James Beckwith Maxwell
Hierarchical Sequential Memory for Music
“And yet the facts show that there are always new aerial dogs about; from which one must conclude that, despite what seem to our mind insurmountable obstacles, no species of dog, once extant, and no matter how curious it may be, ever dies out, or at least not easily, at least not without there being something in every species which puts up a long and successful resistance.”
- Franz Kafka, Investigations of a Dog
The MusicDB: A Music Database Query System for
Recombinance-based Composition in Max/MSP
To my horror and profound irritation, my recent paper titled “Hierarchical Sequential Memory for Music: A Cognitive Model” contained a couple of glaring errors. Such is life, I suppose. The corrections have been made and can be downloaded here:
JBM_HSMM_ISMIR_2009 - Revised.pdf
ABSTRACT
We propose a new machine-learning framework called the Hierarchical Sequential Memory for Music, or
HSMM. The HSMM is an adaptation of the Hierarchical Temporal Memory (HTM) framework, designed to make it better suited to musical applications. The HSMM is an online learner, capable of recognition, generation, continuation, and completion of musical structures.
A pdf of the ISMIR 2009 conference poster can be found here:
An updated paper on the HSMM, which details some changes and refinements made to the underlying algorithms can be found here:
ABSTRACT
We propose a design and implementation for a music information database and query system, the MusicDB,
which can be used for data-driven algorithmic composition. Inspired by David Cope's ideas surrounding composition by “music recombinance”, the MusicDB is implemented as a Java package, and is
loaded in MaxMSP using the mxj external. The MusicDB contains a music analysis module, capable of
extracting musical information from standard MIDI files, and a search engine. The search engine accepts
queries in the form of a simple six-part syntax, and can return a variety of different types of musical
information, drawing on the encoded knowledge of musical form stored in the database.