Chord Recognition

Overview of the Problem

  • How can a computer recognize chords in music?
    • Applications:
      • Automated transcription
      • Live tracking for improvisation
      • Computer-assisted (synthetic) performers
    • Previous work:
      • Analyze pitches
        • Unsolved for polyphony
        • Unclear if this is the human model
      • Inspiration from recent papers:
        • Musical Key Extraction from Audio, Steffen Pauws
        • Chord Segmentation and Recognition using EM-Trained Hidden Markov Models, Alexander Sheh and Daniel P.W. Ellis
        • SmartMusicKIOSK: Music Listening Station with Chorus-Search Function, Masataka Goto
        • A Chorus-Section Dectecting Method for Musical Audio Signals, Masataka Goto
Our Approach
    • Choose a domain which is the subset of the problem
      • Beatles Tunes --- Training/Validation/Testing performed two ways:
        • By song:
          • 16 training songs
          • 1 validation song
          • 2 testing songs
        • By feature sample:
          • 2700 training samples
          • 150 validation samples
          • 246 testing samples
      • Note: The songs chosen are intentionally the same as those from Dan Ellis’ paper cited above
    • Create a feature model based on a simplified version of how we think humans hear sound
    • Segment music; label chords for each segment.
    • Create/compare learning systems which process the labeled data

List of songs (find your favorite):

  • Training:
    • Beatles for Sale:
        • I don't want to spoil the party
        • I'll follow the sun
        • I'm a Loser
    • Help:
        • Help
        • I've Just seen a face
        • It's only love
        • Ticket to ride
        • Yesterday
        • You're going to lose that girl

·        You've got to hide your love away

    • A Hard Day's Night:
        • A hard day's night
        • And I love her
        • Can't buy me love
        • I should've known better
        • I'm happy just to dance w/ you
        • If I fell
        • Things we said today
  • Validation:
    • Tell me why
  • Test:
    • Eight Days a week
    • Every little thing