Epi

A tine piano makes its harmonics in the pickup, not in the metal. That is the whole trick of the instrument, and it is why the little voicing screw on a real one changes its entire character, and why leaning into a note makes it growl rather than simply getting louder.

A sample cannot do that. It has one answer per velocity layer and it gives you that same answer every time you ask. You can stack layers until the library is forty gigabytes and you will still be crossfading between photographs of a sound instead of playing one.

So Epi computes the instrument. A hammer strikes a piece of steel, the steel rings, and a pickup reads its motion. Raise the pickup and the instrument re-voices exactly the way the real one does, because the model is solving the same problem the metal is.

Every note runs its own hammer, its own resonator and its own pickup, all at the same time. Nothing is pooled and nothing is stolen, so a dense chord under the pedal is eighty-eight separate mechanisms doing eighty-eight separate things.

That has consequences you can hear immediately. A repeated note lands on steel that has not stopped moving, so fast repeats build instead of restarting. Hold the pedal down and the rest of the instrument answers underneath — the octave clearly, the twelfth quieter, everything else as a wash below that, at the strengths measured off real instruments rather than at whatever sounded nice.

The sustain pedal is read as a position rather than a switch, so half-pedalling works the way it does on an acoustic: the damper felt compresses, and the useful part of that spans the pedal’s actual travel instead of crushing into the last centimetre.

Hold a chord down silently, strike a note somewhere else, and the strings you are holding answer. It is the reason a pedalled piano sounds like a room rather than a stack of notes, and it is the first thing that goes missing when an instrument is sampled — because a recording of one note cannot know which other notes you are holding.

Every instrument in Epi does it, and each one does it through its own real path. On the tine piano the coupling runs through the harp. On the electric grand it runs through the bridge frame. On the reed piano it is proximity between the bars. On the acoustic grand it is the soundboard, with the whole set of strings opened when the pedal is down.

How strongly a note answers depends on how well its partials line up with the one you struck, and that hierarchy is measured rather than invented: a struck note rings its octave partner about nineteen decibels down, the twelfth twenty-six down, and everything non-coincident thirty-eight down as a wash underneath. On the grand it goes further — the full course resonates, including the coupled prefixes and the slow beating between the slightly detuned strings of a single note.

The coupling only runs one way, and that was a decision rather than a shortcut. A two-way spring between string and body was built, measured, and thrown away: it pumps against the string’s own Q, and on resonators that ring for seconds that is a generator waiting to happen. What is there now is passive by construction.

The nice consequence is that a lot of things nobody programmed simply appear. The boom you hear when the sustain pedal goes down is not a sample of a pedal — it is the sound of eighty-eight strings suddenly being allowed to move. The wash as the pedal lifts, and the soft shh of the dampers landing back on the strings, come out of the same place.

A string on its own is a thin, unconvincing sound. What you actually hear is the string through whatever it is bolted to — a harp, a frame, a soundboard, a wooden case — and that body has its own modes, its own mass and its own losses.

Epi models that as a bench you can change. Eight materials, and a size you can move continuously. Both do what physics says they should: the ladder of body modes scales with stiffness over density divided by size, the modal mass scales with density and the cube of size, and each material brings its own internal loss. Shrink the body and everything moves up and gets tighter; build it out of something lossier and the bloom dies away faster.

It applies to all five instruments — the tine’s harp, the electric grand’s frame, the reed’s frame, the acoustic grand’s soundboard and radiator, and the clav’s case. Leave it alone and you get the stock instrument, bit for bit.

The grand’s board is the elaborate one: a fitted modal soundboard, coupled to the strings through the bridge as a two-port, and above about 1.3 kHz — where a real board stops moving as a coherent thing — it hands over to a statistical radiator instead of pretending the modes still line up. And the electric grand’s famous endless sustain is not a reverb hiding behind the strings. It is the absence of a soundboard: with nothing efficient to radiate into, the string simply has nowhere to spend its energy.

The tine piano, with its magnetic pickups and the stereo panner its amplifier insisted on calling vibrato. The electric grand, with the long singing sustain that made those things worth carrying up stairs. The reed piano, which barks when you lean on it, because the gap in its electrostatic pickup is not symmetrical and the asymmetry is the sound.

Then an acoustic grand, played through a pair of microphones you can actually move around the instrument, with the lid, the soundboard and the distance all doing what they do. And a clav, with the twin bar pickups at their measured spacing, the four tone rockers computed as the RC networks that sit behind them, and the three-semitone pitch drop you hear when the tangent lets go and the dead length rejoins the string.

They are not five presets on one engine. Each one is its own mechanism, with its own way of turning motion into a signal.

This is the part I could not stop building. The panel gets you the classic sounds; the workshops let you take the instrument apart.

Re-cut any note’s steel and it retunes by the beam and string equations. Paint a microtonal scale in by hand, or drop just intonation, Pythagorean, quarter-comma meantone, Werckmeister III, slendro or pelog across the whole bank in one click, rotated to any root. Change the wire gauge and the pitch stays put while the overtones move — fat wire turns the tine bank into gongs and makes the grand bell-like.

Decide how worn the pickups are, with tolerance templates that paint the manufacturing scatter of a well-kept, a worn or a neglected instrument — deterministic, so it is the same instrument every time you open the session. Size a speaker cabinet by its dimensions rather than loading an impulse response of one. Or drag up to five microphones around a top-down view of the grand and hear the image move, with arrival delays at the speed of sound and the lid brightening the open side.

All of it is saved inside the preset. A sound you save is the whole sound, not a panel snapshot with the real work missing.

The tine itself is very nearly a pure sine wave once the strike transient has passed. That is the published finding, and it is the thing that reorganises how you build one of these: every harmonic you associate with a tine piano is made by the pickup’s field, not by the metal. Which is why the pickup height control is not a tone knob. It is the voicing screw, doing the job the voicing screw does.

The hammers strike where the real ones strike. The line runs from 57.15 mm from the end in the bass to 3.175 mm in the treble, straight off the service manual, because where a hammer lands decides which partials it can excite at all.

The reed piano’s bark is a single asymmetry. Its pickup is a capacitor whose gap changes as the reed swings, and the response to closing the gap is not the mirror of opening it. Bypass that one nonlinearity in the model and the bark disappears completely — which is a good test of whether you have modelled the mechanism or decorated around it.

The clav drops about three semitones when the tangent lets go and the yarn-wrapped dead length rejoins the string, which is the noise everyone recognises and almost nobody reproduces. And on the grand, a microphone placed under the soundboard reads the low end inverted, because a soundboard is a dipole — while the open lid throws about three and a half decibels of 2 to 6 kHz back at whatever is sitting on that side.

None of that is expensive to play. A ten-note chord on the acoustic grand with the pedal down costs about a third of one core at 48 kHz, and the absolute worst case — all eighty-eight notes fortissimo with the pedal down — is under half a core. Deterministic, every time.

Epi ships with a document that lists what the models cover, what they do not cover yet, and how every claim in the first list is verified. The gaps are named alongside the physical mechanism behind each one, and they print on every test run, so none of them can quietly disappear between releases.

The constants that come from published measurements say so, and cite the paper. The ones that are calibrations say that too. Every factory sound is rendered and measured before it ships — and not only for level: a hardened-felt preset has to measure its zing, a worn keybed its thump, a parlour soundboard its thinner bottom end.

I would rather tell you where the edges are than let you find them in the middle of a session.