How to add swing to trap drums: MIDI 70 ghost tier vs rigid hits

This guide delivers a strict method for adding swing to trap drums.

It uses a 70% ghost-note velocity tier while keeping the kick and snare locked to the grid.

Measure Baseline Timing Drift

This section alone defines the baseline timing measurement protocol. Before any swing offset can be tested, the trap part must be measured against its intended 16th-note grid. To do this, export MIDI or audio onsets and compare each hit to the 16th-note grid, recording absolute timing error in milliseconds. The purpose of this step is not to correct the part, but to establish a concrete starting point for the layers that are candidates for swing, while keeping the low-end untouched by this measurement.

Once those errors are collected, calculate mean absolute deviation separately for hats and ghost notes; do not average kick or snare into the same MAD. This separation is required because the deviation of the swing candidates must be isolated from the anchors.

The reason for this separation is practical, not analogical: the swing candidates are the only layers that will receive a timing offset later, so their baseline drift is the number that matters. Kick and snare are grid anchors under this method and are excluded from the measurement entirely. Keep the two MAD values distinct so any later swing test can be judged against the correct layer's starting point.

If MAD is below 3 ms, treat the part as rigid for testing purposes; this cutoff is a method assumption, not a measured population value. This gate does not declare the part unmusical, it only determines how strict the test must be. A part that already sits this close to grid will require a more careful check later to prove that any applied swing improves groove without smearing transient clarity. A part above this gate will still need the same separation, but has more measurable drift to work from.

Finally, log the two separate MAD values, one for hats and one for ghost notes, with the date of measurement and the source of onsets used. Do not collapse them into one number, and do not pull kick or snare into this log. With these two values recorded, the baseline timing drift is fully established and ready to be held against any swing attempt without guessing at the starting point.

Set Velocity Tiers, Not Rigid Hits

This section alone defines the velocity-tier starting method. For trap hats and ghost layers, the starting ratio is to set primary hat accents to 100 MIDI velocity and ghost or secondary hits to 70 MIDI velocity. This split is meant to create motion without forcing every sixteenth to the same hit strength, and it is only applied to those layers, not to the anchored low-end.

Before rendering, scan for any adjacent sixteenth-note pair with identical velocity. If one is found, reject that pairing and replace the second hit with 70 or 100 before rendering. This scan is a strict filter: a run of equal adjacent hits can read as rigid even if the rest of the pattern has some variation, so the pair must be broken by one of the two tier values.

To measure the spread, compute velocity standard deviation across non-kick steps. If SD is below 8 MIDI units, widen the tier gap by 5 units and recheck. For example, adjust the ghost tier down by 5 from 70 to create a larger gap from the 100 accents, then recompute the velocity standard deviation across those same non-kick steps. This is a numeric check, not a matter of taste, and it must be passed before moving forward with the tier set.

The goal of the tier gap is contrast without over-separating the hat body. The tiers must be distinct enough to register against the 8-MIDI-unit SD floor, but the split should not be pushed so far that the ghost layer loses its connection to the hat motion. If widening the gap past 60/100 starts to make the ghosts sound like a different part rather than a softer layer of the same motion, that is a sign the numeric floor has been met and the gap should not be widened further.

After any adjustment, confirm two things for non-kick steps only: that both 100 and the adjusted ghost value are present, and that the velocity standard deviation is at or above 8 MIDI units. If the pattern cannot form that two-tier set without collapsing the gap, do not render with a single-tier set. The tiering must exist as a measurable split, or it does not meet this starting method.

Compare Swing vs Rigid A/B

This section alone defines the controlled comparison protocol. To prevent any bias from familiarity, both versions must be created from the same part set with no mid-test edits. Render Version R with all hits on-grid at 100 velocity. Render Version S with the swing offset plus 70/100 tiers applied only to the layers meant to carry motion. Both versions must share the exact same processing chain and the exact same monitoring level before any scoring takes place.

Before blind-scoring, verify the playback path is truly identical. Switch between Version R and Version S without touching faders, inserts, or sends. The test must judge the change in motion, not a change in level. If switching reveals an obvious level difference, correct it before moving forward. This check is necessary because a level bump can be mistaken for improved groove.

