DRUM TO MIDI · BROWSER TOOL · RECHECKED 2026-10-30
Drum to MIDI Converter: Turn a Drum Recording Into MIDI
To convert drums to MIDI, a drum transcriber finds the moment of every hit, decides which drum it was, and writes General MIDI drum notes on channel 10: 36 for kick, 38 for snare, 42 for closed hi-hat. The converter below does this in your browser, with no upload. Use a drum loop or stem, not a full song.
Convert a drum recording to MIDI in your browser
Choose a drum loop or drum stem (MP3, WAV, OGG, FLAC or M4A, up to 10 minutes). The tool finds each hit and writes kick (note 36), snare (38) and closed hi-hat (42) on MIDI channel 10, the General MIDI drum channel. Everything runs on this device; the audio is never uploaded.
Raise it if quiet hits (ghost notes) are missing; lower it if you get extra hits.
Only sets the grid your DAW shows; hit times stay where they were played.
Audio is decoded at 44.1 kHz for consistent analysis. Test scores below use the same analysis rate; results on your recording may differ.
| Drum | MIDI note | Hits | First hit |
|---|
How it works: onset detection in three frequency bands plus a spectral-shape check for each hit (our own code, MIT licence: drum-to-midi.js). Toms, cymbals other than the hi-hat, and drums mixed under bass or guitar are not separated; see the test results below.
Published 2026-10-20. The tables below retain the historical programmed-drum tests. Rechecked 2026-10-30: we reran the archived FluidR3 WAV and MP3 samples and reproduced their published scores. The original raw results for TimGM6mb, MuseScore General Lite and the regular-converter comparison were not recovered; those historical numbers have not been reverified in this recovery. The results file identifies the rerun data and evidence gaps.
Current browser check: found hits and extra notes
Rechecked 2026-10-30 by converting the eight downloadable MP3 samples in Chromium and scoring the MIDI files actually downloaded. Analysis used 44.1 kHz; a correct drum class within 50 ms counted once. Results differ slightly from the historical FFmpeg-decoded tests because decoding can change the detected onsets. The drum detector itself is unchanged.
| Drum | Found / true | Extra notes | Recall | Precision |
|---|---|---|---|---|
| Kick | 83 / 83 | 0 | 100.0% | 100.0% |
| Snare | 58 / 58 | 12 | 100.0% | 82.9% |
| Hi-hat | 184 / 184 | 9 | 100.0% | 95.3% |
The mixed bass-and-piano clip found 4/8 kicks, 8/8 snares, 32/32 hi-hats. These are programmed samples related to the detector’s tuning data, not an independent test on real recorded drums. Extra notes still need manual correction; 100% recall here does not mean a perfect transcription.
Check the inputs, per-clip scores and downloaded-MIDI note data in the rerun results.
How does drum to MIDI conversion work?
Drum to MIDI conversion works in two steps: find the hits, then name the drum. A hit is a sudden rise in energy, so the converter watches three frequency bands, where a kick (30–120 Hz), a snare (150–1,000 Hz) and a hi-hat (7–16 kHz) put most of their attack. When a band jumps, it checks the shape of the sound just after the jump: a kick has much more low end than midrange, a snare has a strong midrange and less top, a hi-hat is mostly high frequencies. Each confirmed hit becomes a short note on MIDI channel 10, with a velocity that follows how hard it was hit.
That is different from normal audio to MIDI conversion, which looks for steady pitches and harmonics. A drum hit has no steady pitch, so a melody converter mostly ignores it (see the test below).
How accurate is drum to MIDI? Our test on 3 drum kits
On our 7 drum clips the converter found 79 of 83 kicks, 58 of 58 snares and 184 of 184 hi-hats with the FluidR3 kit, and saving the same audio as a 128 kbps MP3 changed almost nothing (81 kicks, 58 snares, 184 hats). A hit counts as found when a note of the right drum starts within 50 ms of it; each found note is used once.
