Spokes.wiki Search About

Synthesis — Music-Tech Wiki

The evolving thesis for this spoke. It was spun out on 2026-07-09 from a burst of five sources the human sent in quick succession, so the founding corpus is small and its shape is provisional — but two through-lines are already clear.

Current thesis

1. Routing beats components — modular-routing

The sharpest pattern across the founding instruments is that character comes from how blocks are wired, not from the blocks themselves, and the interesting designs expose that routing. filtopia wires five filters in parallel (not one serial VCF) and lets each oscillator hit a different one, which is exactly what buys it formant/vocal timbres a linear synth can't reach. polyuanalog builds polyphony by orchestrating many identical voice boards from a conductor board — modularity at the voice level. shinronin wraps ordinary delays/filters/saturators in a software patch matrix so the user assembles the effect; cyma-forma-alt does the same in hardware with a physical pin matrix. Four scales of one idea (filter bank / voice architecture / software effect graph / hardware pin matrix). next-oxx is the useful counterexample: a fixed-architecture effect whose value is a tuned engine, not routing freedom — proof that patchability is a deliberate design choice, not a default.

A sharper sub-pattern within this: how a synth filters is often its identity, and the two all-analog flagships bet oppositely. filtopia runs five filters in parallel (each oscillator to a different one → formant timbres); cyma-forma-alt stacks two MS-20-style multimode filters in series. Parallel vs serial is the same "routing defines character" claim narrowed to the filter section.

The limit case of this thesis is a device that is only routing — a mixer. Korg's NTS-4 (2026-07), a six-channel DIY performance-mixer kit, makes no sound; its whole value is signal flow (levels, effect sends, cueing, muting). Where filtopia/shinronin expose routing inside an instrument or effect, a mixer is character-from-wiring with the instrument removed — so it's the purest instance of the thesis, not a weaker one. Two wrinkles come with it. First, it's a performance mixer, not a passive summing box: per-channel CUE (private pre-listen) and MUTE make routing a played gesture, which ties this back to thesis #3's who moves the controls — here the answer is "the mix engineer, live." Second, the NTS-4 is the spoke's second DIY/self-assembly device after polyuanalog, but from a major maker's kit line (Korg Nu:Tekt) rather than an academic open-hardware project — a build-it-yourself sub-thread that now spans both origins. (It's an analog mixer path with two digital FX engines, so also a small hybrid on the thesis-#2 axis.)

And the unbounded case, which crosses the Eurorack edge (2026-07-22). Every instance above exposes routing inside something bounded — one instrument, one effect, one mixer, one DAW's device catalogue. VCV Rack removes the boundary: modules from any developer, audio and CV as the same cable, no fixed signal path, the patch itself as the instrument. That's the Eurorack model proper, and it makes Rack the strongest instance of this thesis in the corpus — stronger than Reason, which uses the rack as an interface metaphor over a fixed first-party catalogue. Reason looks like a rack; Rack is one.

It arrived from an unexpected direction. The spoke reached it via RackDroid, an Android port that runs the whole engine on a phone, so the modular entered as free software rather than hardware — a free APK against the thousands of pounds a comparable Eurorack case costs. That splits two properties the corpus had only ever met together: modularity is architectural (open routing, mixed sources, CV as a first-class signal), analog is a signal-path property. A virtual rack keeps the first and drops the second, which turns it into a control on thesis #2 — whatever people value in modular that survives in software wasn't coming from the circuits. (T2, and a caveat worth carrying: RackDroid has 9 stars and nobody in the sources has run it.)

2. The analog-vs-digital sound argument

A live tension runs through the corpus. filtopia sells an all-analog signal path; the test-equipment scene pushes the purist claim to its limit — "these are better sine waves, they just are" — arguing raw lab oscillators beat modelled ones on tone. polyuanalog hedges: it is analog but uses DCOs (digital control of analog oscillators) for tuning stability, conceding a bit of "character" for reliability. So the spoke isn't uniformly pro-analog; it holds the argument rather than settling it. Watch whether later sources bring measured evidence (not just taste) to the "better sine waves" claim — right now it is an aesthetic assertion (T2, opinion-heavy). The digital pole now has an instrument, not just plugins (2026-07-09): gr-mega is a fully-DSP granular synth whose entire premise (granular-synthesis — grain-cloud resynthesis of samples) can't exist in analog. So the axis isn't analog-good/digital-compromise; digital enables sounds analog can't make. That reframes the tension from "which is better" to "different tools for different sounds," with the plugins and the GR-MEGA on one side and the boutique analog synths on the other.

The argument leaves the instrument and lands on the effect (2026-07-19). The guitar-pedal gives the tension its first effects-side, non-synth source analog-vs-digital-pedals, and it maps almost the same way — analog for overdrive/fuzz (the circuit's own clipping nonlinearity is the tone, the effects echo of the "better sine waves" purism) versus digital for delay/reverb/complex modulation (which need memory and math). Two things this adds. First, it promotes the through-line to its own node, analog-vs-digital, now that it spans instruments and effects — with the software audio-effect-plugin as the far-digital end and the analog stompbox as the far-analog one. Second, it strengthens the reframing above: pedals sort effects by type onto each format (dirt→analog, time/pitch→digital), which is the "different tools for different sounds" reading stated as a rule of thumb rather than a taste claim — and the hybrid pedal (analog path + digital control) is the effects-side version of polyuanalog's DCO hedge, dissolving the binary from the other end.

