We ran one 4-second clip with a known answer — 12 written notes: a six-note melody, two three-note chords, then two more single notes — through this converter twice, once in monophonic mode and once in polyphonic, with every other setting identical. Monophonic returned 6 notes, a 50.0% recall: it kept every single-note line perfectly and dropped all 6 chord tones. Polyphonic returned 11 of 12 (91.7%): it caught 5 of the 6 chord tones and missed only the highest note of the last chord. The difference is not a preference slider — it is a hard gate: monophonic mode cannot emit two notes in the same moment, so anything played as a chord simply does not come out. Every number below was measured by running this site’s own analyser (dsp.js) in Node on a clip it had never seen.
| metric | monophonic | polyphonic |
|---|---|---|
| notes detected | 6 | 12 |
| matched the written notes | 6 | 11 |
| written notes missed | 6 | 1 |
| extra notes | 0 | 1 |
| recall | 50.0% | 91.7% |
| precision | 100.0% | 91.7% |
| pitch accuracy (of matched notes) | 100.0% | 63.6% |
| onset error, median | −24.8 ms | −70.9 ms |
| processing time | 4 ms | 1 ms |
1. The clip: single notes and chords in one file
The test material is melody-and-chord.wav, 4.0 s of piano, built so that the two modes disagree in an obvious way. It alternates between material a single-line detector should nail and material it physically cannot:
| section | content | written notes |
|---|---|---|
| A | four single notes — C4, E4, G4, A4, 0.30 s each | 4 |
| B1 | C major triad — C4 + E4 + G4 together, held 0.60 s | 3 |
| B2 | F major triad — F4 + A4 + C5 together, held 0.60 s | 3 |
| C | two single notes — G4, C4, 0.30 s each | 2 |
| total | 6 single notes + 6 chord tones | 12 |
Because we wrote it, “correct” is not a judgement call: the answer is 12 notes at known times, and a mode either reproduces them or it does not.
2. What each mode actually detected
Monophonic returned six notes — exactly the six single notes. It did not miss a single one, and it did not invent a single one: 100% precision. Its only failure is that the two chords, six notes in total, produced nothing at all. Polyphonic returned the six single notes plus five of the six chord tones, at the cost of one extra note that does not belong to the written music.
| mode | detected notes |
|---|---|
| Monophonic | C4@0.13 · E4@0.48 · G4@0.82 · A4@1.17 · G4@2.98 · C4@3.32 |
| Polyphonic | C4@0.07 · E4@0.44 · G4@0.78 · A4@1.13 · G4@1.52 · E4@1.53 · C4@1.54 · C5@2.17 · A4@2.23 · F4@2.24 · G4@2.93 · C4@3.29 |
The chord results tell the whole story. The C major triad at B1 came back as three notes — G4@1.52, E4@1.53, C4@1.54 — essentially simultaneous, which monophonic mode cannot represent. The F major triad at B2 came back as F4, A4 and C5, but the C5 was detected 129 ms before its written time, outside the 100 ms window we count as a match — so it scores as one missed note and one extra note, and the pitch-accuracy column drops to 63.6%.
3. The difference is a gate, not a preference
Monophonic mode is not “polyphonic but sloppier”. It is a different rule. In monophonic mode the analyser commits to exactly one pitch per analysis frame; when a frame contains a three-note chord, it must pick one frequency and drop the other two. In polyphonic mode it can emit several pitches per frame, so the chord survives. The result is not a small recall gap — it is the difference between a mode that loses every chord tone and one that keeps almost all of them.
| mode | chord tones written | chord tones found | missed |
|---|---|---|---|
| monophonic | 6 | 0 | 6 |
| polyphonic | 6 | 5 | 1 |
Put the other way round: for material that is genuinely a single line, both modes are perfect on this clip — 6 of 6 single notes either way. The modes only part company where more than one note sounds at once. That is the single fact that decides which switch to flip, and it is worth being precise about because “polyphonic is better” is not true either: polyphonic added one ghost note that monophonic never would.
