skip to content

Centre for Music and Science

 

Empirical methods allow us to scale music analysis to large datasets. Instead of focusing on a single musical composition, we can do things like analyse an entire composer’s oeuvre, all charting pop songs from 1980 to 1990, or the musical traditions of 30 cultures from around the world. This allows us to gain valuable systematic insights into musical styles.

The ingredients of musical styles

One primary interest is in identifying the ingredients that define a given musical style, be that of a composer, improviser, performer, genre, or culture. We aim to develop algorithms capable of quantifying the presence of these ingredients in musical examples, enabling us to examine their prevalence across musicians and cultures and trace their development over time.

One approach we use involves constructing supervised machine-learning models that learn to associate a given musical excerpt with a specific performer or style. We then investigate the features the model relies on to make its predictions. For example, Cheston et al. (2024) used this approach for analysing rhythmic style in jazz improvisers.

A traditional method for building these supervised models involves creating a large, hand-crafted feature set informed by music-theoretical and cognitive principles, such as pitch intervals, n-grams, metrical positioning, and syncopation. This approach makes the model’s classification strategies more interpretable but limits its capacity to capture more sophisticated musical concepts. Alternatively, we can use modern deep-learning techniques which allow models to learn their own representations from large datasets, and try retrospectively to interpret these representations in terms of music theory and cognition.

These machine-learning approaches help uncover objective differences between musical styles. An interesting follow-up question is how these relate to listener judgments of musical styles. Some cues that the machine finds helpful might be imperceptible to the listener, whereas others might disproportionately influence their judgments. To explore such questions, we can run experiments (perhaps online with PsyNet) where listeners make stylistic judgments, and compare these judgments to the judgments of our machine-learning models.

Biological and psychological influences on musical styles

A second interest is how musical styles are shaped by biological and psychological factors. To study this question, we take large datasets of existing music and try to explain stylistic patterns through the application of relevant computational models, which could include both factors constraining the production of musical sounds (e.g. how the larynx produces pitches), the perception of those sounds (e.g. how roughness appears on the basilar membrane) and the cognition of those sounds (e.g. how a melody is encoded in memory). To complement such studies, we also run simulated cultural evolution studies, where large groups of online participants interact in artificial cultures and together produce musical artifacts (e.g. songs) that are shaped by biological and psychological factors.

Example publications

Anglada-Tort, M., Harrison, P. M. C., Lee, H., & Jacoby, N. (2023). Large-scale iterated singing experiments reveal oral transmission mechanisms underlying music evolution. Current Biology: CB, 33(8), 1472-1486.e12. https://doi.org/10.1016/j.cub.2023.02.070

Cheston, H., Schlichting, J. L., Cross, I., & Harrison, P. M. C. (2024). Rhythmic qualities of jazz improvisation predict performer identity and style in source-separated audio recordings. Royal Society Open Science. https://doi.org/10.1098/rsos.240920

Harrison, P. M. C., & Pearce, M. T. (2020). Simultaneous consonance in music perception and composition. Psychological Review, 127(2), 216–244. https://doi.org/10.1037/rev0000169

Harrison, P. M. C., & Pearce, M. T. (2020). A computational cognitive model for the analysis and generation of voice leadings. Music Perception, 37(3), 208–224. https://doi.org/10.1525/mp.2020.37.3.208

See also

Mauch, M., MacCallum, R. M., Levy, M., & Leroi, A. M. (2015). The evolution of popular music: USA 1960-2010. Royal Society Open Science, 2(5), 150081. https://doi.org/10.1098/rsos.15008

Savage, P. E., Brown, S., Sakai, E., & Currie, T. E. (2015). Statistical universals reveal the structures and functions of human music. Proceedings of the National Academy of Sciences, 112(29), 8987–8992. https://doi.org/10.1073/pnas.1414495112

Savage, P. E., Passmore, S., Chiba, G., Currie, T. E., Suzuki, H., & Atkinson, Q. D. (2022). Sequence alignment of folk song melodies reveals cross-cultural regularities of musical evolution. Current Biology. https://doi.org/10.1016/j.cub.2022.01.039

CMS Logo

Latest news

New paper: Artist identification using rhythm via machine learning

21 October 2024

We are excited to share our new paper appearing in the journal Royal Society Open Science, entitled “Rhythmic Qualities of Jazz Improvisation Predict Performer Identity and Style in Source-Separated Audio Recordings”. This was completed by Huw Cheston during his PhD at the CMS, and builds from two earlier publications...