Understanding music is a fundamental challenge to artificial
intelligence.
It is a task that most humans are able to perform.
It shares some similarities with natural language understanding (both
extract understanding from a stream of data),
but is different in fundamental ways.
On the one hand, music does not require the same sort of general
real-world knowledge as natural language.
On the other hand, music is more highly structured, with multiple
layers, and with much repetition and variation.

Statistical modeling has revolutionized the area of natural language
understanding.  
Standard models such as hidden Markov models and context free
grammars, can be applied to music, but they fail to capture the
particular structural qualities of music.
This project seeks to develop probabilistic models of music, 
by creating models that capture our knowledge of musical form and
properties.