Why Do Finger-Drumming Pads Clip Only When Hit Together? A Peak-Collision Matrix

MIDI velocity, perceived loudness, and audio peak are different quantities

Yamaha’s FGDP-30 manual provides one concrete hardware example: pad force can pass through A/D Gain, a fixed velocity, a velocity curve, and minimum and maximum values before becoming MIDI velocity 1–127. That proves the relationship between the same strike and a MIDI number can depend on device settings. It does not create a universal conversion from velocity to dBFS. The audio response to a velocity-127 event must be checked in the receiving instrument and measured; it does not automatically mean the output reaches exactly 0 dBFS.

Apple defines Peak in Logic Pro’s Level Meter as the highest individual sample value and recommends it for tracking peaks while recording or mixing. True Peak looks for inter-sample peaks that can exceed the sample peaks and is intended for checking the output during a bounce. RMS describes average level more closely. For a short drum transient that “flashes red,” log peak hold and the clip indicator first; do not rely only on a quiet overall impression or an RMS reading.

Apple also states that Logic Pro’s output channel strip represents the summed level of all channel strips in the signal flow and that exceeding the reproducible limit at that output causes distortion. Ableton adds an important product-specific qualification: Live 12’s 32-bit floating-point engine can carry internal track levels above 0 dB without necessarily clipping immediately, but Main, physical inputs and outputs, and saved or exported files remain critical boundaries. That is a documented implementation detail, not permission to treat every red meter in every product as harmless.

Create a repeatable input before deciding which hit is too loud

Two live finger-drumming takes will differ in timing and velocity. Comparing them cannot tell you whether a peak changed because of your fix or your fingers. Record or draw one short MIDI test containing the light, medium, and strong hits you actually use—not a row of forced 127s. Preserve the unedited take. Every matrix condition must play this same MIDI.

1. Freeze sound identity. Keep the samples, velocity layers, start points, envelopes, pan, and voice limits for K (kick), S (snare), H (closed hat), and C (crash or another long-tailed voice).
2. Freeze the route. Record instrument output, track faders, drum bus, sends and returns, main output, sample rate, and export format. Do not reroute inside the matrix.
3. Freeze processing state. Write down every EQ, saturation, compressor, and limiter in order and whether it is bypassed. If dynamics remain in the release chain, observe their input, output, and gain reduction.
4. Declare this project’s ceiling. Use the current device or delivery chain’s clip indicators and output target. Do not copy a supposedly universal dB value from this guide. Keep the target identical in every round.
5. Clear every cell. Reset peak hold, clip flags, and gain-reduction hold, play through the full decay, and only then log. An uncleared red flag cannot identify the current combination.

If you cannot insert a meter at an intermediate stage, do not invent the reading. Log the earliest source stage and final output you can observe, and mark the unavailable B stage U. An honest unknown is more useful than guessing that a Main failure belongs to one pad.

Five rounds turn “it clips sometimes” into one repeatable combination

Play every cell at least twice. If its two results disagree, label it unstable and rerun instead of averaging the conflict away. S0 establishes solo baselines; P0 changes only simultaneity; T0 introduces three voices and a continuing tail; G0 restores real musical density; S1 returns to the exact S0 solos to detect project or meter drift.

For every row, log: round and combination; E event count and velocities; S input/output peak; B input/output peak; M peak/true peak; clip flags; maximum compressor or limiter gain reduction; the specific audible change; and one first-failure code. Peaks depend on signal interaction. Do not predeclare that “two equal peaks always add 6 dB,” and do not assume another snare will reproduce a lower K+S result from this kit.

Code only the earliest failed layer so five controls do not move together

Classify from upstream to downstream. An earlier failure contaminates later observations, so each row gets one primary code. Fix it, rerun, and only then interpret the next layer.

