A polyphonic synthesizer
We took a beloved classic and threw it into a shredder.
We cloned a legend down to the exact circuit: eight voices, each one the whole of a 1974 American expander module, solved part by part from its schematic. Then we added a bunch of weird stuff that made it better, and over all eight voices a peak-trough cascade filter, five notches and five peaks in stereo.
VST®3 instrument · Windows 64-bit No activationmacOS VST3 and Audio Unit are in development.
From the schematic set
Every part from the original schematics is in the circuit, under its own reference designator. The only things left out are the power supply, whose rails are ideal here, and a gate lamp.
Listen
Seven takes from the Eight-Tone, thirty-two seconds each, in stereo. They are here as they were given, at the levels they came at, so some play quieter than others.
And with the cascade
Four takes with the peak-trough cascade on over the voices, thirty-two seconds each. They came in hotter than the seven above, and they are here as they came too.
Stock, and then us
Left of the line, the panel as the circuit came: the new stages dashed, bypassed, and the cascade off. Right of it, all four turned up, the cascade on and five matrix slots set. Turn the four down and each voice is the stock circuit, but for a shift of a percent or two at the very top of the filter's range, where FIZZ copies its current.
First, the clone
The source is the module's full schematic set, 8 sheets of it, and every original part keeps its reference designator from those sheets. Nothing was drawn as a curve and fitted. Each voice is the circuit, solved.
One voice holds two oscillators with their exponential converters and reset comparators, two envelopes, an LFO, the 12 dB state-variable filter on its transconductance integrators, the notch blend and the VCA: 74 transistors and 310 unknowns, all of them solved together at every step, so the filter hears the oscillators exactly as they are and the VCA hears the filter the same way.
There are eight of them. Every voice draws its own part tolerances: seed 1977 is the netlist's own set, any other seed draws a fresh one, and CHARACTER sets how far the parts stray. Each oscillator is tuned to its own circuit, measured off it, and DRIFT lets the tuning and the filters wander a little. Hold a chord and you are hearing eight instruments that nearly agree.
The power supply and its regulator are left out (the rails are ideal), and so is the pot board's gate lamp. The keyboard and the case mixer were never on these sheets, so the notes, the voice allocation and the mix are ours, in code.
| BLOCK | THE PANEL | HERE | THE REFERENCE | |
|---|---|---|---|---|
| VCO 1 and VCO 2 | FREQ centred, the key voltage from -1 to +5 V | 62.68 to 4013.1 Hz | 62.6 to 4007.8 Hz | every point within 0.53 percent |
| FILTER | FREQ 4.0 to 10.0, RES 0 | 72.0, 253.1, 937.3, 3689 Hz | 72, 253, 936, 3700 Hz | the cutoff, four settings |
| FILTER | FREQ 6.0, RES 10.0 | +11.9 dB near 643 Hz | +11.7 dB near 650 Hz | the resonant peak |
| ENV 1 and ENV 2 | 10 and 30 ms into the attack | 2.539 V, 6.404 V | 2.54 V, 6.40 V | after two seconds at rest |
| VOICE LFO | two settings of RATE | 14.18 Hz, 4.338 Hz | 14.2 Hz, 4.34 Hz | both swinging 3.89 V either way |
| WAVE WRAPPER | the window it wraps into | 2.44 V either way | 2.44 V either way | NEW |
| RES FOLDER | the folded resonance | -1.574 to 1.498 V | -1.57 to 1.50 V | NEW |
Then
Four new stages, built as circuits, placed inside the voice and solved with the rest of it. They are not effects after the synth: the wrapper feeds the filter, the folder lives on the filter's resonance, the sub divides an oscillator and the fizz rides into the VCA.
On the oscillators' mix, before the filter
Six differential pairs fold the oscillators' mix back on itself whenever it leaves a 5 V window, so a plain saw comes out with a new edge every time it would have run off the end. DRIVE pushes it from ×1 to ×6, SYM tilts the window, and MIX sets how much of the wrapped wave reaches the filter.
