How the music is made
When you press play on one of our timelines, a written score turns into sound you can hear — live, in your browser. It can feel like magic. It isn't, and it isn't a mysterious "AI" either. This page explains how it actually works — the general idea, told at whatever level you like. (This is about turning notes into sound; how the notes themselves are chosen is a separate story.)
There's no neural network inventing audio here. What you hear is built from real recordings of real instruments, replayed and shaped by ordinary audio software — the same proven approach behind film-score mock-ups and sampled orchestras. The only "intelligence" is careful engineering.
-
1
Reading the score
Every piece starts as sheet music — the same dots on lines a musician reads. A program reads that notation and turns it into a simple list: which note, and when it happens.
We start from the notation itself. An engraving engine (Verovio) parses the score exactly as you would read it — pitches, rhythms, bar lines — and hands back the note events, plus a timing map that drives the highlight that follows along on screen.
-
2
The sound is a real instrument
The sounds are real. A real cellist, violinist, flautist and so on sat in a studio and recorded every single note, one at a time. We're not faking the tone — we're replaying real recordings.
The timbre comes from real recorded samples — a player recording each pitch (bowed, and now plucked/pizzicato) from the University of Iowa's instrument library. We tune each recording to equal temperament, trim the attack, and match levels so the set is even.
-
3
Filling in the notes between
We didn't record literally every possible note. When one wasn't recorded, the computer takes the nearest recording and nudges its pitch up or down a little — like speeding up or slowing a tape by a hair.
A sampler maps the recordings across the range and pitch-shifts the nearest one to reach any note in between — the same principle as a sampled orchestral library or a keyboard workstation. We space the recordings closely so no note is stretched far.
-
4
Playing it in time
Then it plays the notes in order, in time — like a very precise musician (or a music box) sounding each note at exactly the right moment, at the tempo you pick with the slider.
A transport/sequencer fires each note on its beat, at your chosen tempo, with a little phrase shaping — dynamics that swell and ease, notes overlapping slightly for legato — so the line breathes instead of sounding metronomic.
-
5
Making it sound like a room, not a robot
Finally we add the touches that make it feel alive: a sense of a room (a little echo), the warm wood of the instrument's body, and a gentle wobble (vibrato), like a real player's finger.
We voice it: reverb for space, peaking EQ to restore the body/wood resonance the dry samples lack, a delayed per-note vibrato that blooms in and is register-aware, and legato overlap — the same calls you'd make mixing a sampled mock-up.
-
6
Live, right in your browser
All of this happens the moment you press play, on your own device. Nothing is downloaded as a finished song — it's built fresh every time. And nothing is invented by an AI.
It's synthesised live from the score each time you play — closer to a film-score mock-up rendered in real time than to a recording of a performance. It is not a captured take, and it is not an AI generating audio.
So — is it AI?
No. It's real recordings of real instruments, replayed in time and dressed with a little echo and warmth — put together by careful software. That's the whole trick.
No. It's a sampled instrument played by a sequencer through an effects chain — the everyday technology of studio mock-ups, not a generative model. If it moves you, that's the writing and the real recordings doing their job.
Curious to hear it? Open any instrument on the Music timelines and press play — now you know what's happening under the hood.