Investigating Style with Scale Embeddings
Format (When): Where
- Proceedings (2022) – Mathematics and Computation in Music, Georgia
- Presentation (2022) – Texas Society for Music Theory
- Title: Probabilistically Modeling Scale Theory
TL;DR
In this paper, we use pitch-class vector embeddings to study scale relationships between composers. Recent research in naturalLanguageProcessing (NLP) has used machine learning to derive vector representations-known as embeddings—for words based on their co-occurrence. Borrowing from NLP, we use the word2vec algorithm to encode windows of pitch-classes, or pitch-class vectors, of music. We show that these embeddings not only replicate the well-known theoretical circle of fifths, but can also capture stylistic nuances between composers’ use of scales.