The mechanism, at last, and it is one hyperbolic tangent. The argument in this thread has always been conducted in adjectives — warm, organic, digital-sounding — with the corpus holding nobody's account of what in a circuit produces the character. moog-ladder-nonlinear-digital (Huovilainen, DAFx-04, T1, via research pass 2026-08-12) supplies one for the case the argument keeps returning to, the filter. Moog's 1965 transistor ladder is four buffered RC stages whose differential pairs pass a tanh of the base-emitter voltage difference; discretized, that is a cascade of one-pole IIR sections with the nonlinearity embedded, and "the 'warmth' is determined by the input amplitude" — no coefficients tuned by ear.

This lands on both sides at once, which is why it belongs in the thesis rather than in a footnote. It grants the purists' premise: the character is in the circuit's nonlinearity, which is exactly where analog-vs-digital-pedals put the dirt-effects case (the clipping is the tone). It denies their conclusion: the nonlinearity is derivable from the schematic, so "unmodellable" is not the reason a model might fall short. What is left of the digital shortfall in this paper is specific and small — oversampling is mandatory, and the feedback path's unit delay pulls resonance off cutoff and off the 3 dB-per-stage attenuation, so resonance must be compensated per frequency. Those are engineering costs with known fixes, not an ineffable gap.

Still absent, and now the only thing edge 1 wants: anyone measuring or ABX-testing a ladder circuit against a model of that same circuit. The corpus has the theory of the character and no test of the copy.

A new pole on the analog axis — cheap, not boutique (behringer-cs-mini, 2026-07-23). Every analog instrument sourced so far sat at the premium/artisanal end: filtopia (£1,799, 200 units), cyma-forma-alt (€1,279), or the open-hardware polyuanalog. Behringer's CS Mini — a reported ~$99 3-voice analog synth cloning the spirit of the Yamaha CS-80 (a five-figure vintage machine) — puts analog at the budget end for the first time. That doesn't resolve the analog-vs-digital argument so much as complicate its economics: the "better sine waves" purism has always been entangled with analog being expensive, and a mass-market clone tests whether the aesthetic claim survives being cheap. It also adds a third maker-type to the roster beside boutique (analogue-solutions, cyma-forma) and DIY-kit (korg Nu:Tekt): the mass-market reissue of a recognizable classic, with Behringer's Synthtribe app as a demand-gauge/pre-release pipeline. One honest asterisk, filed on the synthesizer voice model: it's paraphonic, not polyphonic (three oscillators through a shared filter/amp path), the exact compromise that buys "3 voices" cheaply — and a departure from the fully-polyphonic CS-80 it invokes. Pre- release and unconfirmed (specs/$99 leaked via the Synthtribe app), so hold the numbers loosely.

Two days later, the opposite pole — and the axis was never really about price (miltone-4exp, 2026-07-25). Miltone's 4Exp is a part-for-part rebuild of the 1975 Oberheim Four Voice at €9,990–11,990, hand-built and numbered by a Paris shop whose day job is repairing the originals. Set beside the CS Mini it forces a distinction the "cheap vs boutique" framing was hiding: both instruments recreate a five-figure 1970s analog polysynth, and they disagree about what recreation owes the original. Behringer keeps the idea and deletes the price, paying in architecture (paraphonic, shared filter path). Miltone keeps the circuit and the price, and sells fidelity as the product. That is now its own axis — vintage-recreation — running from part-for-part copy through budget clone and architecture homage (ouranos7 and m-vave-fm-1 both rebuilding the DX7 spec, minichord the Omnichord) down to software emulation, where it collapses back into the analog-vs-digital argument.

What the 4Exp adds to that argument is a third reading of "analog": not boutique-new (analogue-solutions, cyma-forma) and not cheap-clone, but archival — the circuit itself is the artifact, and restoration expertise is the manufacturing qualification. It also sharpens a shortage the spoke had not named: the instruments people keep rebuilding are nearly all analog polysynths from 1975–1983, expensive when new and unobtainable now, so recreation is what a design does after it outlives its own manufacturability. Sourcing is weak on this one — dj-mag relays the announcement with no architecture at all (no voice/oscillator/filter/MIDI detail, only "analog"), at T3, a year before delivery.

A smaller cross-instrument motif worth watching: controlled randomness ("spray") and user-defined microtuning both recur independently — spray in gr-mega (pan/pitch/patch) and beast-repeat (glitch scatter); user scales in gr-mega and cyma-forma-alt. Randomized-scatter control and alternative tuning look like shared idioms of this gear, not one-offs.