Blind-score both versions on groove, clarity, and head-nod stability using the same 1-5 scale and the same monitoring level. For each category, assign a score to Version R and a separate score to Version S without knowing which is which. Groove is the pull to move, clarity is the ability to hear each intended transient without blur, and head-nod stability is whether that pull remains consistent across the full loop without forcing it. Write all six scores down at the time of judging, with no mental averaging.

Apply the acceptance check with no exceptions: keep Version S only if it wins by at least 1 point on groove without losing more than 1 point on clarity. If Version S gains groove but loses more than 1 point on clarity, reject it. If Version S gains less than 1 point on groove, reject it even if clarity stays the same or improves. Head-nod stability can be noted, but it cannot override this pairwise condition between groove and clarity.

Finally, only commit to Version S if the blind scores meet that condition. If the labels become known at any point during the test, redo the blind pass from the start. The entire purpose of this A/B is to prove the swing earns its place against the rigid-hit control, not to justify it by preference alone.

Check Low-End Stability

This section alone defines kick and snare anchoring rules. For trap drums, the kick and snare must remain on the grid at fixed velocity unless a deliberate flam is notated. Under no condition should a swing offset be applied to either of them. These two elements are not meant to be pushed by swing, they are meant to hold the rigid pulse that the rest of the part can move around.

With those anchors locked in, the kick-to-snare phase relationship must remain unchanged between the rigid-hit control and the swung version. Any shift in that distance, or any flam introduced between the kick and snare by the swing process, must be flagged immediately. The low-end cannot afford to gain motion at the cost of its own fixed positions, as that is the exact thing that will read as smearing instead of groove.

If the low-end smears despite keeping both the kick and snare on grid with fixed velocity, the correct action is to reduce the swing offset by 2 ms rather than changing kick velocity. This keeps the anchors untouched while trimming the motion applied to the other layers. After each 2 ms reduction, re-run the A/B scoring gate before deciding whether a further reduction is needed.

Verify in Mix Context

This section alone defines final mix-context verification. The first action is to audition both versions under the full mix, not soloed hats. Masking from other elements can change the perceived swing, so a motion that reads clearly in isolation may not read at all once vocals, percussion, or synths are present. The question to answer here is not "does it swing?" but "does it swing in this mix?"

Once the full-mix audition shows a possible improvement, transient sharpness must be measured with intent. Route only the hat bus to a peak meter and compare Version S to the rigid control at matched playback gain. Reject Version S if peak level drops more than 2 dB.

This is not about loudness, but about attack retention. A peak drop beyond 2 dB means the swung hats are losing their attack edge, which can smear the transient enough to undo any gain in groove. Measure the peak difference at matched playback gain so the comparison reflects timing and velocity changes only, not level changes.

With peaks checked, the next gate is timing lock. If any manual nudging was done to help Version S sit in the mix, confirm the final groove still locks to the DAW tempo map after those moves. Do not assume the offset holds just because it was placed before. Scan at least one dense passage of the section and check for any pull against the click reference. A version can pass full-mix feel and peak checks, but still fail this if nudging has caused it to drift from the map.

All three checks must be satisfied together to accept Version S. Full-mix audition, hat-bus peak retention within the allowed drop, and tempo-map lock after nudging form one gate. If any one of them fails, the swing version should not be kept over the rigid control, no matter how appealing it felt at any single point of playback.

Worked 16-Step Example

This section alone provides the concrete worked example and checkpoints. To test swing without disturbing the low-end, take a 16 sixteenth-step grid and apply it strictly to the hat and ghost layers. For those layers, use offsets of 0, +7, 0, +7 ms and velocities of 100, 70, 100, 70 repeating for the full pattern. Keep kick on steps 1 and 9, and snare on steps 5 and 13, with no offset applied to either of them. This isolates the motion to only the layers that can take it, while leaving the anchors untouched.