DRIVE · SYM · MIX
And over every voice
After the eight voices are mixed and before OUTPUT, the signal runs through ten filter sections in series, one cascade for each side of the stereo: five troughs that cut a notch, then five peaks that ring. Every one has its own position and its own depth, and the whole cluster moves together: SHIFT slides it up or down the spectrum, SCALE spreads it out, gathers it to one point or turns it over, and AMOUNT takes it from flat to full depth.
A handle on every feature
On the card, T1 to T5 are blue and P1 to P5 red. Drag one across to move it, up and down for its depth, up to 40 dB for a trough and 24 for a peak; the wheel moves it half a decibel, and a double-click puts it back. Q TROUGH and Q PEAK set how narrow they are, every feature stays between 20 Hz and 20 kHz, and SPREAD sets the two sides up to an octave apart, the right one drawn in pink.
AMOUNT · SHIFT · SCALE · SPREAD · DRIVE · Q TROUGH · Q PEAK · OFF or ON
IN THE MATRIX
CAS SHIFT, CAS SCALE, CAS AMOUNT and CAS SPREAD, and every feature's own position, T1 POS to P5 POS. The cascade is one for the whole instrument, so a voice's own sources, its envelopes, its LFO, velocity and the key, reach it from the newest note.
DRIVE
DRIVE sets the level into the cascade and takes the same back out after it. Quiet, the features are as deep as they are drawn. Driven, each section's gain cell starts to compress its peak and fill its notch, as the circuit does.
SOLVED ON A BOUNCE
Set it to NETLIST and an offline render solves the cascade's circuit, 251 unknowns a side. Played live, a model of the same circuit runs within 0.09 dB of the solved one at every feature. Turned off, it leaves the voices exactly as they were.
And around the circuit
The circuit keeps its own LFO and both envelopes, wired as they were. Around it we put three more LFOs, ten digital ones, four noise sources with a sample-and-hold each, a sixteen-slot matrix and a panel for every voice.
LFO 1 to 3, dLFO 1 to 10
Six shapes, SINE, TRI, SAW UP, SAW DN, SQUARE and S&H, from 0.02 to 30 Hz, each FREE or KEY, restarted with every note, and all thirteen shared by the eight voices. LFO 3 does nothing but bend the rates of LFO 1 and LFO 2, three octaves either way at full BEND. The ten dLFOs, digital, each have a PHASE as well, so ten can be set apart, and they start at rates spread from 0.25 to 9.6 Hz. A bounce starts every one of them afresh, so it renders the same every time.
NOISE 1 to 4
Digital white noise, a fresh draw at every tick of its own clock and held until the next: RATE runs the clock from 0.05 Hz to 5 kHz. SLEW glides from one draw to the next instead of stepping, and KEY draws afresh and restarts the clock with every note. The four are separate streams, and a bounce starts them afresh too, so even the noise renders the same every time.
MATRIX
Eight slots show at once; the box scrolls to the rest, and a line under it counts the ones in use out of view. Each destination enters the circuit where a hand or a patch cable would: PITCH and CUTOFF as voltages into the circuit's own control inputs, the rest by moving a pot along its track, and LEVEL and PAN in the mixer. In a bounce the pots' resistors move inside the solved circuit.
PANEL: SHARED OR PER-VOICE
SHARED plays all eight voices from one panel. PER-VOICE gives each voice a full panel of its own, kept in the preset either way: click a voice in the VOICES strip and the panel edits that one. Eight voices can be eight patches.
SOURCES
DESTINATIONS (red: a new stage's)
The keyboard
VOICE MODE deals notes five ways. ROTATE goes round the free voices in turn; REASSIGN sends a repeated note back to the voice that played it; LOW takes the lowest-numbered voice that is free; UNISON stacks every voice in play on one note, DETUNE spreading them, with PRIORITY LAST, LOW or HIGH and LEGATO if you like; SPLIT puts voices 1 to 4 below a key and 5 to 8 above it. When every voice is busy, the quietest released one is taken first, and if all are held, the oldest note. VOICES sets how many of the eight are in play.
