Labelling the Structural Parts of a Music Piece with Markov Models

Authors:
Jouni Paulus;Anssi Klapuri
Affiliations:
Department of Signal Processing, Tampere University of Technology, Tampere, Finland;Department of Signal Processing, Tampere University of Technology, Tampere, Finland
Venue:
Computer Music Modeling and Retrieval. Genesis of Meaning in Sound and Music
Year:
2009

Citing 4
Cited 0

The power of amnesia: learning probabilistic automata with variable memory length

Machine Learning - Special issue on COLT '94
Speech and Language Processing: An Introduction to Natural Language Processing, Computational Linguistics, and Speech Recognition

Speech and Language Processing: An Introduction to Natural Language Processing, Computational Linguistics, and Speech Recognition
Automatic Structure Detection for Popular Music

IEEE MultiMedia
On prediction using variable order Markov models

Journal of Artificial Intelligence Research

Quantified Score

Hi-index	0.00

Visualization

Abstract

This paper describes a method for labelling structural parts of a musical piece. Existing methods for the analysis of piece structure often name the parts with musically meaningless tags, e.g., "p1", "p2", "p3". Given a sequence of these tags as an input, the proposed system assigns musically more meaningful labels to these; e.g., given the input "p1, p2, p3, p2, p3" the system might produce "intro, verse, chorus, verse, chorus". The label assignment is chosen by scoring the resulting label sequences with Markov models. Both traditional and variable-order Markov models are evaluated for the sequence modelling. Search over the label permutations is done with N-best variant of token passing algorithm. The proposed method is evaluated with leave-one-out cross-validations on two large manually annotated data sets of popular music. The results show that Markov models perform well in the desired task.