With this exact setup, there are three checks to run against the rigid-hit control. Checkpoint 1: hat MAD should be about 3.5 ms. This gives a measurable shift without pushing every hat hit to the same late position. Checkpoint 2: velocity SD across 12 non-kick steps should be at least 8. This confirms the 70% ghost-note tier is creating actual motion, not just a uniform drop. Checkpoint 3: peak hat level should stay within 2 dB of the rigid version. This prevents the groove from being faked by a level boost, and keeps the focus on timing and velocity.

These checkpoints are not meant to be eyeballed from a waveform glance. They must be pulled from measured onsets and measured velocities for those specific layers. If any one of them falls outside of its target, the test is no longer valid for this example. The point is to prove the shift exists with numbers, not to chase the largest offset possible.

The cost of the wrong choice is very real. Too much swing smears transients and weakens the trap pocket, making the part feel washed out instead of pushed forward. Too little swing leaves the part rigid and less dynamic, giving none of the motion that was being tested for. By holding to these checkpoints, the swing version can only be accepted if it improves groove without smearing transient clarity, and rejected the moment it crosses that line.

Decision Rules

This section alone states the final if/then decision rules. With the swing offset applied only to the hat and ghost layers, the decision to keep or reject it cannot be based on preference alone. Every action must be measured against the rigid-hit control, with the sole purpose of gaining groove without smearing transient clarity. The gate is strict, because any gain in motion is only valid if the trade-off stays within the allowed loss.

The first rule is the acceptance gate: if groove score improves and clarity loss is 1 point or less, keep the swing version. This is the only path to immediate acceptance. Do not combine partial improvements to bypass this condition, and do not allow a clarity loss of just above 1 point to be treated as equal to it. The margin is small by design, so the measurement must be exact before moving forward.

The second rule is the timing correction: if clarity loss exceeds 1 point, reduce the offset by 2 ms and retest before abandoning swing. This is not a full removal of swing, but a step down in the applied offset. The retest must be done with the same scoring method, and only after that retest can the first rule be applied to the corrected result. This correction is meant to find a lower point of motion without stripping the offset entirely.

The third rule is the velocity correction: if velocity SD remains below 8 after tiering, widen the gap to 60/100 and recheck. This targets motion in the ghost and hat layers without touching the kick and snare anchors. The recheck must be against the same control, and just like the timing correction, the first rule must be applied to that new result. The rule exists to find the smallest gap that produces measurable spread, rather than widening the tiers indefinitely.

These three rules are the only decision paths. A swing version must either pass the first rule outright, or be corrected by the second or third rule and pass the first rule after correction. If neither correction brings it to groove score improvement with clarity loss of 1 point or less, the swing version must not be kept. No other justification is allowed for acceptance.

What to do next

Frequently Asked Questions

Can the kick and snare be shifted when adding swing?

No, the kick and snare grid anchors must be preserved with no timing offset.

Where is the 70% ghost-note velocity tier allowed to be applied?

The 70% velocity tier must be applied only to ghost-note layers, not the kick or snare.

Which layers are permitted to receive a measurable timing offset?

Swing is only allowed for hat and ghost layers, with the measurable timing offset strictly restricted to them.

How must mean absolute deviation be calculated during baseline timing drift measurement?

Mean absolute deviation must be calculated separately for hats and ghost notes, and kick or snare must not be averaged into the same MAD.

What is the purpose of measuring baseline timing drift before applying swing?

The baseline timing measurement is not to correct the part, but to establish a concrete starting point for the layers that are candidates for swing while keeping the low-end untouched.

Under what condition can the swing version be accepted?

The swing version can only be accepted if it improves groove without smearing transient clarity.

Quick answers

Also worth reading: AI Drum Humanization: 8ms Offsets Beat Swing Templates: AI Drum Humanization: 8ms Offsets · Swing Quantization and Lo-Fi Drums: How to Verify a 10% Change in Ableton Live 12: Swing Quantization and Lo-Fi Drums: · Magenta Groove MIDI: 12ms Timing, Swing Knob vs Microtiming: Magenta Groove MIDI: 12ms Timing,