EVIDENCE Behaviour, measurements, limits
Evidence behind the instruments.
Measurements protect musical behaviour: how a filter holds its body, how a glide moves, how a voice changes with sample rate and whether the same session comes back the same way.
From first principles
P01
Behaviour before labels.
Voltage Synths begins with the relationships inside an instrument: where a nonlinear stage sits, how a filter feeds back, which capacitor sets an envelope time and how several voices differ without losing their shared identity. Those relationships become the model, rather than being added later as a general-purpose vintage control.
Musical consequence: a control changes the instrument for a circuit reason, so the useful interactions survive beyond one carefully tuned patch.P02
Different filters remain different instruments.
Substance uses a saturating four-pole transistor ladder with pole-mixed outputs and compensation for low-frequency body. Quartet uses a two-pole OTA state-variable structure that moves continuously from low-pass through notch to high-pass, with band-pass available separately and no structural bass loss as resonance rises.
Musical consequence: the family can share an engineering standard without flattening every instrument into the same oscillator and filter with a new panel.Substance: the original voice
S01
Drive adds colour before the filter shapes it.
The mixer applies a small bias as it saturates, producing even harmonics before the signal enters the ladder. A second output stage adds density after the filter, with gain compensation that keeps the control from becoming only a loudness change. Engine checks confirm that increasing drive changes the harmonic balance.
Musical consequence: clean weight, warm compression and dirty bass occupy different parts of one continuous drive range.S02
Resonance can rise without emptying the bass.
A textbook four-pole ladder loses low-frequency level as resonance increases. Substance compensates part of that loss inside the feedback design, retaining more of the fundamental at low and middle cutoff positions as resonance rises.
Musical consequence: a resonant bass can keep its weight while the filter adds pitch and movement.S03
The note stack and glide behave like part of the voice.
Substance remembers up to eight held notes with last-note priority. Releasing the top note falls back to the previous one legato, without retriggering the envelopes. Glide runs at a constant rate, so time grows with interval, and downward movement is around twenty percent slower than upward movement.
Musical consequence: a bass line can lean on held notes, slides and direction changes without the articulation falling apart.S04
Eleven distinct starting points.
The factory bank moves from Deep Sub and Fat Stack through Dirty Growl, Funk Pluck, Warm Pulse and three lead voices. The patches separate through spectral balance, crest behaviour and attack, while the non-sweeping bass patches keep their energy concentrated below 250 Hz.
Musical consequence: deep basses, percussive plucks and lead sounds all come from the same focused architecture.Quartet: the oscillator specialist
Q01
The oscillator meets its four audible-band targets.
The discrete saw core is checked at 48 kHz with legitimate harmonics removed from the analysis. Free-running saw, hard sync and full-depth pulse-width modulation all sit below their declared audible-band limits.
| Condition | Target | Audible band | Full band |
|---|---|---|---|
| Free saw, 1000.5 Hz | -96.0 | -109.3 | -60.0 |
| Free saw, 4000.5 Hz | -84.0 | -97.8 | -47.8 |
| Hard sync, worst ratio | -66.0 | -104.0 | -59.4 |
| Pulse-width modulation | -90.0 | -124.8 | -124.8 |
The full-band values are shown because they are materially higher in three cases. A short band-limiting kernel has a transition region around Nyquist; later nonlinear processing can fold some of that ultrasonic energy back into the audible band.
Musical consequence: bright notes, sync sweeps and pulse-width movement are controlled where they can be heard, with the ultrasonic limitation stated beside the result.Q02
The filter follows an independent continuous-time reference.
The state-variable filter is probed from the outside with steady-state tones and compared with a separately evaluated continuous-time version of the same topology. It holds within 0.02 dB at the centre frequency and within 0.15 dB across the tested band, at six sample rates and in both current oversampling configurations.
What this establishes: the digital filter conforms to the intended circuit equations. Hardware calibration is a separate development phase and has not yet begun.
Musical consequence: resonance and cutoff stay where the circuit says when a session changes sample rate.Q03
Pitch and envelope timing hold across six rates.
Oscillator pitch is checked from 44.1 to 192 kHz to within 0.005 percent relative error. Attack, decay and release are compared with the resistor-capacitor trajectories they model over time constants from 1 millisecond to 12 seconds; timing holds within 0.5 percent and the sustain transition within 2 percent.
Musical consequence: a pluck stays a pluck, a long release stays long and a reopened project stays in tune.Q04
Eight related voices, not eight identical copies.
Voice allocation, stealing, unison and glide are part of the polyphonic engine rather than wrappers around a mono model. Each voice draws stable component variation from the patch seed and its voice index, giving the stack independent character while preserving repeatability.
Musical consequence: the width and movement of a chord can belong to the patch and return with the session.Engineering and quality
E01
Buffer boundaries cannot change the performance.
The same notes, controls and seed are rendered in one continuous pass and again through changing buffer sizes with events landing inside the buffers. The resulting samples must be identical. That condition includes voice allocation, stealing and control updates, not only the oscillator output.
Practical consequence: changing a DAW buffer or reopening a session cannot quietly alter which voice played a note.E02
The audio path does bounded work.
Processing avoids memory allocation, locks, background threads, panics and library transcendental calls. Nonlinear feedback solves use a fixed amount of work per sample, coefficients are prepared away from processing, and denormal states are flushed.
Practical consequence: the difficult note or the resonant peak does not trigger a surprise burst of work in the middle of a take.E03
The panel and parameters stay aligned.
Quartet's panel and editor share one control definition, including geometry and parameter bindings. The complete interface is checked at multiple scales as the design changes.
Practical consequence: the control you see and the control the engine receives cannot quietly drift apart as the interface changes.Development
L01
MADRE-8 is the planned polyphonic partner to Substance.
The product definition and current interface study establish an eight-voice instrument around Substance's familiar two-oscillator, drive and ladder workflow. The audio engine has not yet been built.
Musical consequence: the design preserves the direct Substance workflow while opening pads, chords, unison and wider performance.L02
Hardware calibration will set Quartet's final voice spread.
Per-voice component spread, warm-up settling and drift depth are provisional while reference-hardware sessions determine their final ranges. The analytic filter result above covers digital conformance, not this calibration phase.
Musical consequence: final voice character must be set by evidence from the circuit being studied, not by a plausible vintage setting.L03
Quartet optimisation continues.
The eight-voice build is currently above its performance target, with the nonlinear filter responsible for most of the cost. Development is focused on reducing that cost while preserving filter conformance.
Practical consequence: an eight-voice instrument needs enough headroom to sit comfortably inside a real session.L04
Ladder begins with a new four-pole branch.
Its interface and control architecture establish the intended SSM2040-class filter and Poly-Mod routing. The audio engine has not yet been built.
Musical consequence: a contrasting filter and modulation structure broadens the family without repeating Quartet.NEXT From evidence to instruments
See what each circuit is built to do.
Compare the musical role, signal path and current development of Substance, MADRE-8, Quartet and Ladder.