3. The synth as instrument you don't play by keys

cyma-forma-alt widens the corpus past the note-in/sound-out model. It is a generative/soundscape instrument driven by sensors — a microphone ("ears") and light sensors ("eyes", theremin-style) — so the room, not a keyboard, moves the controls. It also treats tuning as a design surface: 14 selectable scales including just intonation, so microtuning is first-class. Both are new axes the two founding synths didn't have, and they point the spoke toward ambient/generative and alternative-control gear (the Eurorack and "manual studio" neighbors). This is one answer to the recurring who moves the controls question — patched automation (modular-routing), hands-only (music-from-test-equipment), the environment (cyma-forma-alt), or the plain dedicated control surface of a midi-controller (keylab-mk3). The controller category now has two shapes that split on keys vs knobs: keylab-mk3 plays notes and is bundled-software distribution, while OXI E16 is a standalone- premium knob box (16 encoders, no keys) whose knob recording + MIDI looper author and replay control gestures — nudging a "makes no sound" controller toward the sequencer end of the chain (its maker oxi-instruments is a sequencer house). So "who moves the controls" gains a wrinkle: a controller that records how you moved them and plays it back. The controller also introduces a distinct business model: sold bundled with software instruments (Analog Lab / V Collection), the hardware doubles as distribution for a maker's synths — a contrast to the give-it-away free-plugin pattern in thesis #4. (Both the KeyLab source and the controller category are thin so far — one T3 colorway launch — so this is a flagged edge, not a load-bearing claim.)

And the thesis now has its foil (2026-09-01). orphea is a pocket instrument built on the opposite bet: keep the piano keyboard, shrink everything else. Its whole design argument, as designboom puts it, is "how far a conventional keyboard can be reduced while retaining its physical and musical functions" — 13 keys, one octave C→C, octave buttons to move the range without growing the box. That makes it the spoke's first digital-piano, a class defined by the keyboard being the product rather than the interface, and it sharpens what thesis 3 is actually claiming. The chord synths and knob boxes are not avoiding keys because keys are bad; they are avoiding them because a keyboard is the expensive, bulky part, and every one of them spends the space it saves on something else — chords, a looper, 16 encoders. Orphea spends it on nothing else. Whether that is enough of a product is the question its Kickstarter will answer.

4. Free plugins as audience-building

All three software effects are free, for different strategic reasons: audio-damage revives discontinued IP (shinronin) as a no-strings freeware release; next-oxx is a Tokyo label seeding producers; beast-repeat is an individual author (dillon-bastan) giving away an M4L device. The set of audio-effect-plugin formats a release targets is a readable signal of how far it is meant to travel — ShinRonin covers desktop + iPad (VST3/AU/CLAP/AUv3), Next OXX is Windows-VST3-only, and Beast Repeat is Max for Live: host-locked to Ableton in exchange for deep integration (it extends a stock Live device). So the free-effect corner already spans standalone cross-host plugins and host-tied M4L devices.

A darker read of "free" arrives with Xnullx (2026-07): a brand-new developer shipping five free plugins at once (multiband granular, multiband saturation, a beat-slicer, a wavefolder, a granular freeze/smear), which Synth Anatomy suspects are AI-generated from the generic site/UIs and the all-at-once drop — with reader reports of install bugs and a CPU-pegging site. Where thesis #4's free software is a deliberate audience play by a known maker, this is plausibly machine-generated shovelware: quantity over polish, uncertain quality and longevity. It's the free-plugin pattern's shadow side, and a candidate through-line to watch — as generation tooling drops the cost of shipping a plugin toward zero, the free channel may fill with batch releases nobody maintains. (Held as a watch node, xnullx; the five plugins aren't paged individually until a cleaner release earns it.) A small irony: the hub that maintains this wiki is itself LLM-run, so "is this artifact machine-made, and does that change how we judge it?" is a question the spoke now shares with its own tooling.

A third read arrives with remo-de-vico (2026-07-26), and it isn't a strategy at all. He is a theatre-and-cinema composer in Cosenza with an electroacoustic training (Berio thesis, Nottoli at Santa Cecilia, masterclasses with Dhomont), who publishes 13 Max for Live devices on the first of eight pages plus free Pure Data patches remo-de-vico-site. Nothing about the collection reads as audience-building: no branding, no pricing, no documentation of any kind. The devices are byproducts of scoring work, given away because a patch costs nothing to copy. That completes a spectrum in this corner — the studio reviving IP, the label seeding producers, the developer courting attention, the suspected machine-generated batch, and now the working composer emptying his tools folder. "Free" turns out to describe the price and nothing else about the intent; each of the five has a different reason and they're not on the same axis. Worth holding against the xnullx suspicion, too: a large, undocumented, all-at-once drop looked like the tell for machine generation, and here is a human one with the same shape.