How we tested: we wrote 7 drum patterns as MIDI files (338 hits, 90–140 BPM, all on channel 10), rendered each one with FluidSynth 2.4.4 through three General MIDI drum kits, and ran the converter's detector (the same file the page loads) on the 44.1 kHz audio. Because the patterns are ours, we know the exact time of every hit and can count what was found, missed and added.
| Drum kit the clips were rendered with | Kick (83) | Snare (58) | Hi-hat (184) |
|---|---|---|---|
| FluidR3_GM (148 MB SoundFont) | 79/83 found, 0 extra | 58/58 found, 12 extra | 184/184 found, 6 extra |
| TimGM6mb (6 MB, the sounds our in-browser player uses) | 83/83 found, 1 extra | 54/58 found, 12 extra | 180/184 found, 28 extra |
| MuseScore General Lite | 67/83 found, 0 extra | 55/58 found, 12 extra | 184/184 found, 3 extra |
| FluidR3_GM, saved as 128 kbps MP3 | 81/83 found, 0 extra | 58/58 found, 12 extra | 184/184 found, 8 extra |
Per clip, so you can see where it fails. Each cell shows FluidR3 · TimGM6mb · MuseScore General Lite as found/true, with extra notes in brackets:
| Clip | BPM | What is in it | Kick | Snare | Hi-hat |
|---|---|---|---|---|---|
| 1-rock-beat | 100 | 8th-note hats, kick on 1 and 3, snare on 2 and 4 | 8/8 · 8/8 · 8/8 | 8/8 · 8/8 · 8/8 | 32/32 · 32/32 · 32/32 |
| 2-kick-snare-only | 90 | syncopated kick and snare, no cymbals | 12/12 · 12/12 · 12/12 | 8/8 · 8/8 · 8/8 | — · +8 extra · — |
| 3-sixteenth-hats | 120 | 16th-note hats with accents | 12/12 · 12/12 · 12/12 | 8/8 · 8/8 · 8/8 | 64/64 · 64/64 · 64/64 |
| 4-hats-only | 100 | hats only, open hat on the "and" of 4 | — · — · — | — · — · — | 32/32 (+1) · 28/32 · 32/32 |
| 5-tom-fill | 110 | 3 bars of beat, then a 16th-note fill on 3 toms and snare, crash | 7/7 · 7/7 · 7/7 | 10/10 (+12) · 10/10 (+12) · 10/10 (+12) | 24/24 (+4) · 24/24 (+4) · 24/24 (+3) |
| 6-ghost-notes | 95 | backbeat at velocity 110 plus ghost snares at velocity 35 | 12/12 · 12/12 (+1) · 12/12 | 20/20 · 20/20 · 20/20 | 32/32 · 32/32 (+12) · 32/32 |
| 7-double-kick | 140 | 16th-note double kick | 28/32 · 32/32 · 16/32 | 4/4 · 0/4 · 1/4 | +1 extra · +4 extra · — |
- Timing: the median timing error was 1.1–7.3 ms per clip and drum, well inside a 16th note at 140 BPM (107 ms).
- Fast double kick is the weak spot: at 140 BPM 16ths the kick was found 28/32 times with FluidR3 and 32/32 with TimGM6mb, but only 16/32 with MuseScore General Lite, whose kick rings into the next hit. The snares in that clip were also missed with two of the kits.
- Toms come out as snare: 12 of the 12 tom hits in the fill were written as extra snare notes, because toms sit in the same frequency range.
- Honest limit of this test: the detector's thresholds were set while testing on FluidR3 and TimGM6mb, so those two results are partly training results; MuseScore General Lite was only tested after, but its drums are close to FluidR3. Real recorded drums with room sound and bleed will score lower. All clips are programmed General MIDI drums.
Can you transcribe drums from a full song?
Only partly: with bass guitar and piano playing over the same beat, the converter still found every snare and hi-hat, but the bass covered kicks. We added a bass line and piano chords to the rock beat (clip 8-beat-with-bass-and-piano, listen) and ran it again:
| Drum kit | Kick · snare · hi-hat |
|---|---|
| FluidR3_GM (148 MB SoundFont) | 4/8 · 8/8 (+2) · 32/32 |
| TimGM6mb (6 MB, the sounds our in-browser player uses) | 8/8 · 8/8 · 32/32 |
| MuseScore General Lite | 5/8 · 8/8 · 32/32 |
For a song, split out the drum stem first with a stem separation tool, then convert the stem; the kick then no longer competes with the bass.
Why does a normal audio to MIDI converter miss drums?
