The Continuator: Musical Interaction With Style

1
The Continuator Musical Interaction With Style

• Presented by Ching-Hua Chuan
• ISE 575, Spring 2007, March 29, USC

2
Introduction

• Real-time interactive musical instruments that
  are able to produce stylistically consistent
  music.
• The system learns music styles automatically, and
  is seamlessly integrated in the playing mode of
  the musicians.
• Interactive v.s. imitation systems

3
Inside the Continuator

• Architecture
• Automatic learning and generation
• Prefix trees with indexing scheme
• Variable-order Markov models
• Reduction functions
• Polyphony and rhythm
• Interactive musical instrument
• Biasing Markov generation

4
Architecture
5
Prefix Trees (learning the sequence)
Input seq 1 A B C D
1 2 3 4
6
Variable-order Markov models(generating the
sequence)
Start again with A B B ? continuation_list 8
Learned prefix trees

• Start again with A B B C
• no match
• continuation_list(B C) 4

Learned seq. A B C D A B B C
1 2 3 4 5 6 7 8
Ending seq A B B C D
7
Reduction Functions

• No continuation is found.

learned
Input
8
Polyphony and Rhythm

• Polyphony legato v.s. musical cluster
• Rhythm (jazz and popular music)
• - Natural, linear, input rhythms
• - Fixed metrical structure

chord
legato
Fixed metrical structure
9
Biasing Markov Generation

• Weighting the nodes according to how they match
  the external input.

S1, we get a musical automaton insensitive to
the musical context, S0, we get a reactive
system which generates the closest musical
elements to the external input it finds in the
database.
10
Experiments
Bernard Lubat
The Continuator and children
Gyorgy Kurtag
Claude Barthelemy
Video from http//www.csl.sony.fr/Research/Experim
ents/Continuator/index.php
11
Demo

• Lets play with it!