Dilla Timefeel Analysis
Twelve J Dilla productions, every drum hit placed against the hi-hat's own grid. Press play and watch where the hits land.
Three things are true at once on almost every record above: the hi-hats are straight, the kick and snare push ahead of them, and the kick's pickups hang behind. Then the whole shape repeats, bar after bar, within a few milliseconds of itself. Here are the four measurements, one at a time.
Layer one: the hats are straight
Read each hi-hat stem's swing off its own placements — its off-beat eighths against its beats, or its e's and a's against its eighth positions where it plays sixteenths — and the answer is between 49 and 52% on every track but one. The exception is Runnin', whose sixteenth-note hats swing at 56%. The clock in these records is straight.
Hi-hat swing, read from the separated hi-hat stem's own placements at the subdivision it actually plays. 50% is straight; the dashed line is triplet swing. The control at the bottom is Questlove on Chicken Grease.
every track — this layer's numbers
Layer two: the backbeat leans in
Now place the kick and the snare against that straight hi-hat grid. On Raise It Up, Didn't Cha Know and Devil's Pie both land 9–17 ms ahead. On Fall in Love the kick alone is 19 ms ahead with the snare dead on. On Players the snare sits 45 ms ahead of the hats and the kick 18 — the most displaced backbeat in the set. That's the "drunk" feel the book describes, and it is early, not late.
Kick and snare placement versus the hi-hat's own eighth-note grid, in milliseconds. Left of zero is early. Confirmed hits only.
every track — this layer's numbers
Layer three: the pickups hang back
The kick's pickup sixteenths — the "a" before the beat — land 26 to 52 ms behind the straight grid. Read as an MPC setting against that grid, that's 58–66% swing. This is the layer people are hearing when they say "swing", and it belongs to one voice, on one subdivision, while the hats above it stay straight.
The kick's pickup sixteenths as an MPC-equivalent swing setting against the straight grid. Only tracks whose kick plays pickups are shown.
every track — this layer's numbers
Programmed — and repeated
Here is the part a swing knob can never explain. Loop to loop, the kick and snare land within about one to four milliseconds of themselves — the same sample, in the same displaced place, every time. Questlove playing Chicken Grease — our control, the same feel performed live — wanders by 13 ms and never repeats at any lag. Dilla's feel is a fixed shape, played back exactly. Two records mix the modes inside one beat: on Players the kick loops every four bars to 1.8 ms while the snare never repeats; on Runnin' the snare loops to 1.4 ms while the kick is the same sample every hit but never lands in the same place twice.
Loop-to-loop residual of the kick and snare, in ms (median), at each track's own loop length. Above the dashed line the hits are not locked and the placement numbers above become unreliable.
every track — this layer's numbers
Is there a single number?
No. The closest any of these records comes to a one-knob swing is Common's The Light, whose sixteenth pickups run about 12 ms late — around 54% — with everything else straight. On the rest, one number would have to describe a straight hi-hat, an early backbeat and a late pickup at the same time. It can't. The feel is three layers, all locked, bar after bar, which is exactly why it is so hard to fake with a knob.
Every track
Offsets in ms versus the hi-hat's eighth-note placement (+ = late). Pickups = the kick's off-sixteenths as an MPC-equivalent setting. Loop σ = kick/snare loop-to-loop residual at the loop length L. † = hits not locked (σ > 25 ms), an unreliable placement. The full 16-slot fingerprints and hit counts per part are in the study's repository notes.
How we measured
- Each track was opened in Practice Room and split into six stems on the Mac — Separate Stems. Nothing was uploaded anywhere.
- The drums stem was split again into kick, snare, hi-hat, toms, ride and crash — Split Drum Stems.
- The hi-hat stem is the clock. Its onsets set the tempo (to two decimals — Raise It Up reads 94.89) and the phase of the grid, sixteen slots per bar.
- Every hit on every part was placed against that grid: a signed deviation in milliseconds, + late, − early. A voice-stem hit counts only when the drums stem has an onset within 12 ms of it — "confirmed" — so separation spill can't invent a hit.
- Swing is read from the hi-hat's own placements — its "&"s against its beats, or its e's and a's against its eighth positions where it plays sixteenths — as the swing setting that would put them there; "MPC-equivalent" restates a pickup's lateness the same way. Loop residual is the median distance between a hit and its counterpart one loop earlier.
- The separation's own timing bias was measured before trusting anything: kick, snare and hat stem onsets sit at 0.0 ms median from the drums stem's onsets.
Limits, honestly
These are measurements of recordings — what reaches your ears — not of an MPC's settings or anyone's hands. The grid is the hi-hat's; where a track's hats are too sparse to read, the table says so rather than borrowing a number from the rest of the kit (the badge's composite, which folds the kick's pickups into one stream, is its own column). Placements marked † come from parts that don't repeat cleanly and shouldn't be quoted as numbers. The corpus is the twelve tracks the book leans on plus one control; it is a study, not a census.
Do it to your own records
Practice Room splits any song you drop in the same way — the drums six ways — and its Timing Analyzer draws each voice's hits against the beat; the selection badge prints a part's swing the moment you select a few bars. The readouts on this page — the hats' own placements, the kick's own grid, the bass notes — are the timing lane in the next update.
Split stems on your Mac, slow the record down in tune, loop a bar locked to the real beat, and see the timing of every hit.
Try free for 14 days$29 once · Apple Silicon Mac, macOS 14 or later · nothing uploads
The records
Why each one is in the study, and the section measured. Excerpts on this page are a few bars each, played as the object of the measurement above; the full recordings belong to their artists and labels — go listen to them.