A normal audio to MIDI converter looks for pitched notes, so drums mostly disappear. We ran the same 7 FluidR3 clips through this site's regular converter (the Basic Pitch model that runs on the MP3 to MIDI converter): it returned 26 notes for 338 hits, 3 clips gave no notes at all, only 11 notes started within 50 ms of a real hit, and every note was a pitched note (MIDI 30, 33, 36, 56, 71), mostly from the ringing tail of toms, crash and kick.
| Clip | Real hits | Regular converter notes | Pitches it wrote | Drum converter notes |
|---|---|---|---|---|
| 1-rock-beat | 48 | 2 | 56 | 48 |
| 2-kick-snare-only | 20 | 0 | — | 20 |
| 3-sixteenth-hats | 84 | 0 | — | 84 |
| 4-hats-only | 32 | 8 | 71 | 33 |
| 5-tom-fill | 54 | 7 | 36 | 57 |
| 6-ghost-notes | 64 | 0 | — | 64 |
| 7-double-kick | 36 | 9 | 30, 33, 36 | 33 |
So for drums, use the drum converter on this page; for a melody, bass line or chords, use the regular converter.
Does drum to MIDI keep ghost notes and accents?
On the archived FluidR3 test clip, all 12 ghost snares were found and written softer than the 8 accented backbeats in our 2026-10-30 rerun. The clip plays accents at velocity 110 and ghost notes at 35; the converter measures relative hit strength rather than copying the original velocity. The other-kit rows below retain the historical test results.
| Drum kit | Ghost snares (played at 35) | Accented snares (played at 110) |
|---|---|---|
| FluidR3_GM (148 MB SoundFont) | 12 of 12: velocity 50–60 | 8 of 8: velocity 102–127 |
| TimGM6mb (6 MB, the sounds our in-browser player uses) | 12 of 12: velocity 61–69 | 8 of 8: velocity 84–127 |
| MuseScore General Lite | 12 of 12: velocity 52–63 | 8 of 8: velocity 104–127 |
What MIDI notes do drums use?
In General MIDI, all drums are on MIDI channel 10, and the note number chooses the drum sound rather than a pitch. These are the numbers you will see most often in drum MIDI files, from the General MIDI Level 1 percussion map:
| Note | Drum | Note | Drum |
|---|---|---|---|
| 35 | Acoustic Bass Drum | 45 | Low Tom |
| 36 | Bass Drum 1 (kick) | 46 | Open Hi-Hat |
| 37 | Side Stick | 47 | Low-Mid Tom |
| 38 | Acoustic Snare | 48 | Hi-Mid Tom |
| 39 | Hand Clap | 49 | Crash Cymbal 1 |
| 40 | Electric Snare | 50 | High Tom |
| 41 | Low Floor Tom | 51 | Ride Cymbal 1 |
| 42 | Closed Hi-Hat | 53 | Ride Bell |
| 43 | High Floor Tom | 55 | Splash Cymbal |
| 44 | Pedal Hi-Hat | 57 | Crash Cymbal 2 |
Some drum plugins use their own map. If the kick plays as the wrong sound, change the plugin to a General MIDI map or move the notes.
Free drum MIDI files to download
The 8 drum MIDI files from our test are free to download and use in your own music; we programmed them for this page. Each one is a short General MIDI groove on channel 10, with an MP3 of the FluidR3 render (128 kbps) so you can hear it first. The test is also repeatable: drop an MP3 into the converter above and compare its MIDI with ours.
| File | BPM | Hits | Groove | Download |
|---|---|---|---|---|
| 1-rock-beat | 100 | 48 | 8th-note hats, kick on 1 and 3, snare on 2 and 4 | MIDI · MP3 |
| 2-kick-snare-only | 90 | 20 | syncopated kick and snare, no cymbals | MIDI · MP3 |
| 3-sixteenth-hats | 120 | 84 | 16th-note hats with accents | MIDI · MP3 |
| 4-hats-only | 100 | 32 | hats only, open hat on the "and" of 4 | MIDI · MP3 |
| 5-tom-fill | 110 | 54 | 3 bars of beat, then a 16th-note fill on 3 toms and snare, crash | MIDI · MP3 |
| 6-ghost-notes | 95 | 64 | backbeat at velocity 110 plus ghost snares at velocity 35 | MIDI · MP3 |
| 7-double-kick | 140 | 36 | 16th-note double kick | MIDI · MP3 |
| 8-beat-with-bass-and-piano | 100 | 48 drum hits + bass and piano | clip 1 with a bass line (program 33) and piano chords | MIDI · MP3 |
Test files: ground-truth.json (every hit), results.json (available rerun scores and evidence gaps), gen.py and gen_mix.py (make the clips), score.js (scoring, Node.js), drum-to-midi.js (the detector).
