Measured, not asserted

We measured it against a real Wurlitzer 200A

Plenty of Reeds is modeled (modal synthesis), not sampled. To keep the model honest, every per-note render is compared against a real Wurlitzer 200A — the Pifinger sample set — across partial tuning, decay, attack envelope, tremolo depth, and velocity timbre. The numbers below come straight out of the analysis scripts in spectral_analysis/; nothing here is rounded in the model’s favor.

The numbers

~4.5c

Partial tuning vs real 200A

median deviation, k1–k2, n=36 notes

src · spectral_analysis/spectral_match_summary.txt

5.77 dB

Tremolo depth

vs real 5.78 dB swing · ~5.5 Hz

src · spectral_analysis/measure_tremolo_depth.py

14.6 ms

Attack peak time

vs real 12.9 ms

src · spectral_analysis/measure_attack.py

+15 dB

Velocity bark (soft→hard)

vs real +18 dB upper-harmonic growth

src · spectral_analysis/measure_bark.py

64 voices

Polyphony

one per 200A reed

src · dsp/Engine.h

The plots

These are the actual measurement artifacts rendered by the analysis suite — real Pifinger 200A (reference) overlaid against Plenty of Reeds (PoR).

Per-partial frequency deviation in cents, real 200A vs Plenty of Reeds, across notes C2–C6 — Per-partial tuning. Each point is one harmonic’s deviation from the ideal integer multiple, in cents (vertical), per note (horizontal). For partials k1–k2 the model tracks the real instrument to a ~4.5-cent median. The visible spread is the real 200A’s own inharmonicity — the model follows it rather than snapping to ideal harmonics. Source: `spectral_analysis/spectral_match_summary.txt`.

T60 decay time per note, real 200A reference vs Plenty of Reeds — T60 decay time (seconds to −60 dB) per note. Reference vs PoR. The model’s reeds decay slightly longer than the real instrument (e.g. A4 5.10 s modeled vs 3.38 s reference) — an honest, audible difference, not hidden. Source: `spectral_analysis/compare_full_summary.txt`.

Tremolo amplitude-swing scatter, real 200A vs Plenty of Reeds — Tremolo amplitude swing (dB) measured per sample. The modeled median lands at 5.77 dB against the real instrument’s 5.78dB (67 samples), at the 200A’s characteristic ~5.5 Hz. The model uses an asymmetric-triangle LFO — amplitude modulation only, no pitch vibrato. Source: `spectral_analysis/measure_tremolo_depth.py`.

Where the model falls short

No model is the instrument. Two gaps are known, measured, and deliberately left visible rather than papered over:

Bark dynamics: +15 dB vs +18 dB. The real 200A grows ~+18 dB of upper-harmonic energy from a soft hit to a hard one; the model reaches +15 dB. The shortfall is a modal ceiling — pushing the velocity→timbre law (kVelTilt = 0.38) harder to close the last 3 dB starts to cost mid-velocity centroid accuracy, so the trade is held where it is. Source: spectral_analysis/measure_bark.py, dsp/Reed.h.
Upper-partial cents drift in octave 6.The per-partial table shows the highest octave’s upper harmonics (k4–k6 around C6/D6) wandering tens of cents off in the reference itself — those partials sit near Nyquist and are noisy to estimate. The model stays close on k1–k3 there but cannot meaningfully chase the reference’s scattered top partials in octave 6. Source: spectral_analysis/spectral_match_summary.txt (Octave 6 rows).

If a number on this page ever drifts from the script output, the script is the source of truth — not the copy.