Show HN: Deterministic analysis of musical tension, novelty, and fatigue

I built a deterministic, offline audio analysis system that models long-horizon musical structure:

tension, novelty, fatigue, and impact.

It analyses full tracks and produces interpretable curves and events (drops, stagnation, transitions). No ML or training data.

This repo is the Python reference implementation used to prototype the kernel behind a real-time C++ DAW plugin. The Python outputs are the "golden" results used to validate the C++ port.

Motivation was simple: I heard Beyoncé’s "Haunted" and wanted to understand why it worked so well structurally.

I was not expecting to LIKE Beyoncé!

That turned into a research artefact I thought others might find interesting.

The c++ repo is defo hard work, will share soon.

The behaviour later turned out to align closely with David Huron’s ITPRA framework, though this project doesn’t attempt to prove it.

PS. I didn't ANY time looking for alternatives.