5. The free-software synth pole, and synthesis method as the axis of difference

The KVR Developer Challenge 2026 roundup (2026-07-09) adds the spoke's first all-free, all-software synth cluster — five instruments given away at once through the KVR Developer Challenge, the biennial free-plugin contest. It sharpens two existing threads and opens one. First, it extends thesis #4 from effects to instruments: free-as- audience-building now covers synths, and the distribution channel is a named institution (a voting contest with sponsor prizes), not just a lone dev's giveaway. Second, it pushes the digital pole of thesis #2 hard — when the signal path is software and the price is zero, the thing a synth competes on is its synthesis method, and the five entries fan out across the method axis: FM (ouranos7 six-op DX7 homage; tetraop four-op FM/wavetable hybrid), additive (the-hydra, built partial-by-partial), formula-oscillator subtractive (identity-synth), and audio-following pitch tracking (double-freak, a sound source driven by external audio — the same synth↔effect boundary double-freak straddles). That variety seeds two new method nodes (fm-synthesis, additive-synthesis) as siblings to granular-synthesis, turning "synthesis method" into an explicit organizing axis alongside routing and analog-vs-digital. The axis grew a fifth node with wavefolding (2026-07, via Xnullx's Foldspace) — West-Coast waveshaping that adds harmonics by folding rather than filtering them out, and that ships paired with phase modulation, a reminder these methods combine rather than partition. And it's the clean counter-pole to the boutique hardware: where filtopia (£1,799) and cyma-forma-alt (€1,279) differentiate on build, panel, and scarcity, these differentiate on method alone at zero cost — the two economic extremes of the same craft. (Method is independent of price/format: tetraop even ships Linux/LV2. All T2 press-roundup claims — no hands-on measurement.)

A third economic point lands between them — cheap hardware: M-VAVE's FM-1 (2026-07). The boutique↔free-software axis was price and format (expensive hardware vs free software); the FM-1 splits those apart — it's physical hardware at software-adjacent prices (~$75–100: a battery, a keyboard, a sequencer, a speaker). What makes it sharp for thesis #5's "method as the axis" claim is which method: its six-operator, 32-algorithm engine is exactly the Yamaha DX7 spec, so the same DX7 configuration now shows up at both the free-software end (ouranos7, an explicit DX7 homage) and the budget-hardware end. FM's canonical config has become a commodity target, hit cheaply in both formats decades after it defined a price tier. It's also the spoke's first hardware FM synth and first pocket/portable instrument, and its three-way simultaneous MIDI (TRS + USB-C + Bluetooth) plus phone support blur it toward the controller role — a standalone voice that doubles as a portable keyboard for a laptop/phone rig.

The free-software pole now extends to a whole platform: RackDroid/VCV Rack (2026-07-22). Thesis #5's free instances so far were single instruments and effects given away by their developers. VCV Rack is free infrastructure — an open engine hosting an ecosystem of third-party modules — so what's free isn't one synth but the environment other people's synths live in. Against the boutique end (filtopia £1,799, cyma-forma-alt €1,279) the price gap widens from "one instrument vs zero" to "a Eurorack case costing thousands vs a free APK." It pairs with m-vave-fm-1 on a second axis the spoke keeps turning up: that one made FM pocket-sized as cheap hardware, this makes modular pocket-sized as free software, and both point at the phone as somewhere instruments now live rather than just a controller for them.

6. The workstation: integration as the value, not a single trick

Akai's MPC One G2 (2026-07-12) is the spoke's first sampler / groovebox — and it doesn't fit the two organizing axes cleanly, which is the point. Its founding instruments differentiate on how blocks are wired (modular-routing) or what synthesis method they use (synthesizer / fm-synthesis / etc.); the MPC differentiates on integration and workflow — sampling, sequencing, arranging, mixing, and plugin hosting all in one standalone box, off the computer. That's a third framing: the workstation, whose value is that everything lives in one machine rather than one clever sound-shaping idea. Two sub-notes fall out. First, the MPC is a host, not just a sound source — it runs 32 plugin instruments and ships a bundled Native Instruments plugin, so it sits over other makers' instruments the way a computer DAW does. Second, MPC 3.9's headline is a DAW-like linear arranger — a timeline, not just loop/pattern grids — so the standalone-hardware end is migrating toward the computer-DAW workflow it historically avoided. This also answers the "who moves the controls" thread (thesis #3) from a new direction: not sensors or a knob box, but a self-contained sequencer that captures and arranges everything you play. (Vendor-announcement T2 — spec/performance claims are Akai's, no independent measurement.)

7. The DAW: integration in software, and the routing thesis at production scale

Reason 14 (2026-07-17, via the MusicTech review) is the spoke's first proper software DAW — the exact case thesis #6's open question named ("whether Eurorack or a DAW proper lands"). It matters in two ways. First, it completes the integration axis thesis #6 opened: the MPC groovebox is integration in a hardware box, a DAW is integration in software — the same "value is that everything lives in one place" claim, now at both ends of the hardware↔software line. A DAW is the widest single instrument the spoke holds: it contains the synths and effects rather than being one.

Second — and this is why Reason is an unusually clean fit rather than scope-drift — Reason lands the DAW on-thesis with modular-routing instead of against it. Its whole interface is a virtual rack you patch with cables (flip it with Tab), and the Combinator saves a whole device graph as one reusable patch. Where filtopia exposes routing inside an instrument and the NTS-4 is routing with the instrument removed, Reason is routing scaled up to a whole production environment — the biggest instance of "character comes from how blocks are wired" yet. So the DAW node doesn't just widen the domain from instruments/effects to the full studio; it shows the founding thesis surviving the jump to the largest tool.

