Information Technology Reference
In-Depth Information
Chapter III
Melodic Query Input for Music
Information Retrieval Systems
Richard L. Kline
Pace University, USA
aBstract
This chapter discusses key issues of building and using a system designed to search and query a music
collection through the input of the actual or perceived melody of a song. It presents a detailed survey
and discussion of studies, algorithms and systems created to approach this problem, including new
contributions by the author. Emphasis is placed on examining the abilities and likely errors of those
with little or no formal musical training to remember and reproduce melodic phrases, as these must be
taken into account for any music information retrieval system intended for use by the general public.
The chapter includes an extensive discussion on human humming as an intuitive input method for the
musically untrained person and an examination of several music information retrieval systems designed
for this type of input query.
IntroductIon
rhythm contained therein, such as that presented
in written form or in formats such as MIDI (1993),
rather than raw or processed digitized audio
samples such as CD or MP3. In this chapter we
will present many of the most important theories,
algorithms, techniques and performed studies for
capturing, representing and performing similar-
ity or identification searches using melodic input
query strings.
MIR systems requiring the highest search effi-
ciencies generally must demand that the collection
An effective music information retrieval (MIR)
system requires suitable mechanisms for speci-
fying input queries. For queries of the nature of
identifying a tune, theme or melody, providing
query input in the form of a melody line or rhythm
usually will be the most intuitive and powerful
means of specifying such a search. Similarly, col-
lections to be searched in this manner must have
some available representation of the melody and
Search WWH ::
Custom Search