere — a flat list you Tab through, a 0-127 digit buffer, a random seed — while pitch got hues, chord morphing, distance and a twelve-rung practice ladder, which is exactly the imbalance Wessel opens his 1979 paper by blaming on acoustic instruments ("very accurate control over pitch but little in the way of compositionally specifiable manipulation of timbre"), and which no longer has an excuse in a fleet where gm_synth exposes ~100 continuous parameters per program and Fluoddity exposes 80 floats. The rhythm shelves of that platter shipped long ago as pop/lib/necklace.mjs; shelf 10 had nothing. So: measure the axes Wessel used to INTERPRET his space — a Bark centroid over a Schroeder-spread excitation pattern for brightness, rise time and cross-band onset spread for bite — and hang the levers off them. Steps 2 and 3 of his method are deliberately skipped: no listeners were asked and no MDS was run, so every generated file says in its header that this is a proxy for a timbre space and not one. The axis was validated before anything was built on it (sine 2.72 < square 4.29 < sawtooth 5.30 < noise 18.86 Bark, with the C harness and its JS port agreeing to three decimals), and its stimulus-set relativity measured rather than waved at (brightness rank correlation against C4 is rho 0.92 at C3, 0.82 at C5). Five levers land: notepat's Tab cycle is now a derived brightness ramp rather than authoring order (sine, composite, triangle, square, whistle, sawtooth, harp — the hand-written list had square and sawtooth swapped); Menu Band's 8x16 picker is re-laid on the measured coordinates with brightest on top and bite increasing rightward per Figure 1, arrows walking the grid so "right" means "more bite", family colours left keyed to program number so the palette scatters and the scatter is the finding, and `defaults write ... instrumentGridLayout catalogue` restoring the old map; streamSplitRisk predicts melodic fission from brightness gap times van Noorden's rate term and reports rather than forbids, because the split is also a device; Fluoddity's genome becomes addressable through lerp, the Ehresman/Wessel parallelogram and a distance, with rule() so a bred instrument can be saved instead of being a one-way door, and a path harness that renders and measures the blend instead of assuming it is smooth (no cliff, worst step 2.2-3.7x mean; parallelogram exact); and p-centres pull each note earlier by its own perceptual attack lag in kidlisp's timeline, mean-zero so tempo never drifts and a single-timbre line comes out bit-identical.

Timbre gets a geometry: AC had been addressing tone colour by catalogue number everywhere — a flat list you Tab through, a 0-127 digit buffer, a random seed — while pitch got hues, chord morphing, distance and a twelve-rung practice ladder, which is exactly the imbalance Wessel opens his 1979 paper by blaming on acoustic instruments ("very accurate control over pitch but little in the way of compositionally specifiable manipulation of timbre"), and which no longer has an excuse in a fleet where gm_synth exposes ~100 continuous parameters per program and Fluoddity exposes 80 floats. The rhythm shelves of that platter shipped long ago as pop/lib/necklace.mjs; shelf 10 had nothing. So: measure the axes Wessel used to INTERPRET his space — a Bark centroid over a Schroeder-spread excitation pattern for brightness, rise time and cross-band onset spread for bite — and hang the levers off them. Steps 2 and 3 of his method are deliberately skipped: no listeners were asked and no MDS was run, so every generated file says in its header that this is a proxy for a timbre space and not one. The axis was validated before anything was built on it (sine 2.72 < square 4.29 < sawtooth 5.30 < noise 18.86 Bark, with the C harness and its JS port agreeing to three decimals), and its stimulus-set relativity measured rather than waved at (brightness rank correlation against C4 is rho 0.92 at C3, 0.82 at C5). Five levers land: notepat's Tab cycle is now a derived brightness ramp rather than authoring order (sine, composite, triangle, square, whistle, sawtooth, harp — the hand-written list had square and sawtooth swapped); Menu Band's 8x16 picker is re-laid on the measured coordinates with brightest on top and bite increasing rightward per Figure 1, arrows walking the grid so "right" means "more bite", family colours left keyed to program number so the palette scatters and the scatter is the finding, and `defaults write ... instrumentGridLayout catalogue` restoring the old map; streamSplitRisk predicts melodic fission from brightness gap times van Noorden's rate term and reports rather than forbids, because the split is also a device; Fluoddity's genome becomes addressable through lerp, the Ehresman/Wessel parallelogram and a distance, with rule() so a bred instrument can be saved instead of being a one-way door, and a path harness that renders and measures the blend instead of assuming it is smooth (no cliff, worst step 2.2-3.7x mean; parallelogram exact); and p-centres pull each note earlier by its own perceptual attack lag in kidlisp's timeline, mean-zero so tempo never drifts and a single-timbre line comes out bit-identical.