A third wrinkle sharpens thesis #4's economics from a new angle: Reason ships in two forms, a standalone DAW and a rack plugin (Reason Rack) that runs inside a rival DAW. That dual form is a deliberate feature-pressure valve — Reason can stay a simpler DAW because anything it lacks (deep audio/MIDI editing, video, Atmos) is covered by the host it plugs into. It's the inverse of the MPC's everything-in-one-box bet: rather than absorb every feature, Reason stays narrow and borrows depth from its host. The review measures that trade against Cubase Artist (£273 vs Reason+'s £299) and names the missing depth precisely — VariAudio-class audio editing, VCA faders, notation, video, surround/Atmos — giving the daw node its first depth-ceiling yardstick. (T2 — a single named practitioner's hands-on review, 8/10; device counts and prices are the vendor's.)

A fourth DAW layout, and the one the thesis predicted (2026-09-14). audionodes makes the node graph the DAW itself: the patch canvas is the whole instrument and the timeline is beside it, not around it. That is modular-routing carried past Reason's rack-inside-a-DAW to a DAW that is the rack — vcv-rack with no host. Two things it adds to the daw node's yardsticks: the price line is drawn under the rendering (free is the whole instrument, $9.99/month buys 96 kHz / 7.1 and custom nodes), and it runs in a browser, which the BPB source records without saying how. Alpha, one review, no test — T2 and volatile.

8. The chord synth: an instrument organized by its interface to harmony

Three portable chord synths (2026-07-17) open a new instrument class, chord-synthesizer, and it's a genuinely different kind of category than the ones before it. The spoke's existing classes split on what makes the sound — synthesizer (oscillators), sampler (recorded audio), mixer (nothing, pure routing). A chord synth splits on how you address harmony: dedicated chord buttons plus a strum surface, descended from the Suzuki Omnichord, so a non-pianist can play coherent harmony without keyboard technique. The sound engine underneath can be anything. That makes it the sharpest answer yet to the recurring "who moves the controls" thread (theses #3): here the interface itself encodes music theory, so the answer is "a non-musician, because the instrument knows the chords." It sits nearer the midi-controller (a control surface) than the synthesizer on the wiki's map — except a chord synth makes its own sound.

What keeps this from being a single stray is that the three examples re-run axes the spoke already has, which is strong evidence the category is real and not a gimmick. HiChord ($320) is the workstation corner — chord buttons wrapped in a looper/drum/sampler/sequencer, "almost a mini groovebox," a pocket cousin of the MPC and the integration thesis (#6). Minichord ($169) is the DIY/open-source corner — an open-hardware Omnichord homage via Seeed Studio, a third origin for the build-it-yourself thread after academic polyuanalog and major-maker-kit korg-nts-4, and its USB-MIDI-out tips it toward the midi-controller role. Orchid ORC-1 ($549) is the harmonic-intelligence corner — a voice-leading engine and inversions/extensions, with the synthesis method (VA/FM/reed) demoted to a menu under the harmony surface, plus a Kevin Parker design credit that adds a small artist-endorsement distribution angle next to the maker-led and free/DIY patterns already here. So one source seeds a class and immediately triangulates it across integration, price/DIY, and musicianship — the same three tensions the rest of the corpus runs on. (T2 editorial comparison; specs/prices are the makers'.)

9. The tracker: a category defined by how you enter time

Five trackers (2026-07-26) add tracker, and the class is the second one here that splits on interface rather than on what makes the sound — after the chord synth (#8), which splits on how you address harmony. A tracker splits on how you address time: a vertical text grid, one line per note division, notes and parameter values and effect commands typed in as numbers. Nothing about it constrains the engine, which is why the five picks are a modular synth (SunVox), a DAW (Renoise), two grooveboxes (Dirtywave M8, Polyend Tracker+) and a sample workstation. Two interface-defined classes is a pattern, not a coincidence: the spoke's real taxonomy may be sound source × control surface, with controllers as the degenerate case where the surface makes no sound at all.

What makes the tracker the strongest instance is the causal claim attached to it. The roundup argues the grid's per-line editing made "pitch rolls, volume swells and other quick changes" cheap to type, and that this is why jungle, breakcore and IDM producers took to the form from the Amiga days (attack-mag-5-trackers dates it to Ultimate Soundtracker, 1987). That reverses the usual direction: the spoke has mostly recorded instruments built to serve a sound, and here an interface's editing cost is credited with shaping which gestures became idiomatic in a genre. It also gives thesis #3 ("who moves the controls") a fourth answer alongside patched automation, hands, and the room — the person typing, whose speed is set by what the layout makes cheap.

textbeat (2026-08-10) separates the two things this thesis had been holding together. If a tracker is its interface, then a tracker without an application ought to still be one — and textbeat is exactly that test: the vertical columns survive as a convention about whitespace, vim does the editing, and the grid program is gone. The cheap-to-type gesture the causal claim rests on turns out not to need the grid at all. It needs a text editor, which producers already had.

