MP3 to MIDI Polyphony: How Many Notes at Once Breaks It
Transcribers blame polyphony and never say how many notes. MIDI puts no ceiling on simultaneous notes, only a 24-voice floor. The count that matters is in the audio, and one note already supplies four pitch classes.
Where the ceiling actually is
- The file has no polyphony limit. MIDI carries 16 channels and a 7-bit note number from 0 to 127; no field counts how many note-ons are active at once.
- 24 is a floor, not a ceiling. General MIDI requires an instrument to sound 24 voices at once — at least 16 melodic and 8 percussive. That is the least a synth must do, and it says nothing about detection.
- One note already supplies four pitch classes. Its first seven harmonics land on the root, the major third, the perfect fifth and the minor seventh — a dominant seventh chord built on the note.
- Count pitch classes, not instruments. Two notes a fifth apart already cover 7 of the 12 pitch classes in the octave.
What the three say
jukeblocks says "Polyphonic or very busy mixes may need note editing after import"; ai-midi warns of "dense mixes"; songscription says "the model has to identify each pitch in a chord". All three name it; none counts it.
The file side: no ceiling, only a floor
A MIDI file addresses 16 channels, and a note-on carries a 7-bit note number, 0 to 127. Nothing counts how many note-ons sound at once. General MIDI points the other way: a GM Level 1 instrument must allow 24 voices to be active simultaneously, at least 16 melodic and 8 percussive — a floor for playback, not a limit on transcription.
The audio side: one note, four pitch classes
A sounding note is a fundamental plus harmonics at whole-number multiples, each a fixed distance above it.
- Harmonic 1 · 1f0.00 semitones — the root
- Harmonic 2 · 2f12.00 — the octave, exact
- Harmonic 3 · 3f19.02 — a perfect fifth, 2 cents sharp
- Harmonic 4 · 4f24.00 — two octaves, exact
- Harmonic 5 · 5f27.86 — a major third, 14 cents flat
- Harmonic 6 · 6f31.02 — a perfect fifth, 2 cents sharp
- Harmonic 7 · 7f33.69 — a minor seventh, 31 cents flat
Read the right column by pitch class and the list is root, major third, perfect fifth, minor seventh: a dominant seventh chord. One note alone hands a detector the pitch classes of a four-note chord. songscription traces octave errors to overtones — "a strong harmonic gets picked up instead of the fundamental" — but never counts the candidates.
Why the partials become ghost notes
MIDI's grid is 100 cents per semitone. Harmonics 2 and 4 are exact octaves and land on real pitches. Harmonics 3 and 6 sit 2 cents sharp, harmonic 5 sits 14 cents flat, harmonic 7 sits 31 cents flat: off the grid, but nearer a real note. A tracker rounding to the nearest pitch snaps each onto a note nobody played — the "short spurious or ghost notes" songscription tells you to delete.
The counting rule
Count pitch classes, not instruments. A note and the note a fifth above it supply the pitch classes at 0, 4, 7 and 10 semitones, and the same set shifted by seven: two notes cover 7 of the 12. So count distinct pitch classes at the densest instant. One melody line supplies four; as the count climbs toward twelve, what is left is what no detector can separate.
What actually lowers the count
Nothing in the file and nothing in the converter settings. The lever is the source: convert one part rather than the mix, since songscription's advice is that "separating the instruments first can help".
Questions
- Does a MIDI file cap how many notes sound at once?
- No. It carries 16 channels and 7-bit note numbers from 0 to 127, and nothing counts simultaneous notes.
The 24-voice and 16-channel requirements are quoted from the General MIDI Level 1 minimal feature list. The semitone values are arithmetic on the harmonic series and equal temperament, against MIDI's 100 cents per semitone.