From melodic note sequences to pitches using word2vec

TL;DR

This paper applies word2vec to melodies, treating notes as words, to capture pitch information and predict notes based on context, demonstrating a strong correlation between semantic vectors and actual pitches.

Contribution

It introduces a novel application of word2vec to musical notes, using small-dimensional embeddings to effectively model pitch relationships in melodies.

Findings

Semantic vectors have about 0.80 correlation with pitches

Notes can be predicted from 2-4 preceding notes

Effective embedding of pitch information in small dimensions

Abstract

Applying the word2vec technique, commonly used in language modeling, to melodies, where notes are treated as words in sentences, enables the capture of pitch information. This study examines two datasets: 20 children's songs and an excerpt from a Bach sonata. The semantic space for defining the embeddings is of very small dimension, specifically 2. Notes are predicted based on the 2, 3 or 4 preceding notes that establish the context. A multivariate analysis of the results shows that the semantic vectors representing the notes have a multiple correlation coefficient of approximately 0.80 with their pitches.