The same source pushes back on the class's other boundary. Every tracker here spells notes absolutely — a cell is a pitch. textbeat's cells hold scale degrees and chord functions, resolved against the current scale when it plays, so a phrase is stored as a relationship and transposition is a scale change rather than a rewrite. That is the chord synth's move (#8, addressing harmony by function) performed inside the tracker's surface (#9, addressing time by grid). The two interface classes the spoke had kept parallel turn out to be composable, and textbeat is the first instance holding both — theory made editable as text, where hichord and minichord make it playable as buttons. Recorded with the caveat that nothing here was run and the repository has been dormant since March 2023.

The DAW thread (#7) gets a cleaner axis out of the same source. Cubase is tracks-and-mixer, Reason a rack you patch, Renoise a typed grid — three production-capable, plugin-hosting DAWs with nothing in common visually. The Reason review sorted DAWs by feature depth; layout turns out to be the axis that separates them, and depth doesn't see it.

Digital makes the board fungible

hermetic-alchemy-lab adds a business-layer reading of analog-vs-digital the corpus didn't have. Its comparison set — Noise Engineering's Versio/Alia/Legio, Bastl's Kastle 2 — are all named as SDK-based modules that "are also based on the Daisy". Four product lines from separate boutique brands, one commodity Electrosmith board underneath.

In analog, the circuit is the product: filtopia's five-filter parallel bank can't be shipped by anyone else, and polyuanalog publishing its board files is a real act of disclosure. Once the sound source is software, the compute becomes fungible and identity moves up into the panel and the code — industrial design, jack layout, DSP authorship. The trade the spoke has been tracking at the level of sound (analog immediacy vs digital recall) shows up here as a supply-chain fact, and it raises a question worth watching: does Eurorack's digital corner homogenise the way browser engines did, or do panels and firmware carry enough identity to keep it diverse?

Open questions

Growth edges

Ranked; each names the kind of source that would close it (see ../QUALITY.md → Growth edges).

  1. Analog purism, measured. CLOSED 2026-08-08 (research pass) by two AES papers — meyer-moran-adda-loop and digital-filter-audibility — which between them answer the sampling half and disagree with each other, recorded as a tension rather than resolved. Successor: the circuit half is untouched. Both papers test conversion; neither tests whether a discrete ladder filter and its DSP model are discriminable, which is the claim instrument makers actually advance. — needs: a bench comparison (THD, noise, drift) of an analog circuit against its digital emulation, or an AES paper on filter-topology audibility. Hunted 2026-08-12 — a T1 source landed and the edge is re-specified, not closed. moog-ladder-nonlinear-digital (Huovilainen, DAFx-04) derives the Moog ladder from the circuit and locates the "warmth" precisely: the tanh of the transistor differential pair, so that "the 'warmth' is determined by the input amplitude" with no coefficients tuned by ear. That is the mechanism, which the corpus lacked. It is not a measurement: no bench figures, no hardware in the loop, no listening test — validation is analytical. — needs now: a discrimination result specifically, an ABX or bench comparison of a ladder circuit against a model of that same circuit. A second modelling paper does not close it. Hunted again 2026-08-28 (research pass) — nothing ≥ bar, and the second failure says something. The virtual-analog literature is large and it consistently evaluates a model against a measurement or against a recording, not against the circuit in a listening booth: the recent work read this pass (deep regularised RNNs for VA modelling; state-space models of optical compressors) reports MUSHRA-style similarity to reference recordings, which cannot separate the circuit from the converter chain that captured it — the exact confound meyer-moran-adda-loop exists to isolate. So the missing result is not merely unfound: the field's standard evaluation design does not produce it, because modellers are asked to match a signal and nobody is asked to defend the hardware. Two hunts, same outcome. — needs, unchanged: an ABX or bench comparison of a ladder circuit against a model of that same circuit. Not eligible again before 2026-09-11. HALF CLOSED 2026-09-05, and by the absence-claim drain rather than by a hunt. The edge asked for two things and has now got one. moog-vcf-measured (Paschou, Esqueda, Välimäki & Mourjopoulos, APSIPA ASC 2017, T1) is the bench comparison: a real Moog Mother-32 against the derived model, sine sweep for the linear response and 12 Vpp waveforms for the nonlinear one, agreeing "within a few dB in the linear regime" with matching odd-harmonic distortion. It also confirms the diagnosis above from the other side — the paper's own evaluation is spectral, and it never asks a listener anything. 1a. The discrimination result, which is the half that is left. New 2026-09-05, residue of edge 1. Nobody has run an ABX between a ladder circuit and a model of that same circuit, and "a few dB in the linear regime" is not a null result: a few dB around a resonant peak is exactly where the character claim lives, and no paper here draws a threshold. — needs: an ABX or MUSHRA against the hardware, not against a recording of it. The two failed hunts explain why it is scarce and do not excuse it. 1b. A control surface nobody documents. New 2026-09-05, and it is a finding rather than a gap. Measuring the Mother-32 turned up three things its own maker does not publish: the cutoff knob labels do not match the measured cutoff, the resonance knob maps to no scale at all (the authors report it in % of travel and fit their own curve), and there is an undocumented attenuator in front of the filter that had to be matched by ear. The spoke treats instrument specs as first-party fact; here a peer-reviewed measurement of a current product contradicts its front panel. — needs: a second instance, any manufacturer, of a measured control disagreeing with its published labelling.
  2. A purely analog Eurorack module. Both Eurorack instances the corpus holds are digital — VCV in software, hermetic-alchemy-lab a DSP slate on a Daisy. The case where circuit and character genuinely cannot be separated is still unsourced, which is exactly the case question 1 turns on. — needs: a first-party module doc or schematic (T3 allowed).
  3. Does a tracker's hardware talk to the studio? Nothing in the corpus says whether the M8 or Tracker+ carry MIDI or CV, and a self-contained box that cannot be patched is a different object from one that can under the modular-routing thesis. — needs: a first-party spec sheet.
  4. Eurorack as a market, not a price point. One module at $299/$389 is not economics. — needs: a retail or market survey.
  5. The T2 grading of gear press. Resolved 2026-08-04 by the curator, and not in either of the two directions the edge offered. Rather than choosing between T2-by-masthead and T4-by-masthead, the ladder in ../QUALITY.md gained a rule: tier the page by what it rests on. Measured, tested or compared → T2; the maker's specs relayed → T3; filler → T4. Applied here, the 23 T2s became 16 T3 / 6 T2 / 1 T1, and one T4 rose to T3. The floor is now honestly low rather than dishonestly mid, which changes what the spoke needs: see edge 1, unchanged and now the only thing that would actually raise it. Hunted 2026-09-16 — nothing ≥ bar. The ladder-filter literature is all modelling (Huovilainen DAFx 2004; D'Angelo & Välimäki 2013) with no listening panel against the circuit; the perceptual studies found are of crossover and reconstruction filters, which is the half already closed. Not eligible again before 2026-09-30.

