madmom-infer -- NOTICE

madmom-infer is an independent, from-scratch reimplementation of algorithms
published by the CPJKU/madmom project:

    https://github.com/CPJKU/madmom

It is NOT an official fork, port, or release of madmom, and is not affiliated
with or endorsed by the Department of Computational Perception at Johannes
Kepler University, Linz, or the Austrian Research Institute for Artificial
Intelligence (OFAI), Vienna. Attribution to the original authors is preserved
in LICENSE and README.md because this project re-derives algorithms they
designed and published, even though no original source code is reused.

No pretrained weights are bundled. madmom's own trained model files (`.pkl`
and similar) are licensed CC BY-NC-SA 4.0 (non-commercial) by the original
authors, separately from madmom's BSD-2-Clause source code license. This
project does not, and will not, bundle, vendor, or redistribute any of
madmom's own pretrained weights, under any phase of its roadmap. See README.md
for details.

Phase 2 (`madmom_infer/models.py`) downloads these same `.pkl` weight files
at RUNTIME, on demand, directly from the official upstream
`CPJKU/madmom_models` GitHub repository, into a local cache
(`$XDG_CACHE_HOME/madmom_infer/models/`, never inside this project's own
package or git history -- see `.gitignore`). The downloaded bytes remain
CC BY-NC-SA 4.0 (Attribution-NonCommercial-ShareAlike) -- **non-commercial
use only** -- regardless of madmom-infer's own BSD-2-Clause license, which
covers only this project's source code (including `models.py`'s download
logic itself), never the weight bytes it fetches. Anyone deploying
`RNNDownBeatProcessor` (or any future processor that downloads a madmom
model) in a commercial context is responsible for either obtaining a
separate license from the original authors for the specific weights used,
or supplying their own non-CC-BY-NC-SA-encumbered weights.
