Information Technology Reference
In-Depth Information
of semantic labels, are a very compact represen-
tation of the document content, because they
can summarize a complete document with few
keywords. Though not the main subject of this
discussion, it is hence worth spending few words
on music indexing through metadata and on its
limitations for an efficient and effective retrieval
task. A number of music digital libraries are ac-
cessible through the use of metadata. For instance,
Cantate (2006) and Musica (2006) allow users to
access to choral music using metadata and lyrics.
Another project based on the use of metadata is
Jukebox (Harvell & Clark, 1995).
As for many other media, music metadata
addresses different characteristics of document
content. In the particular case of music, it can be
roughly divided in three categories:
some music features also in other music genres:
terms like “bossa”, “waltz”, and “blues” appear
often in jazz compositions with that particular
feature, and terms like “jig” and “reel” are often
part of the title of the respective dances in Irish
music tradition.
General information is often too generic to
be a good discriminator between different music
works. For example in tonal music there are only
21 major and 21 minor different tonalities, while
thousands of compositions of tonal Western music
can be labelled with the term “cantata” or “con-
certo”, and the same applies with terms such as
“up tempo” or “slow” for pop and rock genres. The
genre information itself groups together hundreds
of thousands of different works. Another problem
that arises with metadata on general information
is that the terminology is not consistent across
genres and historical periods. For example, the
term “sonata” has different meanings for Baroque
and Romantic repertoires and the term “ballad”
refers to different characteristics in jazz and in
folk music. This kind of metadata can be useful
to refine the description of a music information
need, but it can hardly be used to completely
define it. Moreover, a preliminary study on users
information needs (Lee & Downie, 2004) showed
that users are interested in retrieving songs by
their specific content.
Additional information in the form of lyr-
ics, when present, can be particularly useful to
describe an information need, yet in this case
the retrieval of music documents becomes an ap-
plication of textual IR. Contextual information,
such as the movie where a particular soundtrack
has been used, or the poem that inspired a par-
ticular composition, can be very helpful as well
to describe a user information need. In many
cases the information need is motivated by the
contextual information itself—that is, a user may
be searching for the theme song of a TV series
or for the music of a known ballet—yet this kind
of contextual information applies only to a small
percentage of music documents.
Bibliographic values: Suthor's name,
performer's name (in the case of audio re-
cordings), title, year of publication, editor,
cataloguing number.
General information on document con-
tent: Time and key signatures, musical form,
structure, music genre, orchestration.
Additional available information: Lyrics
and, if applicable, related documents that
create a context for the music work (e.g., a
drama, a movie, a poem).
The search through bibliographic values can
be very effective in terms of retrieval effective-
ness, even if in this case a database approach to
match exact values in predefined fields would be
more suitable than an IR approach. On the other
hand, the user is required to have a good knowl-
edge of the music domain, being able to clearly
describe the documents of interest. In the case
of tonal Western music, the title of a music work
is often nondescriptive, describing part of the
general information on document content, and
it is typical to have titles such as “Sonata in D
flat, Op. 5” or “Fugue”. The title can be based on
Search WWH ::




Custom Search