Coverage edges (added 2026-08-08, at the curator's request for a wider backlog). These widen what the spoke covers instead of answering an open question above; one ordinary solid source closes any.

  1. The plugin formats themselves. audio-effect-plugin treats plugins as products; VST3, AU, CLAP and AAX are the interfaces that let them exist and none has a page. — needs: the format specifications, plus one developer's account of supporting several.
  2. MIDI 2.0. CLOSED 2026-09-01 (quality cycle) by the standard's own text — midi-2-0-specification (M2-100-U v1.1, T1, free from the MIDI Association and AMEI) read with the UMP specification beside it, and written up as midi-2-0. Successor, new: the spec says what a device may implement and the badge is satisfied by either half of it, so "MIDI 2.0 compatible" on a spec sheet carries no fixed meaning. Nothing in this spoke's gear pages records which half any product implements. — needs: a device's own documentation naming its MIDI-CI or UMP support, or a survey of what shipped, so the label can be graded rather than repeated.
  3. Loudness, as a delivered quantity. CLOSED 2026-08-22. ebu-r128 (T1, the 2023 V5 text) and spotify-loudness-normalization (T3, first-party) give both halves the edge named, and loudness-normalization holds the concept. Successor, new 2026-08-22: every number here is a target, and nobody in the corpus has measured what normalisation does to a catalogue — how much gain reduction real masters receive, or whether release loudness actually fell after the platforms switched. — needs: a measurement over a body of releases, T1/T2, not another platform's published target.
  4. Generated music. The largest change to music production since the DAW has no page anywhere in 99. — needs: a model's technical report or a rights-holder analysis with figures.
  5. What makes the sound in a pocket instrument. Two of these are now paged with no engine described anywhere — md-7-pocket-synthesizer (a store page taking pre-orders) and orphea (a maker's site, a Kickstarter listing and a design-press feature, none of which says sampled, modelled or synthesized). The corpus can describe these objects' enclosures and interfaces and cannot say what class of instrument either is. — needs: a spec sheet, teardown or hands-on that names the engine; and, for the digital-piano class, the sampled-vs-modelled argument the category runs on outside this corpus.

The loudness war ended in a spec, not an argument (quality cycle, 2026-08-22)

This wiki has spent most of its pages on whether a circuit sounds different from a model of that circuit. ebu-r128 arrives from a direction none of that predicted, and it settles a different analog-versus-digital fight completely.

The EBU's own reasoning names the cause: peak normalisation "has led to considerable loudness differences between programmes," those differences are "the cause of the most viewer/listener complaints," and the quasi-peak meter everyone was using "does not reflect the loudness of an audio signal." The industry had been competing on a number that measured the wrong thing. Once ITU-R BS.1770 defined LUFS and R 128 fixed a target, the competition had nothing to win.

spotify-loudness-normalization shows the same mechanism working on records rather than programmes. A master above −14 LUFS is turned down by exactly the amount it was pushed up, with no added distortion, so the limiting is still audible and the loudness it bought is gone. That is a technical answer to a question this spoke usually treats as taste, and it points the other way from the spoke's usual finding: here the digital, standardised, measurable side is the one that stopped a real degradation of recorded sound.

