Transcript jSymbolicx - McGill University

jSymbolic
Cedar Wingate
MUMT 621
Professor Ichiro Fujinaga
22 October 2009
Types of Features
 Low Level
 High Level
 Cultural
What are High Level Features?
 Information that consists of musical
abstractions that are meaningful to
musically trained individuals.
 Examples include instruments present,
melodic contour, chord frequencies and
rhythmic density.
Why High Level Features?
 Musicological and music theoretical value
 Great deal of music already encoded in
MIDI or Humdrum’s kern
 Optical music recognition can provide
even more symbolic musical data
jSymbolic application
 Application to extract high-level features
from MIDI files
 Many high-level features cannot be extracted
from audio recordings
 Open Source
 Designed to easily add new features with
basic JAVA and MIDI skills
Building a feature set
 Goals
 Single software system that could be applied to classical music, jazz and
a wide variety of popular and traditional musics
 Use this software without needing to make any manual adjustments or
adaptations in order to deal with different types of music
 Issues to consider
 “Curse of dimensionality”
 Many systems of analysis
 Many of which rely on intuitive subjective judgment
 jSymbolic solution
 Large catalogue of general features
 User can select which ones to include or exclude
 Concentrate on features that can be represented by relatively simple
statistics
 Intermediate representations
 Histograms
Feature characteristics
 Features that can be represented by simple
numbers or small vectors
 One-dimensional features
 Means, standard deviations, true/false values
 Multi-dimensional features
 Histograms
Example: Beat Histogram
(McKay, C., and I. Fujinaga. 2007)
The Features
 Drawn from musical research
 Music Theory (Julie Cumming), Ethnomusicology (Alan Lomax,
Bruno Nettl), Music Cognition (Bret Aarden and David Huron)
and Popular Musicology (Philip Tagg)
 160 Total Features (111 implemented)
 Instrumentation
 Pitched/Unpitched
 Note and Time Prevalence and Variability of Note Prevalence
 Fraction
 Texture
 Independent Voices
 Voice equality
 Range of Voices
 Rhythm





Strength
Looseness
Polyrhythms
Density
Tempo, Meter
The Featues (continued)
 Dynamics
 Range
 Variation
 Pitch Statistics





Common Pitches
Variety
Range
Glissando
Vibrato
 Melody





Intervals
Arpeggiation
Repetition
Chromaticism
Melodic Arc
 Chords (not implemented)
More examples
Twenty sample features extracted from the first two measures of Fryderyk
Chopin’s Nocturne in B, Op. 32, No. 1 (McKay, C., and I. Fujinaga. 2007).
More examples
Twenty sample features extracted from measures 10 and 11
of the first movement of Felix Mendelssohn’s Piano Trio No.
2 in C minor, Op. 66 (McKay, C., and I. Fujinaga. 2007).
Application: Automatic Genre
Classification
 Automatic music classification and the importance
of instrument identification (McKay, C., and I.
Fujinaga. 2005)
 Able to correctly classify MIDI recordings among 9
categories 90% of the time and among 38 categories
57% of the time.
 Root genre identified 90% for 9 categories, 80% for
38 categories.
 Better than audio based classification systems (below
80% among 5 categories)
 Found that features relating to instrumentation
performed significantly better than other features in
automatic genre classification.
Application: Automatic Genre
Classification (continued)
(McKay, C., and I. Fujinaga. 2005)
Application: Automatic Genre
Classification (continued, again…)
(McKay, C., and I. Fujinaga. 2005)
Bibliography
McKay, C. 2004a. Automatic genre classification of MIDI recordings. M.A.
Thesis. McGill University, Canada.
McKay, C. 2004b. Automatic genre classification as a study of the viability of
high-level features for music classification. Proceedings of the International
Computer Music Conference. 367-70.
McKay, C., and I. Fujinaga. 2005. Automatic music classification and the
importance of instrument identification. Proceedings of the Conference on
Interdisciplinary Musicology.
McKay, C., and I. Fujinaga. 2006. jSymbolic: A feature extractor for MIDI files.
Proceedings of the International Computer Music Conference. 302-5.
McKay, C., and I. Fujinaga. 2007. Style-independent computer-assisted
exploratory analysis of large music collecitons. Journal of Interdisciplinary
Music Studies 1 (1): 63-85.