Play along
1 · Setup
bars
2 · Calibration
Run these in order, on the mic and pads you are actually using. Every number the app reports afterwards depends on them.
Noise floor
Not measured.
Input latency
Not measured.
Pad separability
Not measured.
3 · Practice
Each dot is one hit, placed left of centre if early and right if late. The centre line is the beat. kick slot snare slot ±1 SD
4 · Results
Mean offset
—
accuracy · rushing or dragging
Std deviation
—
consistency · the one that matters
Hits matched
—
0 missed · 0 extra
Distribution of this run's offsets. Earlier runs are listed below, not mixed in — a histogram spanning two tempos would describe neither.
When you want this to be exact
- Two mics into a 2-in interface. One on each pad: the channel is the drum, so classification stops being a guess. Biggest single improvement available.
-
Piezo contact triggers. A trigger on each pad into a cheap MIDI module removes
acoustics from the problem entirely.
MidiHitSourceis already wired for it. - An e-kit. Same path as triggers, no changes needed here.
Diagnostics
Record pad fixture
Records 15 seconds while you hit one pad about 20 times, then downloads a .wav.
Drop both files into
tests/fixtures/real/ and the classifier test reports on
your actual pads instead of the synthetic stand-ins.