toxadev/aura-engine-filters

Aura Engine FIR filters

Pre-computed FIR upsampling filters, stored as raw float64 NumPy arrays. These are the filters the Aura Engine convolves with, kept here so the web converter can fetch only the ones a conversion actually needs instead of carrying all of them.

Naming

fir_<TAPS>_<TARGET_HZ>_<PHASE>.npy

• TAPS — 5k, 1M, 5M, 10M or 30M
• TARGET_HZ — output sample rate: 88200, 176400, 352800, 705600 for the 44.1 kHz family, and 96000, 192000, 384000, 768000 for the 48 kHz family
• PHASE — linear_phase or minimum_phase

So fir_30M_352800_linear_phase.npy is a thirty-million-tap linear-phase filter designed for 44.1 → 352.8 kHz.

Each file is a plain .npy of float64 coefficients — 8 bytes per tap, no compression, nothing to unpack:

import numpy as np
h = np.load("fir_1M_352800_linear_phase.npy")   # shape (1000000,), dtype float64

Why a filter per ratio

A filter designed for one ratio has the wrong cutoff for another. Using the ×8 blob at ×2 would pass everything between the two Nyquist limits that the lower ratio is supposed to remove, so every ratio carries its own design rather than sharing one.

Sizes

Taps	Per file	All 16 files
5k	40 KB	0.6 MB
1M	8 MB	123 MB
5M	40 MB	611 MB
10M	80 MB	1.2 GB
30M	240 MB	3.6 GB

License

Same terms as the engine: PolyForm Noncommercial 1.0.0.