Information Technology Reference
In-Depth Information
than the NONM group. When considering all
trials together, musicians chose starting pitches
2.1 semitones higher than BTWN subjects and
2.4 semitones higher than nonmusicians.
Perhaps more important than the actual starting
pitches is the similarity of choices across subjects
for a given song. In examining the starting pitches
for a given song across all subjects, in most cases
the standard deviation was quite small. When
considering all subjects for a single song, the
performances which resulted in the lowest stan-
dard deviation included both songs with small
overall pitch ranges (i.e., d02, d08) and those with
large ranges (i.e., d01, d05). In the same way, the
songs with the largest deviation fell into both of
these categories. Overall, four of the songs had
a standard deviation below 2.0, while two songs
had values over 3.0, and the remaining six were
in between.
When considering performance by subject
group, the musicians showed the highest variabil-
ity in pitch selection at 3.53, while the BTWN and
NONM groups had standard deviations of 2.66
and 2.69, respectively. Only two subjects had vari-
ability below 2.0, one of who had the extremely
low value of 1.22: this subject chose nearly the
same starting pitch for every performance.
We meticulously compared the audio record-
ings of each subject's rendition of Take Me Out
To The Ballgame (d01) with the automated tran-
scriptions produced by
Autoscore
to determine
the number of errors introduced through the pitch
tracking process. On average, subjects hummed
26.5 notes of the melody. When compared to our
manual transcriptions, the
Autoscore
transcrip-
tions contained an average of 3.75 extraneous
notes, while also failing to record 2.2 of the notes
which were hummed. The transcription exactly
matched the vocal rendition for only one subject.
At least one extraneous note was introduced to
the automatic transcription in 13 of 15 trials, and
10 of the 15 subjects' transcriptions were missing
at least one hummed note clearly audible in the
recording.
The difference between note durations and in-
ternote onset times was noticeable for all subjects
and all groups. Overall, the average hummed note
had a duration value 74% of the length of its cor-
responding INOT. Table 3 shows as percentages
the ratio of note duration to INOT for all notes
hummed by all subjects in the experiment. The
values are shown categorized for each of the three
subject groups, as well as in groups based on the
INOT time value (reported in milliseconds). The
different columns for each row give the percent-
age for all notes whose INOT values were in the
ranges shown. The cutoff values were chosen
based on the mean INOT value of 313ms for all
notes in all trials. As expected, longer notes gener-
ally displayed a larger difference between the two
values (both in absolute time and as a percentage)
than exhibited by shorter notes.
Table 2b. Average starting pitch selection and
standard deviation, tabulated by subject, from
Experiment 1
Subject Avg.Pitch Std.Dev.
1 48.83 3.71
2 46.33 2.71
3 47.30 3.27
12 50.60 1.96
14 57.08
3.26
MUSI 47.54 3.53
7 47.83 2.25
8 46.10 2.77
9 56.92
2.43
11 44.64 2.54
13 55.91
1.22
15 45.18 2.96
BTWN 45.45 2.66
4 44.33 2.77
5 57.17
2.52
6 45.83 2.98
10 57.08
2.61
NONM
45.10 2.69
ALL 46.04 3.15
Notes:Pitch values are given as MIDI note
numbers;Female subjects are shown in italics.
Search WWH ::
Custom Search