Controls

A channel strip

Velocity and pressure, per partial

Velocity being per partial is what lets a soft note be a different timbre rather than just a quieter one. Set the fundamental to 0% and the upper partials to 100% and the tone opens up as you play harder, which is roughly what a struck string or bar does.

Both controls run either side of zero. A positive amount means harder or heavier is louder. A negative amount inverts that, so the partial is at its loudest when you play softly. The two halves are exact mirrors, so -50% at a given velocity matches +50% at the opposite velocity. The reason to want the negative half is crossfading: give one set of partials a positive amount and another set a negative one, and the two timbres trade places across the velocity range instead of one simply fading in.

Aftertouch works the same way but adds to the fader instead of scaling it, and it ignores velocity entirely. A strip with its fader all the way down is silent until you lean on the key, and then it fades in under your finger, while a negative amount fades an open strip back out. Both channel pressure and polyphonic aftertouch are accepted and whichever is higher wins. Pressure is smoothed over about 15 ms, so seven-bit MIDI does not step the gain.

The mod wheel stands in for it by default, since most keyboards have no aftertouch at all and the wheel is the control your hand already goes to. A wheel left alone reads zero, so nothing changes for a controller that does send pressure. Settings has an "Aftertouch from" entry if you would rather have one or the other on its own.

The two-part envelope

Delay, attack, decay and sustain on the way in. Letting go of a key then runs its own little envelope rather than simply fading out: the KEY OFF rows say where the level goes when the key is released and how long it takes to get there, and only then does the release run from that level down to silence.

Above the sustain that is a release click or a bloom, the sound a damper landing or a hammer returning makes, and it can be louder than the note was while you were holding it. Drawbar Organ uses it that way: the upper drawbars jump to 55% for two milliseconds as the contacts break. Below the sustain it is the fast initial drop into a long tail that a piano or a struck bell actually has, which is probably the half that gets more use. A key-off level of zero skips the stage entirely and releases from wherever the level sat.

LIFT aims release velocity at the key-off level, the same way the velocity row aims key velocity at the fader. Zero ignores the gesture, positive means a faster release is louder, negative inverts it. On a harpsichord the jack falls back harder when you snatch the key away and barely speaks when you let it up slowly, and that is one control rather than a second envelope. A controller that ends notes with velocity zero has said nothing about the gesture, so it gets the middle value rather than the slowest.

The delay stage holds a partial silent before its attack begins. Staggering it across the series makes the spectrum unfold rather than arrive all at once, which is how Slow Pad and Shimmer open up. It is latched in samples at note-on, so moving the knob cannot retime a note already waiting.

One shape worth knowing, because it is what a music box or a thumb piano has: no sustain at all, so the partial decays to silence while the key is still down, and then a key-off level that brings it back. Reaching zero is not the same as being finished.

Ganging the channels

LINK in the top bar gangs the strips: dragging any knob moves the same knob on the others. It works relatively, applying an offset to wherever each strip already sits rather than dragging everything to one shared value, so a spectrum you have shaped by hand keeps its shape. The offset is always measured from the values captured when the drag started, so returning the knob to where you began restores the strips exactly, even if some of them hit an end stop along the way.

The button opens a menu rather than toggling, and the same menu is on a right-click anywhere in the mixer, which is where you are when you want it. With LINK engaged, pointing at a knob arms every knob a drag from it would move, before you touch anything, and how brightly each one lights follows the curve.

Taper is anchored on the strip you grabbed, which always gets exactly its full share of the drag. That has to be true, since the knob under the mouse follows the mouse. From there a strip's share falls off in a straight line with distance, reaching nothing thirty-one channels away.

Where you reach in is therefore the whole control. Grab channel 1 and the mixer tilts down across its full width, from everything to nothing. Grab channel 32 and it tilts up instead. Grab channel 10 and the curve peaks under your hand and falls away on both sides, to 0.71 at channel 1 and 0.29 at channel 32. Distance is measured against the full width rather than against whichever end is nearer, so the reach stays the same wherever you grab: scaling it to the nearer end would make a grab on channel 2 drop channel 1 to nothing in a single step.

LINK gangs the 32 partials only. The noise channel has no tuning, pitch modulation or drift to gang.

Reading the mixer

The two readouts

Each channel carries two figures, the cents its tuning knob is worth and the level its fader is at, both drawn as seven-segment displays rather than as text: a number the instrument is reporting back should look different from a number you typed. Two of the things they show are not numbers. A partial left in equal temperament reads Et and a fader all the way down reads -inF, both dimmed, because a statement of fact is not a level. The fundamental and the octaves read a dimmed 0.0, since their just interval is the note itself.

Lamps on the rules

The rules dividing a strip into groups carry a lamp each, showing what the group under them is doing to this partial right now.

They read from the voice pool rather than from the knobs, so they describe a note rather than a setting and nothing pulses over silence, and they follow the loudest voice on that partial, which is the one the meter follows. The needle's scale is fixed and the same on every strip, so two channels can be compared by eye. Full deflection is 225 cents, the widest the two controls can produce together.