It answers the pitch bend, the sustain pedal, the mod wheel (vibrato), channel pressure and CC74, and velocity, aftertouch and CC74 can each open the filter. KBD TRK sets how far the filter follows the keyboard, and VEL>VCF how far velocity pushes it.
Along the bottom of the panel, each voice shows its note, its level and its pan, with a lamp while it sounds. Click one to edit it in PER-VOICE; the wheel sets its level, and shift with the wheel its pan.
Playback and bounce
Eight solved circuits cannot keep up with your hands, so the Eight-Tone keeps two engines and tells you which one you are hearing. BOUNCE, in the header, opens a menu of the voice's 9 blocks and the cascade, each LIVE or NETLIST, and under its name the panel says what an offline render will be:
renders offline: 3 blocks solved, the rest live
PLAYBACK
While you play, each voice runs a real-time model of the same circuit, every stage's law worked from the netlist's own resistors and measured levels, with the four new stages modeled from their circuits too. Against the circuit on one voice it is a little brighter, by 0.31 dB.
BOUNCE: NETLIST
When your DAW renders offline, the Eight-Tone solves the blocks the BOUNCE menu sets to NETLIST in every voice that sounds, the voices in parallel, and runs the rest as the live model beside them, with your automation moving their parts as it goes. By default the wave wrapper, the filter and the resonance folder are solved. Both engines have the same delay, 29 samples at 44.1 and 48 kHz, so a bounce lines up with playback to the sample.
| BOUNCE | A BOUNCE RUNS | HOW LONG IT TAKES |
|---|---|---|
| ALL LIVE | The live model, exactly as it plays. | Faster than the song |
| DEFAULT, 3 NETLIST | The wave wrapper, the filter and the resonance folder solved, the rest live. | 3.1 to 9.0 times slower than real time |
| ALL NETLIST | The netlist of every voice that sounds, new stages and all, and the cascade. | 11.2 to 19.9 times slower than real time |
The panel
The panel is the signal path. Click a block and its card opens below it, with its controls and a picture of what it is doing; MODULATION sits beside it, with VOICE, LFOs, dLFOs, NOISE and MATRIX tabs. Blocks in red are the new stages, dashed while they are bypassed.
VCO 1, VCO 2
FREQ, two octaves either way in semitones; MOD and its SOURCE, LFO, EXT or an envelope; PW, the pulse's width. SYNC locks VCO 2 to VCO 1.
MIX
VCO 1 and VCO 2 each from pulse through off to saw, and EXT, the external input's level.
WAVE WRAPPERNEW
DRIVE, ×1 to ×6; SYM; MIX, dry to fully wrapped.
SUB OSCNEW
LEVEL, and the OCTAVE keys, -1 OCT or -2 OCT, under VCO 1.
FILTER
FREQ, RES, MOD and its SOURCE, LFO, EXT or ENV 2; OUTPUT, BP or NOTCH; and NOTCH, which blends the lowpass into the highpass.
RES FOLDERNEW
DRIVE, ×1 to ×9.3; MIX; PARALLEL beside the main path, or IN-LOOP inside the filter's damping.
FIZZNEW
FIZZ, DRIVE and SYM; DRIVE CURRENT, FIXED or ENV 2; BAND, CUTOFF or +1.5 OCT.
VCA
ENV 1, or ON for a drone.
CASCADENEW
OFF or ON; AMOUNT, SHIFT, SCALE, SPREAD, DRIVE, Q TROUGH and Q PEAK; a handle for each of T1 to T5 and P1 to P5 on its response.
ENV 1, ENV 2
A, D/R and S, and GATE or TRIG. ENV 1 opens the VCA; ENV 2 sweeps the filter and VCO 2.
VOICE LFO
The circuit's own LFO and its RATE, one in every voice.
LFO 1, LFO 2, LFO 3NEW
SINE, TRI, SAW UP, SAW DN, SQUARE or S&H, a RATE from 0.02 to 30 Hz, FREE or KEY. LFO 3 has BEND, which moves the other two's rates.
dLFO 1 to dLFO 10NEW
Digital LFOs: SHAPE, the same six as LFO 1 to LFO 3; RATE, 0.02 to 30 Hz; PHASE, 0 to 360 degrees; FREE or KEY.
