Emerging Technologies
 
 
  > Presentations  
  > Emerging Technologies
   Committee
 
Music Interaction with Style

The Continuator system learns musical styles from music played by users and then produces music in the same style. Users can play with the system in a question-and-answer mode or together, and they can influence its generation in real time. They can literally perform with virtual musicians, or with themselves.

Innovations
  • A robust and efficient representation of musical phrases, taking into account polyphony, noise, and arbitrary rhythm.

  • An extended multi-layer Markov model to efficiently learn arbitrary corpuses of musical phrases in arbitrary styles. This model allows the system to generalize from musical phrases that are not exactly similar, and thus speed up the learning phase drastically. As a result, the system is able to respond immediately to musical phrases in unknown styles.

  • A biasing mechanism that forces the Markov generation to specific harmonic regions. This point is very important, as it allows users to control the generation of the system in real time , and therefore avoid the mechanistic effect of traditional music-generation systems. This is achieved by introducing a probabilistic scheme in the Markov generation process based on a compromise between Markovian probability (the most expected continuation) and a fitness function (the most fitting continuation with regards to external input). Thanks to this probabilistic scheme, the Markov generation becomes elastic, and it can be twisted in a flexible way.
Vision
There are two implications of this work:
    It shows that a new species of musical instruments is emerging. Current technology allows solutions for the issues related to efficient learning and real time. New issues pop up with the availability of such "devices," in particular regarding new musical-collaboration modes and the nature and property of music, especially real-time performance.

  • The Continuator model opens new perspectives regarding interactive systems that learn. With such models, the primary issues shift from the technical (pattern matching and retrieving) to the behavioural: how to design systems that exhibit "interestingness" and attachment.
Goal
A dynamic system of variable quantities and awareness.

Contact
François Pachet
Sony Computer Science Laboratories

Contributor
Atau Tanaka
Sony Computer Science Laboratories




   presentations

   experiences    services
   > COURSES    > ACADEMIC EXHIBITS    > ACM SIGGRAPH
 
   > EDUCATORS    > ART GALLERY       FORUM  
      PROGRAM    > COMPUTER    > BIRDS OF A FEATHER  
   > EXHIBITOR TECH
      ANIMATION FESTIVAL    > GET INVOLVED  
      TALKS    > EMERGING    > GRAPHICSNET  
   > FUNDAMENTALS
      TECHNOLOGIES    > INTERNATIONAL  
      SEMINAR    > sigKIDS       RESOURCES  
   > KEYNOTE ADDRESS /
   > GUERILLA STUDIO    > PATHFINDERS  
      AWARDS    > VORTECHS      
   > PAPERS          
   > SKETCHES &          
      APPLICATIONS          
   > SPECIAL SESSIONS          
   > WEB GRAPHICS