How do I get a better drum transcription?
- Use drums only. A drum stem or a solo drum loop gives the best result; bass and guitar cover kicks and snares.
- Adjust sensitivity once. If quiet ghost notes are missing, raise it; if cymbal wash turns into extra hi-hats, lower it, then convert again.
- Set the tempo. Enter the song's BPM so the MIDI lines up with your DAW's bars and beats. The hits keep their real timing; quantize in the DAW if you want a strict grid.
- Fix toms and cymbals by hand. Move tom hits from note 38 to 45, 47 or 50 and crashes from 42 to 49.
- Check the result by ear. Use the listen button to hear the MIDI next to your recording before you export.
When should you not use this converter?
- When you need toms, ride and crash on their own notes: it writes only kick, snare and closed hi-hat.
- For a full mix you cannot separate: expect missing kicks, as in the mixed-clip test.
- For drum notation: the MIDI can be opened in a notation program, but this page does not draw a drum score.
- For pitched instruments: use the MP3 to MIDI converter, piano to MIDI or guitar to MIDI.
Frequently asked questions
Can I convert drums to MIDI for free?
Yes. The converter at the top of this page turns a drum recording (MP3, WAV and other formats your browser can decode) into a MIDI file with kick, snare and hi-hat notes on the General MIDI drum channel, free and without uploading the file. It found 99% of the kick, snare and hi-hat hits in our FluidR3 test clips.
Why does my audio to MIDI converter not detect drums?
Most audio to MIDI converters, ours included, are built to find pitched notes, not drum hits. On our 7 drum clips (338 hits) the site's normal converter returned 26 notes, 3 clips gave none at all, and the notes came out as pitched notes (MIDI 30, 33, 36, 56, 71) rather than kick, snare and hi-hat.
What MIDI notes are kick, snare and hi-hat?
In General MIDI, drums play on channel 10 and the note number picks the drum: 36 is Bass Drum 1 (kick), 38 Acoustic Snare and 42 Closed Hi-Hat; 46 is Open Hi-Hat and 49 Crash Cymbal 1. The converter writes 36, 38 and 42, so the file plays as drums in any GM player or drum sampler.
Can it transcribe drums from a full song?
Only partly. In our mixed clip with bass and piano, the snare (8 of 8) and hi-hat (32 of 32) were found with every kit, but the bass covered half of the kicks with FluidR3 (4 of 8). Separate the drum stem first with a stem-splitting tool, then convert the stem.
Does it detect toms and cymbals?
No. Toms are written as snare (in our fill test 12 tom hits became extra snare notes) and crashes and rides are not separated from the hi-hat. Move those notes to the right drum in your DAW after converting.
Does drum to MIDI keep ghost notes and accents?
Yes, as velocity. In our ghost-note clip every ghost snare was found and written with a lower velocity than the accented backbeat, so the MIDI keeps the dynamics. Raise the sensitivity slider if very quiet hits are missing.
Is my drum recording uploaded?
No. The file is decoded and analysed by JavaScript in your browser tab, and the MIDI file is created on your device; nothing is sent to a server. The detector source is linked under the converter.
Sources and tools
- MIDI Association, General MIDI Level 1 specification (RP-003): percussion on channel 10 and the percussion key map.
- FluidSynth 2.4.4 (render), FluidR3_GM, TimGM6mb and MuseScore General Lite SoundFonts (drum sounds for the test clips).
- Spotify Basic Pitch, the model behind this site's regular converter (used in the comparison test).
Kit note map: what this converter writes vs where to move notes
The converter writes only three General MIDI drum notes on channel 10: 36 (Bass Drum 1), 38 (Acoustic Snare), 42 (Closed Hi-Hat). Everything else that the detector cannot separate still lands on one of those three. This table is a remapping checklist from those outputs to the rest of a kit — it is not a second copy of the GM excerpt already on the page, and it is not a claim that the detector classified the piece correctly.
