Drums to MIDI: what a pitch detector does with no pitch to find

The short answer: Drums to MIDI does run — drop a kit in and the converter writes notes out — but a drum hit has no fundamental to lock onto, so the analyser reads its noise as if it were pitched material. The kick comes back as a short bass line in the 73–87 Hz range and a cymbal comes back as hundreds of high tones. What you get is not a drum track; it is a pitched translation of unpitched sound. We measured a full kit on this converter’s own code, and the numbers are below.

1. What this converter is actually built to find

Every mode here is a pitch detector. Before a single note is read, the audio is reduced to one fixed stream, and then the engine hunts for stable fundamentals inside a fixed band:

F_MIN    = 55;     // dsp.js — monophonic floor, about A1
F_MAX    = 1800;   // dsp.js — monophonic ceiling, about A6
TARGET_SR = 22050; // dsp.js, line 6 — every file is resampled to this
HOP      = 256;    // → 11.610 ms frame, about 86 frames per second
MAXP     = 6;      // dsp.js — at most 6 notes per frame
SAL_GATE = 0.20;   // salience floor — a peak below this is dropped

A pitched instrument — a piano, a voice, a bass — gives the detector something to grab: a clear fundamental and a harmonic stack it can score. A drum hit gives it the opposite. A kick is a pitch sweep with a fast decay, a snare is filtered noise, a hi-hat is broadband noise with almost no tonal content at all. There is no single frequency the detector can call “the note.” So it does the only thing it knows: it finds the strongest peaks in each 11.6 ms frame and calls them notes. That is the entire mechanism, and it is the reason drum machine to MIDI and percussion to MIDI land in the same place — none of those sources has a pitch to recover.

2. A full kit, measured piece by piece

We built a synthesised kit at sensitivity 60: a kick of 32 hits sweeping 48 to 120 Hz with an 85 ms decay, 16 snare hits, and 64 hi-hat hits with a 20 ms decay, then 112 hits as the whole kit. Here is what the detector returned.

Two things stand out. First, the polyphonic column is busy but meaningless: the snare alone becomes 156 notes and the hi-hat becomes 360, because each broadband hit throws up many peaks the detector scores as separate tones, most of them up at 2–3 kHz where cymbal energy lives. Second, the monophonic column is the reverse — it keeps only the kick, one low G1 per hit, and throws the snare and hi-hat away entirely because their broadband content never clears the 0.20 salience floor.

3. What those two columns actually mean

Read them as two different kinds of useless, not as two different quality settings. Polyphonic mode keeps most of the kit but turns it into pitched material it never was: the kick reads as a short bass line sitting at 73–87 Hz, right in the register a bass guitar occupies, and the cymbals become hundreds of high notes scattered around G7 and F7. Monophonic mode keeps only the kick and deletes everything else. Neither column is what anyone means by drums to MIDI, and no sensitivity slider fixes it, because the problem is the question — “find the pitch of this drum” — not the answer.

The one thing the converter does usefully with a drum-heavy recording is pick out the low end underneath it. The kick is the only piece that produces a sustained-ish tone near the floor of the detector, so with the right mode you can recover a rough bass line from a kit-heavy mix. That is a side effect of chasing pitch, not a drum-transcription feature.

4. Drum machine to MIDI, percussion to MIDI, and convert drums to MIDI

The same verdict covers the rest of the family. A drum machine to MIDI request hits the same wall: a sampled or synthesized kit is still unpitched hits, so it still comes back as a bass line plus a scatter of high tones. Percussion to MIDI is broader — shakers, tambourines, toms — but every one of them is noise-dominated, so the result is the same kind of pitched translation, not a percussion part. Wherever you type “convert drums to MIDI,” the engine underneath is still the pitch finder in section 1, and a pitch finder has nothing to find.

This is the gap most pages skip. Pages that rank for drums to MIDI present it as if you will get a usable drum track, with a button and a before-and-after. They do not say the converter is looking for pitch where there is none, and they do not show what actually comes back — a bass line and hundreds of stray high notes. The honest version is the one above: the tool runs, it returns notes, and the notes are the wrong kind.

The numbers above are the what; the exact why — how low a kick can fall before it vanishes, why a cymbal’s energy sits above the capture ceiling, and how the 11.61 ms frame grid erases a groove — is measured separately in What drum transcription actually catches (and what it misses).

