Submitted by: Submitted by quiroule
Views: 378
Words: 4608
Pages: 19
Category: Science and Technology
Date Submitted: 04/16/2013 02:11 AM
Using Fisher Kernels and Hidden Markov Models for the Identification of
Famous Composers from their Sheet Music
David R. Hardoon
University of Southampton
ISIS Research Group
Building 1, Highfield
Southampton, SO17 1BJ
drh@ecs.soton.ac.uk
Craig Saunders
University of Southampton
ISIS Research Group
Building 1, Highfield
Southampton, SO17 1BJ
cjs@ecs.soton.ac.uk
ABSTRACT
We present a novel application of Fisher kernels to the
problem of identifying famous composers from their sheet
music. The characteristics of the composers writing style
are obtained from note changes on a basic beat level, combined with the notes hidden harmony. We are able to
extract this information by the application of a Hidden
Markov Model to learn the underlying probabilistic structure of the score. We are able to use the model to generate new sheet music based on a specific composer. Furthermore we do an identification comparison using Fisher
kernels and a Hidden Markov Model on limited data.
Keywords: Fisher Kernel, HMM, SVM, Composer
Identification.
1
Introduction
In the last several years music information retrieval,
generation and classification has become a major topic
of interest. The analysis of music has been addressed
using natural language modelling and statistical models
that have been mostly focused on digital audio resources.
Batlle and Cano (2000) have applied Hidden Markov
Models (HMM)(Rabiner, 1989) on audio fragments
for music segmentation and classification (Batlle et al.,
2004). While others have applied grammatical induction
to learn the genre and style of MIDI files. An interesting
review of music generation from statistical models is
given by Conklin (2003).
In previous work (Saunders et al., 2004) we have
presented a novel application of string kernels to the
problem of recognising famous pianists from their style
of playing. This was done using a measurement of the
changes in beat-level tempo and beat-level loudness
(Zanon...