| You hear / expect | GM note | Converter writes? | Typical wrong landing on this page | Manual remap to |
|---|---|---|---|---|
| Kick | 36 | Yes | — | Keep 36 (or 35 Acoustic Bass Drum if your sampler prefers it) |
| Snare / backbeat | 38 | Yes | — | Keep 38 (40 Electric Snare only if your kit map needs it) |
| Closed hi-hat | 42 | Yes | — | Keep 42 |
| Open hi-hat | 46 | No | Often 42 | 42 → 46 on sustained open-hat hits |
| Pedal hi-hat | 44 | No | 42 | 42 → 44 for chick-only pedals |
| Low / mid / high tom | 45 / 47 / 50 | No | 38 (fill test: 12 toms → extra snare) | 38 → 45, 47, or 50 by pitch |
| Floor tom | 41 / 43 | No | 38 | 38 → 41 or 43 |
| Crash | 49 / 57 | No | 42 (not separated from hat) | 42 → 49 (or 57) |
| Ride / ride bell | 51 / 53 | No | 42 | 42 → 51 (bell → 53) |
| Splash | 55 | No | 42 | 42 → 55 |
Evidence already published on this page: the tom-fill clip added 12 extra snare notes; crashes and rides are stated as not separated from the hi-hat. Use this map after you listen with the in-page TimGM6mb button — remap by ear, not by assuming every 42 is a closed hat.
False-positive hi-hat vs ride: checklist before you trust note 42
When the MIDI is crowded with 42s, some of them may be ride or crash energy that the three-band detector parked on the hi-hat class. This checklist is a manual review order. It does not add a fourth detector class.
- Count vs arrangement: If you played a ride pattern and almost no hats, but the hit table shows a large Closed hi-hat count and near-zero other cymbal notes, treat the 42 row as “cymbal bucket,” not as proven hats.
- Listen before quantize: Use Listen to the MIDI. A ride that was written as 42 will sound like a closed hat on TimGM6mb — wrong timbre is the cue to remap 42 → 51, not to raise sensitivity.
- Sensitivity direction: Extra hats from cymbal wash usually need a lower sensitivity and a second convert. Missing ghost snares need a higher sensitivity. Do not chase ride separation with the sensitivity slider; the tool has no ride class.
- Stem first: Ride wash over a full kit is worse in a mixed loop. Prefer a drum stem. The mixed bass-and-piano clip already shows non-drum energy changing kick recall; cymbal bleed is the same honesty class.
- Grid is not proof: A 42 that lands on eighth-note ride positions is still a 42 until you remap it. Tempo metadata does not classify the drum.
- When to stop: If you need separate hat / ride / crash tracks as a first-pass output, this converter is the wrong tool — see “When should you not use this converter?” on this page.
Precision on the 2026-10-30 browser recheck was 95.3% for hi-hat (184 found, 9 extra). Those extras are exactly the class of note this checklist is for: found energy, wrong drum label.
Grid-snap residual (ms): what BPM does and does not move
The optional BPM field only writes tempo metadata so your DAW shows bars and beats. Hit times stay where they were played — this tool does not quantize. If you later snap notes to a grid in the DAW, the residual is the milliseconds removed by that snap. That residual is separate from the 50 ms window used when we score a note as “found” in the test tables.
| BPM (MIDI tempo) | 8th-note cell | 16th-note cell | If a hit is 20 ms early vs the 16th grid | If a hit is 40 ms late | vs 50 ms “found” window |
|---|---|---|---|---|---|
| 90 | 333.3 ms | 166.7 ms | residual 20 ms | residual 40 ms | Both still “found” if class matches; snap still moves them |
| 100 (clip 1) | 300.0 ms | 150.0 ms | 20 ms | 40 ms | Same |
| 120 (clip 3) | 250.0 ms | 125.0 ms | 20 ms | 40 ms | 40 ms < 50 ms score window, but audible if you snap hard |
| 140 (clip 7) | 214.3 ms | 107.1 ms | 20 ms | 40 ms | 40 ms is a larger share of a 107 ms cell (~37%) |
Honesty rules used here: (1) leaving BPM blank defaults the file tempo to 120 without moving onsets; (2) scoring tolerance (50 ms) is not a quantize grid; (3) hard 16th-note snap at 140 BPM can erase almost two-fifths of a cell when the hit was 40 ms off — that is a feel change you choose in the DAW, not something the converter already applied. Keep unsnapped MIDI if you want the played timing; snap only after you have remapped 38/42 misclasses.
Related: audio to MIDI converter · WAV to MIDI · MIDI to MP3 and WAV · MIDI converter · audio tools hub