Brad's verdict after the kick and snare were rebuilt: "Much improved. We've got more work to do on the sound of it, but it works for now." This is the list, so it is not carried in anyone's head.
Everything below is open by ear, not by any failing test — the twelve traits pass and all seven planted faults land.
Toms have no pitch envelope
The kick got one and it is most of why it now reads as a kick. The toms did not, and real toms do fall a little when struck hard.
Careful with the measurement, though: a zero-crossing sweep currently reports 18–52% drops across the four toms, and that number is an artifact. It is the fast-decaying attack mode dying away and leaving the fundamental behind, not a bend. Nothing in the tom voices moves in pitch at all. So the open question is whether a real, shallow bend (10–20%, much less than the kick's 53%) makes them better — and that is an ear question, because the measurement above cannot tell the two apart.
The hats and cymbals are untouched
Still the original all-bank voices. Cymbals have their own ceiling, which is #64 — a linear bank of 64 modes cannot make a convincing wash and the literature says so. The hi-hats are a smaller version of the same thing and might respond to the noise-layer treatment the snare got, since a hat is closer to a burst of broadband metal than to a set of tuned modes.
The snare's wires are still unsourced
The head is real — Skrodzka, Hojan & Proksza (2006), six modes with measured damping. The wires are band-passed noise at 3.2 kHz with a 0.26 s decay, chosen by ear and by what made the spectral flatness look like noise. No academic source for the wires was reachable, and the dossier records why: every source treats them as inseparable from the shell and head.
What would make the next pass cheaper
A way to A/B two versions of one drum without rebuilding the kit — render both to a WAV pair at matched gain and listen back to back. tools/render_component.py does one-shots per note but not variants of the same note, and the DAW A/B layout convention in the workspace anchor already says how such a pair should be laid out. That would turn each of the questions above from a rebuild into a listen.
Brad's verdict after the kick and snare were rebuilt: "Much improved. We've got more work to do on the sound of it, but it works for now." This is the list, so it is not carried in anyone's head.
Everything below is open by ear, not by any failing test — the twelve traits pass and all seven planted faults land.
Toms have no pitch envelope
The kick got one and it is most of why it now reads as a kick. The toms did not, and real toms do fall a little when struck hard.
Careful with the measurement, though: a zero-crossing sweep currently reports 18–52% drops across the four toms, and that number is an artifact. It is the fast-decaying attack mode dying away and leaving the fundamental behind, not a bend. Nothing in the tom voices moves in pitch at all. So the open question is whether a real, shallow bend (10–20%, much less than the kick's 53%) makes them better — and that is an ear question, because the measurement above cannot tell the two apart.
The hats and cymbals are untouched
Still the original all-bank voices. Cymbals have their own ceiling, which is #64 — a linear bank of 64 modes cannot make a convincing wash and the literature says so. The hi-hats are a smaller version of the same thing and might respond to the noise-layer treatment the snare got, since a hat is closer to a burst of broadband metal than to a set of tuned modes.
The snare's wires are still unsourced
The head is real — Skrodzka, Hojan & Proksza (2006), six modes with measured damping. The wires are band-passed noise at 3.2 kHz with a 0.26 s decay, chosen by ear and by what made the spectral flatness look like noise. No academic source for the wires was reachable, and the dossier records why: every source treats them as inseparable from the shell and head.
What would make the next pass cheaper
A way to A/B two versions of one drum without rebuilding the kit — render both to a WAV pair at matched gain and listen back to back.
tools/render_component.pydoes one-shots per note but not variants of the same note, and the DAW A/B layout convention in the workspace anchor already says how such a pair should be laid out. That would turn each of the questions above from a rebuild into a listen.