Information Technology Reference
In-Depth Information
were able to remain in key and whether their pitch
contours remained correct over the course of their
performances. Subjects were given time to prac-
tice each song before it was recorded, and they
were asked to sing as much as they knew of the
songs. They discovered that their subjects were
able to sing at least one phrase of each song with
two or fewer pitch contour errors, and that small
pitch intervals were often exaggerated while large
ones were often compressed. They described and
analyzed how songs with large or unusual pitch
intervals were much more likely to cause subjects
to end their performance out of key. They also
noted that singing experience played a larger factor
than overall music training in the performance of
the subjects. Unfortunately, most of these results
were reported only anecdotally, with data reported
only for how well subjects ended their singing in
the same key in which they started.
The most recent study was performed by Unal,
Narayanan, Shih, Chew and Kuo (2003) as part of
their efforts to collect a corpus of hummed input
queries from a variety of subjects. It is intended
that their collection will be made available for
further study, analysis and use in testing other
MIR systems.
The collective insights of all of these stud-
ies paint a picture of our intended user, but not
enough of the results apply to the vocal skills of
the musically untrained. They provided the direc-
tion and many good ideas which we drew upon to
develop and carry out our own study, which we
describe in the next section. In addition to the
intrinsic value of the knowledge gained through
our experiments, we used this opportunity also to
collect a significant body of sample queries with
which to evaluate our later efforts at developing
effective new MIR matching algorithms.
In summary, these are some of the important
shortcomings of the available body of research
which we attempted to address in the design and
execution of our own study:
•
The majority of the studies available from
psychology and music cognition sources
focus more on perception and recognition
of music rather than its reproduction.
•
Studies from these sources on music repro-
duction skills have typically been restricted
to certain individual aspects such as select-
ing a starting pitch or tempo. Many of these
have used instruments or other devices rather
than subjects' voices.
•
Some studies specifically targeting human
humming ability have been performed
anecdotally or used very small subject sets
with a disproportionate number of highly
skilled musicians represented.
our studY of human hummIng
skIlls
Our study consisted of four separate experiments
in which subjects were asked to hum both familiar
and unfamiliar melodies and rhythms. The experi-
ments were conducted together under the same
conditions and with the same group of subjects.
For the first experiment, subjects hummed songs
whose titles and lyrics were presented to them
on paper. In the second, subjects first listened
to recordings or MIDI renditions of songs, and
then hummed them. The third experiment was
a modified version of a study by Lindsay (1996)
in which subjects heard 32 unfamiliar five-note
phrases and hummed each one in turn. The fourth
was a modified version of the Drake (1993) study
on reproduction of rhythms. Space considerations
allow us here to report on only the first and third
experiments of the study, where most of our sig-
nificant findings were discovered.
subjects
A group of 15 subjects, all within the ages of
18-35, participated in the study. The subjects
Search WWH ::
Custom Search