4. The timing and the extra note
Both modes placed the matched notes slightly early — monophonic by a median 24.8 ms, polyphonic by 70.9 ms — which is within the tolerance that matters for a piano roll. The one thing polyphonic got wrong that monophonic could not: a C5 detected 129 ms early on the last chord, the highest and quietest tone, whose onset the detector pulled from the chord’s attack rather than its written slot. That is the cost of emitting several notes per frame — occasionally the boundary of the weakest one lands where the transient is, not where the note was written.
| written note | written start | detected start | offset | verdict |
|---|---|---|---|---|
| C5 (top of F major triad) | 2.30 s | 2.17 s | −129 ms | outside 100 ms window → miss + 1 extra |
5. Which mode to choose, from the measurements
The practical rule that falls out of this test is simple and matches the numbers:
- One line — a voice, a whistle, a solo — use monophonic. On this clip it was 100% precise on single notes and never invented a ghost. There is no reason to accept polyphonic’s extra note when nothing is sounding at once.
- Anything with chords or overlapping notes — use polyphonic. Monophonic loses every chord tone by construction; polyphonic kept 5 of 6 and the price was one ghost note.
- Mixed material — lean polyphonic, then delete. The melody was safe in both modes; the difference is that only polyphonic keeps the harmony, and its errors (one early note) are easier to fix than monophonic’s total silence on the chords.
This is measured on a synthetic piano clip, not a promise about a real recording. Reverb, room tone and human timing will shift the exact recall figures. But the underlying rule — monophonic emits one pitch per frame and cannot represent a chord — holds for any input.
Questions people actually ask
Is polyphonic always better than monophonic?
No. On a genuinely single line, both modes were perfect — 6 of 6 single notes each — and polyphonic added one ghost note that monophonic did not. Polyphonic is better specifically when more than one note sounds at once, because monophonic physically cannot emit a chord. For a solo, monophonic is the cleaner choice.
Why did monophonic mode miss my chords completely?
Because monophonic mode commits to one pitch per analysis frame. When a frame holds a three-note chord, it must choose one frequency and discard the other two. On our clip it returned zero of the six chord tones while keeping all six single notes perfectly. It is not a bug; it is what the mode is for. If you have already picked the right mode and notes are still missing, A conversion dropped notes — here’s the order to fix it walks the format, sensitivity and shortest-note checks before you give up. If what you actually want is to identify chords from audio — what the engine finds, what it cannot name, and when harmonic suppression eats stacked octaves — the chord identifier page measures that end to end.
How much does polyphonic mode actually recover?
On our 12-note clip, monophonic recalled 50.0% and polyphonic 91.7% — the whole gap is the chords. Polyphonic found 5 of 6 chord tones, missing only the highest, quietest tone, whose onset it placed 129 ms early (outside our 100 ms match window). The single-line notes were identical between the modes.
Does polyphonic mode add wrong notes?
It can. On this clip it added one: a C5 whose timing did not match the written note, a side effect of emitting several pitches per frame and reading the transient as the onset. One ghost note against five recovered chord tones is a trade worth accepting, but it is a real one — polyphonic output expects a quick cleanup pass.
Which mode should I pick for a full song?
For a song, choose polyphonic: real music almost always has overlapping notes, and monophonic will silently drop every chord. Then expect clutter and plan to delete — the trade this clip shows is that polyphonic keeps the harmony and adds the occasional stray note, while monophonic stays clean but throws the harmony away.
Related guides
- Audio to MIDI free: what the top tools return — one fixed clip through the tools that rank.
- How many notes a converter can hold at once — the polyphony ceiling and the harmonic rule.
- Where the note boundaries land — the frame grid and the beat grid, measured.
- What is actually inside a MIDI file — the bytes, offset by offset.
- Convert a file now and read the CSV — every number on this page can be checked against your own export.
- Pitch bend detection — What a glide adds to your MIDI, and when it matters
- Sample rate & MIDI — Whether 44.1k, 48k or 22.05k changes the notes
- Song to MIDI — What a full mix comes back as
- Clean up a converted MIDI — The 5 passes that turn a raw export into a usable part