• Save
Identifying Successful Melodic Similarity Algorithms for use in Music
Upcoming SlideShare
Loading in...5
×
 

Identifying Successful Melodic Similarity Algorithms for use in Music

on

  • 2,337 views

Margaret Cahill, Donncha Ó Maidín

Margaret Cahill, Donncha Ó Maidín
Centre for Computational Musicology and Computer Music, Department of Computer Science and Information Systems, University of Limerick, Ireland.

Statistics

Views

Total Views
2,337
Views on SlideShare
2,335
Embed Views
2

Actions

Likes
0
Downloads
0
Comments
0

1 Embed 2

http://www.instac.es 2

Accessibility

Upload Details

Uploaded via as Microsoft PowerPoint

Usage Rights

© All Rights Reserved

Report content

Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
  • Full Name Full Name Comment goes here.
    Are you sure you want to
    Your message goes here
    Processing…
Post Comment
Edit your comment

Identifying Successful Melodic Similarity Algorithms for use in Music Identifying Successful Melodic Similarity Algorithms for use in Music Presentation Transcript

  • Identifying Successful Melodic Similarity Algorithms for use in Music Retrieval Margaret Cahill Donncha Ó Maidín University of Limerick, Ireland. [email_address]
  • Introduction
    • MIR - Music Information Retrieval
    • Field of research concerned with searching digitized collections of music
    • Various formats:
      • raw and compressed audio
      • scores
      • MIDI
  • Musical Scores
    • This research is concerned with music in score format
    • Collections of scores online e.g.
      • KernScores http://kern.humdrum.org/
      • MuseData http://www.musedata.org/
    • Music notation editors and scanning software allow saving/exporting in score format
  • Melodic Similarity
    • The ability to detect similar melodies allows:
      • searching for query melodies
      • analysis of compositional style
      • analysis of musical form/structure
      • identification of recurring themes and motifs
  • Current Approaches
    • Many current algorithms are based on string-matching techniques from the IR world
      • e.g. edit-distance algorithm
      • calculates the cost of turning one string into another using delete, replace, and insert actions
    • Operations suited to text are not always suited to melody
  • Features of a score
    • A wide range of features are available from a melodic score:
    • Which features
    • are important?
    • How should they
    • be combined?
    • Which algorithms
    • are suitable?
    Table 1: Some of the musical features available from a score implicit pattern of strong and weak stresses – metrical stress beaming ties tempo directions note duration phrasing accidentals ornamentation note pitch dynamics instrumentation articulation key signature rest time signature barlines clef
  • Our Approach
    • Use music perception research as a guide to choosing suitable features
    • Use a test-bed of melodies for which we already know the degree of similarity
    • Identify potential algorithms
    • Test use of features, method of combining features, values of internal weights and parameters
    • Tweak algorithms to improve performance
  • The Listening Experiment
    • Variations on Twinkle, Twinkle, Little Star for recorder by Duschenes.
    • Theme and Variations provides a natural range of similar melodies.
    • Provides a context for the listener – all melodies are related to the Theme.
    • The Theme is 8 bars long.
    • 9 variations.
    • Good examples of some musical issues melodic similarity algorithms face:
      • different time signatures
      • different keys
      • augmentation of theme (1 bar stretched to 2 bars)
      • triplets
      • single notes replaced by repeated rhythmic repetitions
      • ornate elaborations of the theme using both stepwise movements and arpeggic leaps
  • Theme Figure 1: The first 2 bars of the Theme and Variations
  • Listening Experiment
    • The Theme and each variation easily segments into 4-bar phrases.
    • Two 4-bar segments were extracted to form Part A and Part B of the experiment.
    • Pairs of melodies
    • played
    • Similarity was rated
    • on a 7-point scale.
    • Questionnaire and
    • comment sheet
    • Time = 24-30 minutes.
    • 34 subjects participated.
    Table 2: Structure of the listening experiment
  • Consistency of Results
    • Spearman’s correlation coefficient (correlation on the ranks) used to measure the consistency of subjects.
      • used the similarity rating given by subjects for the sequential and random (repeat) playing of melodies.
    • Only retained data from highly-correlated subjects:
      • >.789, significant at the .01 level
      • Part A: 26 subjects.
      • Part B: 16 subjects.
    • Ratings for random (repeat) playing discarded following the consistency check.
  • Consistency of Results
    • Inter-subject correlation used as
    • a rough measure of the overall
    • reliability of the data.
      • calculated from the correlation
      • matrices of each section
      • how each subject correlated
      • with every other subject
    • Median ratings used for
    • comparison with algorithmic
    • measures of similarity.
    Table 3: Inter-subject correlation Table 4: Results of listening experiment – median ratings.
  • The Algorithms where: k = the time windows of the score p 1 , p 2 = pitch values of the first and second melodies in a window w k = the weight associated with that time window totaldur = the duration processed Figure 2: Algorithm - Ó Maidín 1998
  • The Algorithms
    • processes the score in time windows
      • the duration of the shortest full note at that particular point in the score.
    • sum of the absolute pitch differences in each window multiplied by a corresponding weight for that window
    • weights can be duration, metrical accents,or combinations of these and other weighting factors
    • currently exploring variations of this algorithm using different combinations of musical features
    • transposition identified using the median value of the pitch differences with the associated weight value (similar to using frequency in statistics)
    • edit distance algorithm to be explored next
  • Comparing the Algorithmic and Human Measures of Similarity
    • Two main approaches taken
    • Correlation
      • gives an overall measure of performance across all variations rather than individual variations
    • Normalize data, then apply metric
      • z-scores or min-max normalization
      • can compare the algorithm performance for
      • individual variations
      • metric used to measure performance across
      • all variations
    Figure 3: Possible metrics for use with normalized data
  • Current Work
    • Have identified the use of duration and metrical stress weights for melodies by comparing algorithmic and human measure of similarity
    • Investigating the use of duration and metrical stress weights for melodies by comparing the algorithmic and human measures of similarity
    • Analyzing results through correlation and other metrics to determine which features are most successful for use in melodic similarity algorithms