Meters

Each channel meters what that partial is actually putting out, on a decibel scale floored at -48 dB, so the envelope, tremolo, velocity, aftertouch and mute or solo all show at once and the spectrum can be watched evolving as a note decays. The meter fills the fader's own track rather than sitting beside it, and the cap is drawn as translucent glass so the meter reads through it. It reads the loudest instance of a partial across the sounding voices rather than the sum, so it shows the shape of the patch instead of pinning itself the moment you play a chord.

Colour and hover

Channels are coloured by interval class, in chromatic order along a narrow band from blue at the octave, through magenta and red, to yellow at the major seventh, so octaves, fifths and thirds stay visible while you scroll. The band is deliberately narrow so 32 channels read as one family rather than a rainbow, and nothing in it enters the green and cyan the global controls use. Pointing at a knob picks out its row all the way across the mixer and brackets the channel it is on, so where the two cross is the control a drag would move.

Folding the mixer down

Clicking a section heading in the caption gutter folds that group of rows away across every channel at once, and the window loses exactly the height those rows were taking. Pitch modulation, envelope, key off, amp mod and output each fold. The tuning at the top and the faders at the bottom always stay. The heading keeps its activity lamp while folded, so a group you cannot see still reports whether it is doing anything, and the fold state is remembered with the session rather than with the patch.

Channel outputs

Off by default. Overtonium is an ordinary stereo instrument until a host asks for more, and then every channel gets an output of its own: thirty-two partials and the noise channel, alongside the stereo mix that is always there. Any subset works, so wanting two partials out and the rest in the mix is a reasonable thing to ask for.

Through the VST3. The Audio Unit declares the same outputs, but Logic offers no multi-output option for it and loads it as a plain stereo instrument. Logic reads VST3s, so the routing is still available on macOS, and LV2 and the standalone are unaffected.

Each one carries one signal and is dry. One signal because a channel's own output has no stereo image to arrange, and pan is an arrangement inside the mix. Dry because it is taken after the envelope, tremolo, velocity, pressure and fader but before the wobble, the echo, the reverb and the converter, which is the point of routing a channel out: you treat it yourself rather than inherit effects somebody else chose. They do not sum back to the mix, and are not meant to.

The noise channel

Pinned on the right, after the series it does not belong to, is a noise channel marked NZ. It has the same envelope, tremolo, velocity, aftertouch, pan, mute, solo, level and meter as a partial, and takes part in solo alongside them. What it does not have is pitch, so the tuning, pitch modulation and drift rows are empty.

COLOUR takes the tuning row instead, tilting the noise from dark rumble at one end, through flat in the middle, to bright hiss at the other. The middle really is flat rather than merely filtered less. Each voice runs its own noise stream, so playing a chord does not layer eight copies of the identical signal.

MPE

Off by default, and under Settings. With it on, a controller that gives every note its own MIDI channel can bend and press each note separately, which on this instrument means each finger gets its own copy of all 33 aftertouch destinations. Press into one note in a chord and only that note's upper partials come in.

Slide, the forward and back axis under a finger, has a destination of its own under Settings. By default it moves that note's brightness, thinning or opening the top of the series, which is what a player's hands expect from that axis. It can be aimed at the tuning instead, so that pushing one finger forward slides that note alone between equal temperament and just intonation while the rest of the chord stays where it is. The middle of the travel is the rest position, which is also where a controller with no slide axis leaves it, so nothing changes for a keyboard that never sends it.

It is a setting rather than something permanently on because it changes what a channel number means. With it off, a channel is ignored: a key-up on channel 7 releases a note that went down on channel 1, which is what a single-channel keyboard needs. With it on, the channel is part of who the note is, so the same key can be held twice on two channels, bent in two directions, as two voices rather than one retriggering the other.

Across the whole instrument

Wobble

A warped record under the whole instrument. Pitch is bent by reading the output back through a delay line whose length keeps moving, which is what happens when a platter runs eccentric or a capstan slips. A slow warp near once round a 33 rpm disc, a faster wander on top, and now and then a sharp slip that bends hard and settles back. The slips grow in both rate and size with the square of the control, so a low setting is a tired turntable and a high one is a transport falling over itself. It sits ahead of the echo, so the repeats inherit what it did rather than wobbling on their own.

Tape echo

Modelled on the parts of a tape loop the ear notices. Moving the time knob winds the tape rather than cutting to the new setting, so the repeats slide in pitch on the way. Stereo is two tape paths side by side, each fed its own channel and coming back on the side it went out on, with motors running at slightly different speeds so the repeat comes back doubled rather than bounced. AGE never quite reaches zero, because a transport that held speed exactly would put the repeat back in mono and there is no such transport.

Reverb

A feedback delay network whose room is sized from its decay time, since a long tail in a small room is a spring rather than a place.

The converter

Shown as two seven-segment readouts under the output meter. Turning the sample rate down is a genuine cut rather than a filter over the top: the whole voice pool renders slower, which costs a fifth as much at 8 kHz, and the partials above the new ceiling fold back down instead of disappearing.

The output meter

Horizontal, split into left and right, because the two channels are identical until something is panned off centre and a summed meter would hide precisely that. It reads the finished output after master gain and the clipper. Lamps run from teal through amber to orange as they approach full scale, anchored to their decibel positions rather than stretching with the level, with a scale beneath marked at -48, -36, -24, -12, -6 and 0 dB. Each bar carries a peak hold for a couple of seconds.