Information Technology Reference
In-Depth Information
were unpaid volunteers who were undergraduate
or graduate students at Rensselaer Polytechnic
Institute in Troy, New York. At the beginning of
each session, the subjects were asked about the
amount of training and experience each had with
musical instruments and with singing. Subjects
were also asked informally about their familiarity
with the tunes selected for the experiment.
Five of the subjects had more than five years
experience with one or more musical instru-
ments; the mean for this group was 14.8 years.
In the reported results for each experiment, we
referred to them collectively as the musician
group, abbreviated as MUSI. Of the remain-
ing ten subjects, six had between two and five
years musical experience (mean 3.2 years); their
results were grouped together as the inbetween
(BTWN) group. The final four subjects had less
than two years experience and were referred to
as the nonmusician or NONM group. Five of
the subjects were female, and they were evenly
distributed among the groups with two musicians,
two nonmusicians and one in the BTWN group.
None of the subjects had formal voice training,
though three of them had experience singing in
volunteer community choirs. Of these, two were
in the MUSI group and one was a nonmusician
who had recently joined a choir.
cassette deck. The recorded input hummed by
the subjects was digitized using Wildcat Canyon
Software Autoscore 2.0 Professional music tran-
scription software after the sessions.
All musical sequences played to the subjects
during the course of the study were played out of
tune relative to the key signature in which they
were written. The degree to which the notes
were out of tune was a randomly selected value
in the range ± 100 cents. This technique was
commonly used in the previously cited studies in
order to eliminate any tonal basis among the trials.
To facilitate accurate note segmentation by the
transcription software, subjects were instructed
to voice each note using the syllable da rather
than by humming. This provides a consonant
stopping sound between successive notes, which
is another method Autoscore employs to improve
transcription accuracy.
experiment 1: humming familiar
tunes from memory
The first two experiments explored how subjects
hummed songs that were familiar to them. Several
factors were taken into consideration when select-
ing the songs to be used for these two experiments.
Many were selected from the Digital Tradition
folksong collection (Greenhaus et al., n.d.). This
collection was first used by McNab et al. (1997) as
one of two components of their test database, and
it has subsequently been used in other research
as well, for example, Downie and Nelson (2000).
Among our considerations in selecting the set of
stimulus songs were these:
apparatus
All music played to the subjects during the course
of the experimental trials was generated by a
midrange sound card and played through inex-
pensive computer speakers (typically owned by
home users) at a comfortable listening level in a
quiet office. Almost all of the musical samples
were synthesized MIDI renditions, though a few
of the stimuli were presented as digitized audio
(.wav) files made from original recordings.
The subjects' responses were recorded with
an inexpensive computer microphone (again,
equipment of the quality typically provided with
computers used in the home) onto a medium-grade
Several of the songs chosen, such as Twinkle,
Twinkle, Little Star (d02) and Yankee Doodle
(d08), contain sequences in which consecu-
tive notes continually increase or decrease
in pitch by only one or two semitones.
Care was taken to ensure that a wide variety
of pitch interval sizes were represented in
the selected tunes. For instance, Take Me
Search WWH ::




Custom Search