Quantifying the evolution of harmony and novelty in western classical music

Code to reproduce the analyses in:

Quantifying the evolution of harmony and novelty in western classical music. Preprint: arXiv:2308.03224

The pipeline represents each piece as a bar-by-bar sequence of local keys (via a variant of Elaine Chew's center of effect algorithm), and from it computes information-theoretic measures of key uncertainty, key diversity, and novelty in key transitions, together with composer-similarity and corpus-filtering analyses.

Where to find the Data

This GitHub repository contains code only. The datasets are distributed separately (they are large and partly third-party).

• Full dataset + code archive: https://doi.org/10.1184/R1/31714561

Download the archive and place its Data/ folder at Notebooks/Data/ so the tree looks like the layout below.

Repository layout

Harmony_Evolution
├── setup.jl / setup.sh / setup.ps1   installers (see Installation)
├── Readme.md
├── LICENSE
└── Notebooks
    ├── Project.toml / Manifest.toml   pinned Julia environment
    ├── ChordAccuracy.ipynb            chord accuracy: CoE vs. Krumhansl–Schmuckler
    ├── KeyAccuracy.ipynb              local-key accuracy: CoE vs. Krumhansl–Schmuckler
    ├── UncertaintyAndDiversity.ipynb  per-piece key uncertainty & diversity
    ├── NoveltyTests.ipynb             novelty / innovation trends (multi-threaded)
    ├── MarkovTests.ipynb              Markov-chain corpus filtering → retained set
    ├── ComposerSimilarities.ipynb     composer similarity matrices & rankings
    ├── BetaSensitivity.ipynb          sensitivity to the β smoothing parameter
    ├── Plots.ipynb                    figures for the main text
    ├── KeyFindingExample.ipynb        tutorial: the CoE key-finding algorithm
    └── Data/                          NOT on GitHub — obtain from the DOI above
        ├── Pieces_data.csv            master metadata (all pieces)
        ├── Pieces_data_reviewed.csv   reviewed metadata
        ├── retained_pieces.txt        corpus retained by MarkovTests
        ├── CSVPieces/                 per-piece key/note tables
        ├── DCML_Corpora/              third-party annotations (see Attribution)
        └── files/                     demo inputs (Mozart_16.mxl, HeyJoe.csv)

Prerequisites

• Julia 1.11 — recommended via juliaup.
• Git.
• A Jupyter front-end is optional: setup.jl registers an IJulia kernel, and IJulia will bootstrap a minimal Jupyter via Conda if you don't already have one. VS Code (Julia extension) works without a separate Jupyter install.

No Python toolchain is needed.

Installation

Clone, then run the installer for your platform from the repository root:

git clone https://github.com/spiralizing/Harmony_Evolution.git Harmony_Evolution
cd Harmony_Evolution

./setup.sh        # macOS / Linux
# setup.ps1       # Windows (PowerShell)
# julia setup.jl  # any OS, directly

setup.jl instantiates the exact pinned environment in Notebooks/ (Pkg.instantiate() + precompile) and registers a project-bound Jupyter kernel named "Julia (Harmony_Evolution)". This kernel is generated per machine and points at this repository's environment, so running setup is required before opening the notebooks.

Running the notebooks

Make sure Notebooks/Data/ is populated (see above), then launch Jupyter with the project active and open a notebook:

julia --project=Notebooks -e 'using IJulia; notebook(dir="Notebooks")'

or open the folder in VS Code and open a notebook in Notebooks/. In either case select the "Julia (Harmony_Evolution)" kernel.

Notes:

• Work from the Notebooks/ directory — data paths are resolved relative to it.
• NoveltyTests.ipynb uses threads. To parallelize it, start Julia with threads before launching Jupyter, e.g. JULIA_NUM_THREADS=auto.

Reproducibility

• Notebooks/Project.toml + Notebooks/Manifest.toml pin the full dependency tree to the exact versions used for the study (Julia 1.11). MusicSpiralRepresentation is unregistered and pinned by URL; Pkg.instantiate() fetches it automatically.
• If you regenerate intermediate files from scratch, run in this order: MarkovTests (writes Data/retained_pieces.txt) → UncertaintyAndDiversity and NoveltyTests (write their per-piece CSVs) → ComposerSimilarities → Plots. Generated CSVs are written under Data/.

License

See LICENSE.

Data attribution

The hand-annotated corpora under Data/DCML_Corpora/ were created and published by the Digital and Cognitive Musicology Lab (DCML), EPFL, and are redistributed here under their respective licenses (see the LICENSE/CITATION.cff files bundled in each corpus directory):

• Beethoven string quartets (ABC). Neuwirth, M., Harasim, D., Moss, F. C., & Rohrmeier, M. (2018). The Annotated Beethoven Corpus (ABC): A Dataset of Harmonic Analyses of All Beethoven String Quartets. Frontiers in Digital Humanities, 5. https://doi.org/10.3389/fdigh.2018.00016
• Mozart piano sonatas. Hentschel, J., Neuwirth, M., & Rohrmeier, M. (2021). The Annotated Mozart Sonatas: Score, Harmony, and Cadence. Transactions of the International Society for Music Information Retrieval, 4(1), 67–80. https://doi.org/10.5334/tismir.6
• DCML corpora meta-repository: https://github.com/DCMLab/dcml_corpora

Citation

If you use this code, please cite the paper (arXiv:2308.03224) and, where relevant, the DCML datasets listed above.