Computer
Audition
Computer Audition (CA) is a general field of audio understanding
by machine that encompasses questions of audio processing, synthesis,
information retrieval, auditory scene analysis and machine listening. Inspired by
models of human audition, it deals with questions of representation,
transduction, grouping, use of musical knowledge and general sound semantics
for the purpose of performing intelligent operations on audio and music signals
by the computer. Moreover, the tasks of a computer audition system go beyond
questions of classification or retrieval, often using audition for performing
intelligent audio activities, such as audition driven sound processing and
sound or music generation.
Accordingly, the research on computer audition builds upon
psychological and cognitive evidence of human listening experience, combining
disciplines of engineering, information processing and artificial intelligence,
cognitive science, music theory and artistic creativity, making it a formidable
interdisciplinary study.
The study of CA could be roughly divided into three main areas:
1.
Representation: signal and symbolic. This aspect deals with
feature extraction, sound descriptors and auditory models. It also concerns
with audio analysis-synthesis and generative models, such as pattern playback
or signal recreation from partial representation.
2.
Signal alignment and comparison: One of the unique
properties of musical signals is that they often combine different types of
representation, from notated score to performance actions in midi files, to
audio recordings and human annotations. We study methods for finding such
correspondences, with applications for intelligent sound processing, performing
with computers, automatic annotation and more.
3.
Musical Knowledge and Audio Semantics: many aspects of topics
1 and 2 depend on human cognitive processing, such as perception of scales,
rhythms and harmonies, and up to modeling of emotions, musical memory and
perception of musical structure. We use machine learning to model human
cognitive modalities of anticipation, familiarity and appraisal, and use them
to describe musical style with applications to machine improvisation and
building of an intelligent musical assistant.
To read
more about CA, check out the Computer Audition
tutorial at ACM
Multimedia, MM'06, October 23, 2006, Santa Barbara, California, USA.
Some Matlab
CA tools can be found in CATbox
Visit also
our new Computer Audition Lab and
read about Music Information
Processing.