Expanding Harmonic Vocabularies in Computational Musicology

You are reading version 1 of this direction, frozen 2026-10-04. The latest version is version 2.

Version 1 of 228 papers29 conceptsfrozen 2026-10-04

Most computational music systems, from automatic chord recognition to generative AI, rely on a simplified "ground truth" of harmony. By reducing the vast landscape of musical expression to a handful of basic major and minor triads, these models gain efficiency but lose the ability to process the sophisticated structures that define genres like jazz, late-Romantic classical, and complex pop. A growing body of research argues that for machines to truly "understand" or create music, they must move beyond these basic triads toward expansive, nuanced harmonic vocabularies.

The Triad Constraint

The central tension in current music AI is the trade-off between model stability and harmonic depth. As noted by Daniel Müllensiefen and Geraint A. Wiggins in Evaluating a Chord-Labelling Algorithm, restricting recognition to a predefined set of basic chords limits a model's applicability to complex genres. When a system only knows how to label a "C Major," it becomes blind to the functional nuance of a "Cmaj9" or a "C7(#11)."

This limitation isn't just a labeling problem; it is a generative one. In Domain Adversarial Training on Conditional Variational Auto-Encoder for Controllable Music Generation, Gus Xia and colleagues highlight that generative models often struggle to synthesize sophisticated structures like extended jazz harmonies because they are trained on datasets where these complexities have been flattened into triads.

Bridging Theory and Computation

To overcome this, researchers are integrating formal music theory into computational frameworks. Damien Sprockeels and Peter Van Roy introduced Diatony, a constraint-programming model designed to formalize the rules of four-voice tonal harmony. Their work in Expressing Musical Ideas with Constraint Programming Using a Model of Tonal Harmony emphasizes that while deep learning acts as a "black box," symbolic constraints can provide the mathematical precision needed for complex modulations and chromaticism.

Similarly, Jonathan Bragg and Elaine Chew demonstrated in Neo-Riemannian Cycle Detection with Weighted Finite-State Transducers that treating music analysis as a "noisy channel" problem allows for the detection of complex harmonic patterns—like those found in Schubert—that reductionist triad-based models would miss entirely.

Learning Nuance from Data

Where symbolic rules are too rigid, researchers are turning to distributional semantics. In From Context to Concept: Exploring Semantic Relationships in Music with Word2Vec, Ching‐Hua Chuan and Dorien Herremans treat musical segments as "words," allowing models to learn the "company a chord keeps." This approach suggests that machines can discover sophisticated harmonic relationships purely from exposure to large datasets, provided the vocabulary isn't artificially restricted to the most frequent (and often simplest) segments.

The challenge remains the "data drought" for complex harmonies. Martyna Majchrzak suggests in Training chord recognition models on artificially generated audio that synthetic audio can help bridge this gap, providing the high-volume, richly labeled data necessary to train models on seventh, ninth, and diminished chords that are often underrepresented in human-annotated sets.

Towards Expressive Autonomy

The goal of this research direction is to move music AI from "predictable" to "professional." By expanding the harmonic vocabulary, systems like the Chord-Conditioned Song Generator (CSG) developed by Shun Lei and others can filter out noise and better align with user-specified, complex harmonies. As these models learn to handle the density of voices and the build-up of tonal tension, they move closer to capturing the full expressive range of human composition.

Papers in this version

History & citationVersion 2 · 2026-10-04

Current · Permanent link

  1. Version 2 (latest) · 2026-10-04 · 36 papers
  2. Version 1 · 2026-10-04 · 28 papers
Cite this direction

Latest version (a living page):

@misc{lacuna_harmonic_vocabulary_constraints_ad326271,
  title = {harmonic vocabulary constraints},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/harmonic-vocabulary-constraints/rdr_ad326271dd034911acf12c521155070d},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 1, frozen text):

@misc{lacuna_expanding_harmonic_vocabularies_in_compu_ad326271_v1,
  title = {Expanding Harmonic Vocabularies in Computational Musicology},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/expanding-harmonic-vocabularies-in-computational-musicology/rdr_ad326271dd034911acf12c521155070d/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-20},
}