Cluster Chair: Elaine Chew, echew@usc.edu
University of Southern California
The issues in computational modeling of music have risen to the forefront due to the entertainment sector's interests in music recommendation (personalized systems for online retrieval of music) and performance rendering systems (impacting the movie scoring and music industries). Many problems in computer-based methods for content analysis and feature extraction, similarity assessment and classification, and generative methods for creating music can be solved using OR approaches. This inaugural cluster on OR in the Arts presents recent research in mathematical models for musical design and interactive music systems.
SESSION 1. Mathematical Models for Musical Design I
Session Chair: Thomas Noll, Technical University of Berlin
Mark Steedman, U. of Edinburgh.
The Grammar of Musical Chord Sequences
The paper shows that chord sequences of the kind that form the harmonic backbone of western tonal music can be characterized by a syntax and semantics of a kind that is standard in natural language. The harmonic semantics is model-theoretic and compositional. The syntax is of low ("mildly context sensitive") expressive power (although it is highly ambiguous), allowing standard polynomial parsing algorithms and techniques of statistical modeling to be applied.
Elaine Chew, USC.
Slicing It All Ways: Mathematical Models for Tonal Segmentation
Tonal music consists of organized sounds that form vertical (synchronous) and horizontal (sequential) structures. Segmentation by tonality is an important precursor to proper labeling of these components for analysis and characterization. The Spiral Array model (Chew, 2000) clusters tonally important entities and allows tonal contexts to be determined computationally. We illustrate by separating bi-tonal compositions, determining key changes and characterizing tonal patterns.
Thomas Noll, T.U. Berlin.
Experiments with Lerdahl's Tonal Pitch Space Model
Fred Lerdahl's (2000) harmonic configuration space consists of 24 major and minor regions and chords within these regions. Harmonic pathways are calculated with respect to a principle of shortest path. The underlying distance combines a weakened hierarchical model and a shortest path principle in a mathematically problematic way. Therefore we experimentally compare two versions of this space: Lerdahl's original one, which does not satisfy the triangle inequality and a proper metric one.
SESSION 2. Mathematical Methods for Musical Design II
Session Chair: Charlotte Truchet, IRCAM
Anja Volk, T.U. Berlin. Investigations in Metric Structure Based on a Mathematical Model
This paper discusses a notion of metric coherence based upon a mathematical model describing the inner metric structure of a piece of music. Inner metric analysis studies the metric structure of the notes without considering the time signature and bar lines. It is opposed to outer metric analysis which refers to a presupposed regular structure of musical time. The notion of metric coherence describes the correspondences of varying degrees between the outer and inner metric structure.
Simon Dixon, Austrian Research Inst for AI.
Tempo induction, beat tracking and rhythm-based music classification
Tempo induction can be performed with autocorrelation of the raw or band-limited audio signal, or by onset detection followed by clustering of inter-onset intervals. Beat tracking methods include multiple hypothesis search, hidden Markov models and Kalman filtering. We review these methods and present recent work on classification of dance music based on distributions of periodicities and characteristic rhythmic patterns.
Charlotte Truchet, Gerard Assayag and Philippe Codognet, IRCAM. Musical Application of Adaptive Search, a Tabu Search Method for Solving CSPs
We present a new application area of constraint programming : music, precisely the field of Computer Assisted Composition. It deals with any symbolic representation of music, for instance at the score level. We have worked with contemporary composers on a dozen of musical CSPs, using a new heuristic method called Adaptive Search. For many reasons, local search techniques are well adapted to musical purposes. We have then designed and implemented a constraint programming system for musicians.
SESSION 3. Interactive Music Systems
Session Chair: Belinda Thom, Harvey Mudd College
Belinda Thom, Harvey Mudd Coll.
A machine learning based computational model for interactive musical improvisation
We present a melody representation scheme and machine learning framework for tightly coupling musicians with interactive software agents. A probabilistic model provides musician-specific perception, automatically mapping solos onto user "playing modes" that differentiate between various pitch class, intervallic, and melodic contour content. Random-walks through probabilistic graphs invert this perception procedure, automatically generating melodic responses to a user's solos in real-time.
Alexandre François & Elaine Chew, USC. Design for Real-Time Interactive Systems
Performer-centered systems require real-time processing and seamless interaction. We introduce SAI, a new framework for the design, implementation and analysis of real-time interactive applications. An open source architectural middleware, MFSM, complements SAI. We illustrate their use with MuSA.RT, an interactive environment for content-based music visualization.
Maverick Shih, USC. What is the title of that piece of music? An application of Query by Humming
Most people have had the experience of trying to find a piece of music in a music store with only salient tunes in mind. They typically do not have any information about the name of the composers and/or the performers. Humming and singing provide the most natural means for the music database retrieval. Can today's technologies help us to find the pieces that we are looking for? The technologies used by "Query by Humming" will be discussed in the presentation.
Please post and distribute this announcement freely. Last update: July 18, 2003
