The Influence of Polyphony on the Dynamical Modelling of Musical Timbre

By Jean-Julien Aucouturier and François Pachet.

Pattern Recognition Letters, 28(5), pp. 654–661, 2007.

Read the paper (PDF)
Publisher record

About the paper

The way a sound changes over time helps us recognize the timbre of an instrument. This paper investigates why techniques that model those changes had brought little improvement when comparing whole polyphonic recordings.

Abstract summary

Aucouturier and Pachet compare static and temporal models on datasets of single-instrument notes, simultaneous notes, and longer musical textures. Their experiments identify the overlap of multiple voices as the main obstacle for the temporal models tested. An instrument’s attack or decay can coincide with a different part of another sound, complicating the timing information in the combined signal.

Related work

See Timbre Similarity and Hubness for the broader research and The Bag-of-Frames Approach to Audio Pattern Recognition for the related comparison of music and urban soundscapes. Improving Timbre Similarity: How High’s the Sky? reports earlier experiments on the limits of these similarity measures.

Back to Publications