NOISE 1 to NOISE 4NEW
White noise, a fresh draw at every tick of RATE, 0.05 Hz to 5 kHz, held until the next; SLEW, from steps to a glide; FREE or KEY.
MATRIXNEW
Sixteen slots, eight in view and a scroll bar for the rest, each a source, a destination and an amount either way: click a source or a destination for its list, drag an amount, double-click for zero.
VOICES
VOICE MODE, PRIORITY, LEGATO and how many VOICES; DETUNE, GLIDE, BEND, KBD TRK, VEL>VCF, DRIFT and CHARACTER.
BOUNCE, PANEL, OUTPUT
In the header: BOUNCE, a menu that sets each block LIVE or NETLIST for an offline render; PANEL, SHARED or PER-VOICE; and the OUTPUT level.
What it is, and what it is not
SOLVED
Each voice's whole schematic: both oscillators and their converters, the envelopes, the LFO, the filter and its integrators, the notch and the VCA, with the four new stages in the same circuit, and the cascade's own deck. On a bounce the blocks set to NETLIST are solved oversampled, four times over at 44.1 and 48 kHz.
ESTIMATED
The discrete transistors and diodes use close stand-in models from published data for parts of their class, not measurements of the originals; the op-amps and transconductance amplifiers are behavioral models carrying their datasheets' figures. The live model is a model, and says so.
NOT YET
MPE. SPLIT with a separate panel for each half of the keyboard. And the Mac: it is Windows only for now, with the Mac VST3 and Audio Unit in development.
No activation, no license server, no phone-home
A legend, cloned down to the exact circuit, a bunch of weird stuff that made it better, and a cascade of peaks and troughs over all of it. Its sibling, the Overtime State-Variable Filter, brings the fizz to anything else.
VST®3 instrument · Windows 64-bit No activationmacOS VST3 and Audio Unit are in development.
At a glance
| Format | VST®3 instrument · Windows 64-bit |
|---|---|
| In development | macOS VST3 and Audio Unit |
| Requires | 64-bit Windows 11, in any version Microsoft still supports. When the Mac version ships: macOS 15 or later, on Apple silicon or Intel. |
| Voices | 8 · ROTATE, REASSIGN, LOW, UNISON or SPLIT |
| The circuit | Each voice a 1974 American expander module, every original part kept · 74 transistors |
| New stages | WAVE WRAPPER · RES FOLDER · SUB OSC · FIZZ, all four in the circuit |
| Modulation | The circuit's own LFO and two envelopes · thirteen more LFOs, 0.02 to 30 Hz · four noise S&H sources, 0.05 Hz to 5 kHz · a sixteen-slot matrix, 26 sources, 34 destinations |
| Cascade | Stereo, after the voices · 5 troughs and 5 peaks, each a position and a depth · AMOUNT, SHIFT, SCALE, SPREAD, DRIVE, Q TROUGH, Q PEAK |
| Panels | SHARED, or PER-VOICE: a full panel for each voice |
| Character | The netlist's own part tolerances, or a fresh draw · CHARACTER spread · DRIFT |
| Playback | The live model, its laws worked from the netlist's own parts |
| Bounce | The BOUNCE menu: nine blocks and the cascade, each LIVE or NETLIST · the wrapper, the filter and the folder solved by default |
| Bounce time | 3.1 to 9.0 times slower than real time with the default blocks · 11.2 to 19.9 with every block (eight voices) |
| Latency | 29 samples at 44.1 and 48 kHz, the same in playback and the bounce |
| The panel | 1240 by 780 |
| Validation | 47 of 47 validator tests · host-tested |
| Activation | None. No license server, no phone-home |
| Security updates | Until November 2031 |
| License | License agreement · Facts sheet |
VST is a registered trademark of Steinberg Media Technologies GmbH.
