Content-Based Indexing of Symbolic Music Documents - Intelligent Music Information Systems: Tools and Methodologies - page 27

Figure 7. Main components of a complete music retrieval engine where multiple indexing schemes are
combined with a data fusion technique

conclusIon

Yet there are a number of factors that need to
be taken into account when extending the indexing
concept to the music domain. First of all, the fact
that it is difficult to state which is the semantic
of a music document, if a semantic exists. Thus
the choice of which are the most representative
index terms have to be carried out with a different
approach. To this end, the concept of lexical units
of the music language has been introduced, taking
into account that music has a multidimensional
nature, and that not all the dimensions may be of
interest for the final user. Furthermore, it is not
clear to which extent the users agree on the way
they perceive lexical units.
To investigate this aspect, a perceptual study
has been carried out on the way a number of
musicians highlighted melodic lexical units of 20
excerpts of music scores. The analysis highlighted
that, even if there are some common trends in
user's behavior, the consistency among subjects
depends on the availability of particular cues in
the music documents. Even if subjects may not
agree when they refer to lexical units, their use
as index terms may be evaluated experimentally,
using a test collection of documents, queries

Indexing is based on the concept that documents
become more accessible if a number of guidance
tools are provided. This fact can be exploited to
improve the retrieval effectiveness, reducing its
computational cost because the content of a col-
lection of documents is accessed through a set of
pointers: instead of browsing all the documents to
find which are relevant to the user's information
need, the system may access only the ones that
are potentially relevant, depending on the set of
indexes that point to them. Indexing is the key
to scalability.
The application of indexing to music retrieval
is motivated by the need for a scalable system to
access music documents, because music collec-
tions are increasingly growing, both in digital
libraries systems at server side and in storage
devices at user side. Given that the main ideas
behind textual document indexing are quite
general, a parallelism can be drawn between the
phases of textual document indexing—namely,
lexical analysis, stop-words removal, stemming
and index weighting—and the phases that may
be required for music document indexing.