Learning the helix topology of musical pitch

Learning the helix topology of musical pitch

Learning the helix topology of musical pitch

In this paper, we present a method to embed frequency bins from an audio dataset in a Euclidean embedding space, testing on music, speech and urban sounds.

We find that highly harmonic instruments such as harp yield a perfectly 'helical' embedding, while urban soundscapes yield rectilinear topologies.

I presented this paper at the virtually-held IEEE ICASSP 2020 Conference.

Companion website for the paper

Helicality: An Isomap-based Measure of Octave Equivalence

in Audio Data

To better compare embedding results from different datasets, we introduce the "helicality" algorithm which aims to quantify the notion of octave equivalence.

We define helicality as a point-cloud's goodness of fit to a Shepard-Risset helix, testing on monophonic instrument recordings, speech and drums.

I presented this Late-breaking demo at the ISMIR Conference 2020.

 

Companion website for the paper