The unresolved part is scale. −23 and −14 LUFS are both published targets and neither is a measurement of what happened to actual releases, which is where growth edge 8's successor now sits.

What the spoke actually rests on (re-tiered 2026-08-04)

The corpus did not change today; the reading of it did. Twenty-three pages had been graded T2 on the strength of their mastheads — Synth Anatomy, Bedroom Producers Blog, Gearnews, Synthtopia, MusicRadar, Attack, MusicTech, AudioTechnology. Applying the new rule in ../QUALITY.md — tier the page by what it rests on — splits them:

The spoke was already inconsistent with itself and nobody had noticed. keylab-mk3 (Mixdown) and miltone-4exp (DJ Mag) were sitting at T3 — the same kind of coverage as a dozen pages at T2. So this pass is less a new scheme than the spoke finally applying its own two exceptions to the rest.

What it costs, honestly. The floor reading goes from T2-heavy and healthy to T3-heavy, which is what it always was. Nothing about any page's content changed and no claim was withdrawn. What changed is that the tier no longer says a press release was independently verified.

What it does not resolve. analog-vs-digital-pedals stays the spoke's lone T4 and is the one page the rule does not cleanly settle: a Guitar World explainer resting on neither testing nor first-party information, just general explanation. Kept at T4 for exactly that reason — and it bears on the spoke's oldest open question (analog-vs-digital) without being able to settle it, which is the whole problem growth edge 1 names.

The stepping objection is a 7-bit artifact, not a digital one (quality cycle, 2026-09-01)

midi-2-0 arrived as coverage, and it lands on the spoke's oldest argument. The analog camp's practical complaint about digital control — that a slow filter sweep zippers, that automation sounds like stairs — is usually filed under analog-vs-digital as evidence about digital audio. It is not. It is evidence about 128 steps, the size of a MIDI 1.0 controller value, and the MIDI 2.0 Protocol raises that to 32 bits with over 32,000 controllers.

That matters for how this wiki reads gear copy. A complaint about stepping is a complaint about a control-data format from 1983, fixable by a protocol revision, and it should not be counted as evidence in the converter argument that meyer-moran-adda-loop and digital-filter-audibility are actually about. Two separate disputes have been sharing one word.

The other thing the spec supplies is a reason to distrust a badge. "MIDI 2.0 compatible" requires MIDI-CI or UMP, not both, plus one feature from that branch — so two compatible devices may share nothing above MIDI 1.0. The spoke's gear pages currently repeat marketing claims about connectivity without recording which half is implemented, which is now a named edge rather than a habit.

Recreating an instrument without imitating it (2026-09-09)

VES arrives as the first source for the bottom rung of vintage-recreation, and it does not fit the rung.

The spectrum this spoke built ran from part-for-part copy (Miltone's €10,000 Four Voice) through budget clone and architecture homage down to "software emulation," and the whole of it was organised by one trade: fidelity against price. Every cheaper rung buys access by giving up voice architecture. The bottom rung was assumed to be DSP modelling, which is the same trade taken to its limit and the reason analog-vs-digital is a founding through-line here.

firmware-emulation does something else. VES embeds a patched MAME and boots the instrument's own ROM on a simulated CPU, so ~45 machines run their original code. For an instrument whose voice was already digital, that is not an imitation of the sound: it is the same computation on different hardware, and the accuracy question the rest of this corner argues about does not arise. The fidelity/price trade stops applying, because nothing is being approximated.

What it pays instead. Three costs, none of them about sound quality:

And the business model is new here too. Autodafe publishes everything under AGPL-3.0, including the MAME patch series and a script that verifies it reproduces against upstream commit f34f02505e32c1993c6a782b6814232cbfc74e36, then sells the compiled builds. Nothing is held back to make that work; the moat is that building a JUCE plug-in around a patched MAME tree on four platforms is tedious. It is also, by some distance, the strongest provenance discipline of any software source in this spoke — and still entirely unmeasured, since no page anywhere compares VES's output to a real CZ-101.

Contradictions / tensions

Cross-spoke adjacency

Index — Music-Tech Wiki

Catalog of every page, grouped by schema.org @type. Spine: synthesis (thesis), log.md (history), this file (catalog). Some wiki-links resolve to neighbor nodes in ../loudspeaker-design-wiki / ../speech-audio-wiki (intentional cross-wiki links).

Product / SoftwareApplication (instruments & plugins)

DefinedTerm (concepts / mechanisms)

ScholarlyArticle (sources)

TechArticle (sources)

Article (sources)

Review (sources)

SoftwareSourceCode (sources)

BlogPosting (sources)

WebSite (sources)

Event

Organization

Person

Synthesis

Neighbor spokes (cross-wiki)

../loudspeaker-design-wiki — sound reproduction (drivers/enclosures/T-S params); this spoke makes the signal, that one plays it back. ../speech-audio-wiki — speech/audio AI models (TTS/STT/audio-gen); a neural audio model routes there, a synth/effect/DSP plugin here.

Related pages

Links to

Linked from