@audio/midi

Symbolic & timed formats — SMF parse/write/render, MusicXML + ABC notation, subtitles. Roundtrip-tested.

Package	Status	What
@audio/midi-parse	✔	SMF → {tracks, tempos, notes, duration}
@audio/midi-write	✔	notes → SMF format-0 bytes
@audio/midi-render	✔	SMF → PCM through the @audio/synth-* atoms — a GM Level 1 map (128 programs + percussion), no soundfont, no samples
@audio/musicxml	✔	notes → MusicXML 4.0 (quantize, ties, chords, voices, key spelling, percussion) + parser; ABC 2.1 writer
@audio/subtitle	✔	SRT / WebVTT / LRC (enhanced word timing) / TTML parse + write; fromWords for ASR/alignment output
@audio/midi-soundfont	deferred	render via SF2 or midi-js-soundfonts banks — needs an asset-strategy decision first (real SF2 engine vs. ~100 MB/instrument pre-rendered banks, neither belongs in an npm atom as-is). midi-render is the asset-free answer.

import { parse, write, render, musicxml, subtitle } from '@audio/midi'

let smf = write(notes, { bpm: 120 })          // notes → SMF bytes
let { notes, tempos, duration } = parse(smf)  // SMF bytes → notes
let { channelData, sampleRate } = render(smf) // SMF bytes → PCM, GM synth map
let xml = musicxml(notes, { bpm: 120, key: 'G' })      // notes → sheet music
let srt = subtitle.write(cues, 'srt')                  // timed text

Render through any synth voice of your own choosing:

import { parse } from '@audio/midi'
import poly from '@audio/synth-poly'

let audio = poly(parse(smf).notes, { voice: pluck })

Consumers: singing-to-midi tool (site todo), signal/strudel-style editors, @audio/note for pitch math.