The short answer: when audio to MIDI drops notes, fix them in this order — (1) format, use a lossless WAV or FLAC instead of MP3; (2) mode, monophonic for one line, polyphonic for chords (we measured a chord returning 0 notes in mono and 3 in poly); (3) sensitivity, raise it for quiet or buried parts; (4) shortest note, lower it for fast or staccato playing. Format and mode fix the large majority of dropped-note cases; sensitivity and shortest-note are the fine adjustments. Each step below is measured on this converter’s own code.
1. Format: send it a lossless file first
This is the cheapest fix and the one to try before touching any control. The converter resamples every file to 22 050 Hz and reads it with a 2048-sample window, so the analysis is identical regardless of your file’s own sample rate. What changes with format is the signal quality going in:
- WAV / FLAC / AIFF carry the full harmonic structure the detector scores, with no codec noise.
- MP3 / AAC / OGG are lossy and psychoacoustic — they add low-level spectral artefacts that perturb the harmonic ratios, most visibly in the bass.
We measured this directly in Why MP3 loses notes that WAV keeps: across WAV and MP3 at 128–16 kbps, MP3 did not measurably drop real notes — but it added a few spurious low ones at 16 kbps and left a noisier harmonic stack. If your source is already an MP3 you cannot undo that loss, but you can stop stacking more: re-export from the original lossless file rather than re-encoding an MP3. Start here, because a cleaner file makes every later step read better.
2. Mode: the biggest lever, and the one most people skip
Monophonic mode emits one pitch per frame; polyphonic mode keeps several. That single fact decides most dropped-note cases. We fed a sustained C4–E4–G4 chord plus an A2 bass to the converter’s own code in each mode:
| Mode | Notes returned | Which |
|---|---|---|
| Monophonic | 0 | nothing — the chord has no single fundamental to track |
| Polyphonic | 3 | A2, C4, G4 (missed E4) |
A chord in monophonic mode returned zero notes. That is not a failure — monophonic mode is built to follow one line, so on a chord it picks nothing. Polyphonic mode recovered three of the four. The rule: one melody, a voice, a solo → monophonic; anything with chords or overlap → polyphonic. If you converted a song or a piano part and got a thin result, switch to polyphonic first. If polyphonic gave you a cluttered mess instead of missing notes, the cleaner move is often monophonic on an isolated line. The full trade-off, measured, is in Mono vs poly.
3. Sensitivity: raise it for quiet or buried parts
Sensitivity sets the voice gate. In the converter’s own segmentation code the two thresholds are computed from it directly:
gate = floor + (rmsMax - floor) * (0.32 - 0.30 * sens/100)
salThr = 0.35 - 0.25 * sens/100Both thresholds fall as sensitivity rises. At sensitivity 0 the gate is 0.32 of the dynamic range and the salience floor is 0.35; at 100 the gate is 0.02 and the salience floor is 0.10. So raising sensitivity lets quieter, weaker frames count as notes. We confirmed the limit of this: a clean isolated tone survived at every sensitivity setting (10, 40, 80 all returned the same note count) — sensitivity does not rescue a note that is simply not there. It matters when a note is buried: under reverb, under a louder part, or in a noisy take, the frame energy dips below the gate and a low sensitivity drops it. If your missing notes are consistently the soft ones, nudge sensitivity up before changing anything else.
One caveat from the same measurement: raising sensitivity also keeps more noise, because the same lower gate now admits noisy frames too. If the count goes up but the result looks worse, you have crossed from “recovering quiet notes” into “keeping artefacts” — back it off a little.
4. Shortest note: lower it for fast or staccato playing
The shortest-note setting discards any note shorter than the threshold. It is measured in milliseconds and converted to frames: minFrames = round((minms / 1000) / 0.011610). We fed six notes lasting 20, 35, 50, 65, 80 and 95 ms and changed only this setting:
| Shortest-note setting | Notes returned |
|---|---|
| 60 ms (default) | 6 of 6 |
| 150 ms | 1 of 6 (only the 95 ms note) |
Doubling the shortest-note threshold from 60 to 150 ms took the result from six notes to one. So if your fast passages — a trumpet fill, a plucked bass line, a vocal run — came back thin, your shortest-note is too high. Lower it toward 30–40 ms to keep the quick notes; the cost is that very short noise blips start counting as notes, so don’t set it to zero on a noisy file. This is the last step for a reason: it only affects transient length, not whether a pitch was found at all.
