Programmatic music with music21

music21 models notation superbly and models performance not at all, and it does not render audio. That gap is where generated music goes wrong: theoretically correct, audibly dead, or silently broken in ways you cannot hear because you cannot listen.

This skill covers both halves — the library in depth, and the performance and rendering layer that turns a Stream into something a person would play twice.

Setup

pip install music21                          # 10.5.0 stable; everything here is verified on it
brew install fluidsynth ffmpeg               # or: apt install fluidsynth ffmpeg
python -c "from m21kit import render; print(render.ensure_soundfont())"   # ~141MB, once

m21kit/ is this skill's helper library — put it on PYTHONPATH or copy it into the project. It is plain functions over music21; nothing is subclassed or monkeypatched.