The existing source-pad-to-victim-pad matrix handles “one pad triggered another.” The layered-kick test handles polarity and relative sample starts. Stay in this guide only when E passes, S0 passes, and P0 fails at a shared gain stage. That keeps physical crosstalk, event collision, phase cancellation, and headroom from collapsing into one vague “simultaneous hit” problem.

Repair the first overloaded stage; preserve musical velocity until it is the actual variable

Yamaha describes headroom as the safety margin between the average signal and the distortion ceiling and notes that poor gain staging reduces headroom while increasing noise or distortion. That does not mean pulling every track arbitrarily low. It means leaving enough room at each real boundary for the combinations and strong hits your performance actually contains.

1. S failure: repair the source. Lower sample or instrument output—or the upstream gain before the failed point. If distortion is already printed into a sample, a downstream fader only makes that distortion quieter. Rerun every S0 solo.
2. B failure: create combination headroom. Decide whether one voice is anomalously hot or the whole kit reaches the bus too high. Change one voice gain, one shared kit trim, or one pre-bus input stage; preserve the intended K/S balance, then rerun S0, P0, and T0.
3. M failure: repair the full-mix relationship. If the drum bus passes alone and Main fails only after bass, accompaniment, or returns enter, adjust the shared-output mix and sends instead of lowering pads at random.
4. Tail-only failure: test a time-domain choice separately. If only C→K+S fails, duplicate the version and change exactly one of C level, decay, choke, or arrangement space. Label it a sound or arrangement choice, not a universal gain answer.
5. Change velocity only for a musical reason. Curves and limits alter the strike-to-MIDI mapping and therefore change the test input. Treat velocity as a separate candidate only when the current mapping cannot express the intended dynamics; consult the receiving instrument’s own documentation and rerun event, timbre, and peak checks afterward.
6. Evaluate limiting last. Apple explains that a limiter reduces peaks above its threshold and exposes a reduction meter; it cannot repair audio clipped during recording. Treat it as audible dynamics processing. Log reduction, run a level-matched A/B, and check distortion rather than replacing upstream diagnosis with “the final meter stayed green.”

Save every change as a candidate instead of overwriting the original kit. After one parameter changes, restart at S0—do not retest only the K+S cell that failed. Lowering one voice may clear the bus while burying a ghost note; changing a velocity curve may protect output while damaging the original dynamic contour. Regression is what exposes that cost.

The release gate includes real strong hits, final output, and an unchanged return

The matrix is a diagnostic, not the finished music. After a candidate passes fixed MIDI, perform the same G0 passage twice on live pads. Include the strongest accents, fill, and crash entrance you genuinely use; do not invent an extreme strike outside the performance. Retain MIDI and audio. Confirm that events remain complete, the earliest overload is gone, and dynamics were not unintentionally flattened.

Export once from the final output, then re-import into an empty project or trusted checker without added gain. Check peak or true peak, the start, the complete tail, and unexpected distortion. If the host resembles Live in having internal floating-point headroom, export and physical output are essential boundaries. If the device does not expose true peak, document that tool limitation instead of inventing a value.

Frequently asked questions

A track is red but Main is not. Must I fix it immediately?

Check the host documentation and the actual boundary first. Apple’s Logic Pro guidance emphasizes the summed output strip; Ableton says internal Live 12 floating-point tracks can exceed 0 dB without immediate clipping. A plug-in input, hardware output, Main, and export may have different limits. The answer is not to ignore red—it is to record the stage and test the next real boundary.

Can I cap every pad’s velocity and solve this permanently?

Not as a universal fix. Velocity is not dB, and a curve or maximum changes the test input and performance dynamics; the receiving instrument’s exact response depends on its own implementation. If S0 passes and only the shared bus fails on combinations, repair shared gain staging first. Change velocity only when the performance mapping itself is wrong, then regress timbre and dynamics.

Why not put a limiter on Main and move on?

A limiter can belong in the release chain, but it actively reduces peaks. A green output does not prove its input was healthy or that the kick/snare transient relationship stayed intact. Find the first failed stage, then decide whether the amount of limiting is the sound you want.