Musical analysis ... in a nutshell ... and otherwise

Where do you put miscellaneous analyses of music? Typically, only a small number are be polished up as a formal analysis in a public setting like a journal. Even on the computational side, it's usually only a small part of the analysis that ends up in a dataset, e.g., one parameter only in an harmonic analysis dataset.

One day, there may be a beautiful, coordinated, open source, data repository to which we can all happily contribute. In the meantime, one answer to this (as to so many questions) is to make a GitHub repo as a proof of concept and see how it goes.

I may use this for short-form observations ("Musical analysis in a Nutshell", hence the title "noteshells"), or long form deep-dives (on which see "Towards 'Everything' Analyses" below), or both. Frankly, I might abandon the effort altogether. In any case, and for now, I begin with the demonstration of an "everything" analysis of a very well known work.

In the early twentieth century, Donald Francis Tovey and others wrote "bar-by-bar" analyses of canonical works like the Beethoven sonatas. It was the typical style for long-form musical analysis at the time to proceed in this fashion from the start for the end of the piece with a fairly high degree of detail. Moreover, this style was to be seen not only in books and editorial commentaries, but even in outlets like programme notes at concerts, which would routinely come complete with examples in musical notation.

"Long-form music analysis" has since taken a different form, with essays now more typically emphasising points of interest and reorganising their narrative around principles other than "real time". Then again (... and here comes the understatement ...) the long-form essay is also only one format for conveying information. The digital age suggests not only alternative output formats (such as on-score commentary) but even alternative use-cases such as text not even intended for human readership at all, but rather as 'annotation' for training neural networks.

It is perhaps slightly ironic that both of these mentalities (on-score commentary and training data) bring us back full circle, inviting us to revisit the bar-by-bar idea and even to extend the principle further: we could (should?) provide both structured and free prose commentary on literally every bar, or even note, and across any parameter we can image. We could (should?) not concern ourselves with readability: the different medium frees us from the requirement to think in terms of a human read making their way all the way through in one sitting. We can instead focus as nerdily as we like on microscopic detail. This does not prelude the pursuit of connections, hierarchies and more, but encourages doing so directly, on their own terms.

Moreover, on the music computing side of this history, we now increasingly have structured mechanisms for organising aspects of such analytical statement with regularised syntax for harmonic analysis, for instance. This too suggests changes and stands to afford specificity and clarity to a field which has been too content to organise arguments over musical analysis via the unwieldy medium of monographs and over timescales that are better suited to measuring the evolution of dinosaur species than vigorous scientific development.

Music nerds unite!

The demo directory sets out a version of this for the case of the Moonlight sonata.

MIT

Mark Gotham, 2026