Information Technology Reference
In-Depth Information
to include the testing of larger pitch intervals. We
also used a significantly larger group of subjects
who possessed a much wider range of musical
abilities.
In the first two experiments the music tracking
software was run in its default
Constrain
mode of
identifying the nearest semitone to the perceived
frequency. For this experiment, in order to dupli-
cate the reporting done by Lindsay,
Autoscore
was
used in its
pitch bend
mode, in which it measures
and reports the precise sound frequency many
times per second.
Autoscore
typically reported
about 20 distinct pitch-bend values per hummed
note; these were averaged to come up with the
pitch value we recorded.
Method
The stimulus data set was a series of 32 five-note
sequences. The notes were all of equal duration,
played at the rate of three notes per second. This
is slightly slower than the four note-per-second
rate used by Lindsay; the change was made after
the faster tempo was tried in a few informal trials
with nonmusicians before the study. We dupli-
cated 17 of Lindsay's original sequences and the
requirement that exactly 2 sequences represent
each of the 16 possible five-note binary pitch
contours. The remaining 15 sequences originally
contained 1 or more instances of a six-semitone
interval. This dissonant interval occurs rarely
in Western popular and folk music because it is
perceived as sounding unpleasant to many ears,
and as a result, it is difficult for an untrained voice
to reproduce it. Wherever this interval occurred,
it was replaced equally by intervals of 9, 10 and
12 semitones. The resulting set of 32 sequences
had approximately 9 intervals each of
±
1, 2, 3,
4, 5 and 7 semitones, and 3 each of
±
9, 10 and
12 semitones.
The sequences were presented in a pseu-
dorandom order so that no trial was presented
near another with similar contour or its opposite
contour. Half of the subjects heard the sequences
in one order, while the other half were presented
them in exact reverse order, thus reducing the
influence one trial has on the trial that followed it
yet retaining the careful ordering based on contour
distances. Subjects listened to each sequence in
turn, and then vocalized the sequence just heard.
They could begin humming whenever they were
ready. Subjects were invited to request a short
break after the first half of the trials if they wished
it; seven of them did so.
Results
In several instances, the transcription process did
not correctly output the exact five distinct notes
hummed by the subjects. Most often, a single
note was incorrectly identified as multiple notes at
one or more pitch values. Each of these cases was
edited by hand, comparing the transcription to the
original recording to ensure the correct pitch values
were retained. In a small number of instances,
an individual trial had to be thrown out because
of transcription difficulties or user errors such as
humming only four notes. One of the subjects
found this experiment too difficult to complete,
so none of her results are included here.
Again here as before, we report results for
subjects in three groups: musicians (MUSI), inex-
perienced musicians (BTWN) and nonmusicians
(NONM). The MUSI group corresponds to the
five musician subjects of Lindsay's study, while
the BTWN group appears to correspond to the
description of Lindsay's nonmusician subject.
Overall, the musician subjects in our study
performed substantially worse than their counter-
parts in the Lindsay study, matching the intended
sequence using pitch interval data 65% of the
time, and giving the correct pitch contours in
only 81% of the cases. The scores for BTWN
and NONM were nearly identical to one another
and were also very close to those of Lindsay's
less experienced subject: combining these two
groups, 48% of the trials matched correctly when
Search WWH ::
Custom Search