5. The order, as a checklist
Work top to bottom — each step is cheaper and fixes more than the one below it:
- Re-export from a lossless source. WAV/FLAC/AIFF. If you only have an MP3, accept that the loss is baked in and move on. (Format)
- Pick the mode for the material. One line → monophonic. Chords or overlap → polyphonic. This alone recovers most “missing” notes. (Mode)
- Raise sensitivity if the missing notes are the quiet or buried ones — and stop if the output gets noisier instead of fuller. (Sensitivity)
- Lower shortest-note if the missing notes are fast or staccato. (Shortest note)
If you have worked through all four and notes are still missing, the cause is structural, not a setting: the pitch may sit outside the 55–4000 Hz band (a sub-bass or a cymbal — see bass extraction and drum limits), or the source is a drum hit that has no pitch to find at all.
6. How to confirm on your own file
Every number above is reproducible on your audio because the constants are fixed. Convert the file, open the .csv it writes — header index,midi_note,note_name,start_sec,end_sec,duration_sec,velocity — and:
- Sort by note name. If whole regions (all the low notes, or all the high ones) are absent, suspect the band limit, not a control.
- Re-run in the other mode. If polyphonic suddenly returns chords mono missed, mode was your problem — the cheapest fix.
- Check
duration_sec. Notes below your shortest-note setting will not appear at all; lower it and re-run. - Compare a quiet passage at two sensitivities. If raising sensitivity adds the soft notes without adding obvious noise, leave it there.
Questions people actually ask
Why did my chord come back as almost nothing?
Almost certainly monophonic mode. It emits one pitch per frame, so a chord — which has no single fundamental — returns nothing. We measured a four-note chord returning 0 notes in mono and 3 in poly. Switch to polyphonic mode.
Does converting from MP3 lose notes compared with WAV?
Not in our measurements — MP3 did not measurably drop real notes versus WAV at any bitrate from 128 down to 16 kbps. It did add a few spurious low notes at 16 kbps and left a noisier signal. So format is worth fixing for cleanliness, but it is rarely the cause of a large note drop; mode usually is. See Why MP3 loses notes that WAV keeps.
My fast notes disappeared but the slow ones are fine. Which setting?
Shortest note. It discards anything shorter than the threshold; we measured six notes of 20–95 ms dropping from 6 to 1 when the setting went 60→150 ms. Lower it toward 30–40 ms to keep the quick notes.
Will maxing out sensitivity recover everything?
No. A clean tone survived at every sensitivity we tested, because sensitivity only lowers the gate for frames that are already near it — it cannot create a note that has no pitch. It helps buried or quiet notes, and past a point it starts keeping noise instead. Raise it for soft parts, not as a catch-all.
I fixed the settings and notes are still missing. Now what?
Then the pitch is probably outside what this converter reads: below ~55 Hz (sub-bass), above ~4000 Hz (cymbals), or it is a drum hit with no pitch at all. Those are limits of the engine, not settings — Song to MIDI shows what a full mix comes back as, and the bass and drum pages cover the two band edges.
Drop a file in and read the CSV — the Audio to MIDI Converter runs entirely in your browser, so every step above is something you can try on your own take in a few seconds. Once the notes are all there, How to clean up a converted MIDI before you use it walks the passes that turn the raw export into a usable part.
Related guides
- Mono vs polyWhat each detection mode misses
- MP3 to MIDIWhy MP3 loses notes that WAV keeps
- Bass line extractionWhy the converter cannot isolate one voice
- Drum transcription limitsThe 55 Hz floor and 4000 Hz ceiling
- Song to MIDIWhat a full mix comes back as
- Clean up a converted MIDIThe 5 passes that turn a raw export into a usable part