5. What you can actually use

If you need drum parts as MIDI, the practical routes are the ones a pitch detector was never going to give you:

1. Recover the bass line, not the kit. Run the drum-heavy mix in monophonic mode and keep the kick-derived low notes; treat them as a guide for the bass, not as drums.
2. Program the kit by hand. Use the melodic instruments the converter can actually read, then write the drum pattern in your DAW’s grid — that is where onset-based drum tools live, not in a pitch finder.
3. Use an onset-detection tool. A converter built around transient detection rather than pitch detection is the right shape for percussion; this one is not, and no setting will make it so.

None of that is a workaround for a bug. It is the correct routing for a question this tool was not built to answer.

6. How to check your own file

You do not have to take the table on faith. The detector is the same code for every file, so the pattern above repeats on your audio. Drop a kit recording in and look at the CSV it writes — the header is index,midi_note,note_name,start_sec,end_sec,duration_sec,velocity.

1. Count the notes per hit. A clean snare or hi-hat that returns eight or ten notes for one hit is the 156-or-360 pattern from section 2, not a transcription.
2. Look at the note names. If a kick reads as F2/D2/E2 and a cymbal reads as G7/F7, you are seeing the detector’s pitch guesses, not drum pitches.
3. Switch to monophonic mode. The snare and hi-hat will drop to zero and only the kick will survive — confirming the broadband content never clears the 0.20 salience gate.
4. Watch the velocity column. The kick’s one surviving note lands near velocity 110 on a loud hit (velocity 127 is reached at −2.22 dBFS on this converter), which is how you tell the surviving note is real low-end and not a cymbal artifact.

Questions people actually ask

Can drums to MIDI give me a usable drum track?

No, and not because the conversion failed. This converter is a pitch detector, and a drum hit has no pitch to find, so what comes back is a bass line from the kick plus a scatter of high notes from the cymbals. A usable drum track needs onset detection, which is a different kind of tool. The most you can recover here is the low end of a kit-heavy mix.

Why does my snare turn into so many notes?

A snare is broadband noise with a fast decay. In polyphonic mode the detector scores many spectral peaks in each 11.6 ms frame and writes each as a note, which is why 16 snare hits became 156 notes in our measurement. None of them is “the snare pitch,” because a snare does not have one. Monophonic mode drops all 16, because the noise never clears the 0.20 salience gate.

Should I use mono or polyphonic mode for drums?

For a kit, monophonic mode is the more honest read: it keeps only the kick and discards the snare and hi-hat, so you at least get one real low note per hit instead of hundreds of false highs. Polyphonic mode keeps more of the kit but turns it into pitched garbage. Neither gives you drums; monophonic just lies less.

Why does the kick survive but the cymbals do not?

The kick has a pitch sweep with enough energy near the 55–1800 Hz band that it reads as low notes (F2/D2/E2, or a G1 in monophonic mode). A hi-hat is centered up at 2–3 kHz and is almost pure noise, so in monophonic mode it falls below the salience gate and is discarded; in polyphonic mode it survives but as 360 scattered high tones. The kick is the only piece with anything the detector can call a pitch.

Is a DAW's built-in drum capture any different?

Yes — DAWs like FL Studio or Logic do not transcribe drums through a pitch finder. Their built-in capture and a separate converter each follow a different path, and a DAW’s grid is where drum patterns are meant to be written. We cover that separately on the FL Studio audio to MIDI page; the short version is that drum parts belong in an onset-based workflow, not in a pitch-based one.

Drop a recording in and watch the piano roll — the Audio to MIDI Converter runs entirely in your browser, so every number above is yours to reproduce on your own kit in a few seconds.

Related guides

• Piano to MIDIWhere the keyboard falls outside the detector
• Song to MIDIWhat a full mix comes back as
• FL Studio audio to MIDIWhat every DAW's capture shares, and when to export
• Mono vs polyWhat each detection mode misses
• OGG to MIDIWhat the container actually changes (and what it doesn't)
• Guitar to MIDIWhat a bend does to your notes
• Vocal to MIDIWhy a sung note comes back an octave up
• Voice to MIDIWhat a single-line converter can follow
• Humming to MIDIWhat wordless vocals come back as