Information Technology Reference
In-Depth Information
In the past few years, interest in music informa-
tion retrieval (MIR) has been steadily increasing.
MIR algorithms, especially when analyzing music
signals in audio format, typically utilize state-of-
the-art signal processing and machine learning
algorithms. The large amounts of data that are
processed together with the huge computational
requirements of audio processing can stress cur-
rent hardware to its limits. Therefore efficient
processing is critical for building functional MIR
systems that scale to large collections of music
and eventually to all of recorded music. Moreover,
MIR is an inherently interdisciplinary field with
practitioners with varying degrees of computer
and programming expertise (examples of fields
involved include musicology, information science
and cognitive psychology). Therefore it is desirable
for MIR systems to support multiple hierarchical
levels of usage and extensibility. These issues
make the design and development of MIR systems
and frameworks especially challenging.
MARSYAS
(
M
usic
A
nalysis,
R
etrieval and
SY
nthesis for
A
udio
S
ignals) is an open source
audio processing framework with specific em-
phasis on building MIR systems. It has been
under development since 1998 and has been
used for a variety of projects both in academia
and industry. The guiding principle behind the
design of
MARSYAS
has always been to provide
a flexible, expressive and extensive framework
without sacrificing computational efficiency.
Addressing these conflicting requirements is the
major challenge facing the software engineer of
MIR systems.
The main objective of this chapter is to describe
the software architecture of
MARSYAS
using
examples from specific MIR applications. We
highlight the design challenges and corresponding
solutions that are probably relevant to any MIR
software system. In many cases these solutions
are informed by ideas originating from other fields
of computer science and software engineering but
have to be adapted to the particular needs and con-
straints of MIR research. After reviewing related
work and background information,
MARSYAS
is
described in the following subsections: History,
Requirements, Architecture and Projects. The
next section (Specific Topics) describes in more
detail specific topics that we believe are especially
important for audio processing and how we have
tried to address them in the framework. The
chapter concludes with a description of future
trends in
MARSYAS
and audio processing soft-
ware frameworks in general. One of the major
dilemmas facing any MIR researcher is whether
to use existing tools or develop their own. By
describing the tradeoffs and challenges we have
faced with the design of our system we hope to
help researchers make more informed decisions.
Finally an underlying theme of this chapter is the
importance of open source software for research
and how it is different from other areas of open
source development.
Background
Music Information Retrieval is a new area of
content-based multimedia information retrieval.
Although there was sporadic earlier work, a good
reference starting point is the first international
conference on MIR (ISMIR) which was held in
2000. These conferences (ISMIR) have been a
forum for bringing together music researchers,
audio engineers, computer scientists, musicolo-
gists, librarians and the music industry (Futrelle &
Downie, 2002). MIR with audio signals typically
requires signal processing and machine learning
algorithms in order to achieve tasks such as classi-
fication, similarity-retrieval and segmentation.
MARSYAS 0.2
, the software framework for
audio analysis and synthesis described in this
chapter, evolved from
MARSYAS 0.1
(Tzanetakis
& Cook, 2000), a framework that focused mostly
on audio analysis. One of the motivating factors
for the rewrite of the code and architecture was the
desire to add audio synthesis capabilities and was
influenced by the design of the Synthesis Toolkit
Search WWH ::
Custom Search