Recognition of Isolated Instrument Tones by Conservatory Students

1
Recognition of Isolated Instrument Tones by
Conservatory Students

• Asha Srinivasan, David Sullivan,
• and Ichiro Fujinaga
• Peabody Conservatory of Music
• Johns Hopkins University

2
Overview

• Background
• Aims
• Method
• Set-up of previous experiments
• Results
• Conclusions

3
Background

• Musicians have a remarkable ability to recognize
  instruments by timbre
• However, previous experiments using isolated
  tones suggest that recognition rates range
  between 36.5 and 90.0.
• Recently, timbre-recognition computer models have
  been able to match or exceed these rates.

4
Aims

• Verify previous experiments
• Measure the effect of ensemble experience
• Generate more detailed baseline data to help
  evaluate computer performance

5
Method

• Eighty-eight subjects participated in the
  experiment. They were undergraduate ear-training
  students (66), composition students (19), and
  faculty (3).
• Personal information was collected
• gender, degree/year, major, primary instrument,
  of years formal training, orchestral/band
  experience, compositional/conducting experience,
  perfect pitch, of years ear-training
• All tones were taken from the McGill University
  Master Samples.

6
The Tests

• Two tests were performed
• The first test included four sections, involving
  2, 3, 9, and 27 instruments.
• In the second test, short training sessions
  preceded each section, involving 2, 9, and 27
  instruments.

7
Training sessions

• Ex announce oboe, play 2 - 3 oboe samples
  announce sax, play 2-3 sax samples
• The 27-instrument sessions were grouped by family
  and by similar sound

8
List of Instruments
9
List of Instruments
10
Previous experiments and Peabody
11
Recognition rates for previous human experiments
12
Overview of Results

• Comparison of previous experiments and Peabody
• Family groupings
• Comparison of different groups of Peabody
  subjects
• Piano, Guitar, Voice (PGV) students vs. Non-PGV
  students
• Effect of the short-term training sessions

13
Recognition rates for previous human experiments
and Peabody results
14
Previous computer experiments
15
Recognition rates for previous computer and human
experiments and Peabody
16
Confusion matrix (2-instr. 3-instr.)
17
Confusion matrix (9-instr.)
18
Confusion matrix (3D-View)
19
Confusion matrix comparison
20
Confusion matrix (27-instr.)
21
Confusion matrix (3D-View)
22
Confusion matrix (Martin)
23
Confusion matrix (Family grouping for 9-instr.
27-instr.)
24
Confusion matrix comparison
25
Family vs. Exact Answers
26
Recognition rates for ear-training students,
composition students, and faculty
27
Piano, Guitar, Voice (PGV) students vs. Non-PGV
students
28
Effects of training on ear-training (47)and
composition (6) subjects
29
Conclusions

• Compared to previous experiments, the average
  scores of subjects in this experiment were
  considerably higher.
• Subjects who play orchestral instruments tended
  to score higher than those who do not.
• The short-term training sessions had a
  significant effect on the subjects performance
  for the 27-instrument test only.
• The excellent average score of the human subjects
  in this experiment presents new challenges for
  timbre-recognition computer models.