2008:Audio Classical Composer Identification Results

From MIREX Wiki

Introduction

These are the results for the 2008 running of the Audio Classical Composer Identification task. For background information about this task set please refer to the 2007:Audio Classical Composer Identification page.

The data set consisted of 2772 30 second audio clips. The composers represented were:

  1. Bach
  2. Beethoven
  3. Brahms
  4. Chopin
  5. Dvorak
  6. Handel
  7. Haydn
  8. Mendelssohnn
  9. Mozart
  10. Schubert
  11. Vivaldi

The goal was to correctly identify the composer who wrote each of the pieces represented.


General Legend

Team ID

GP1 = G. Peeters
GT1 = G. Tzanetakis
GT2 = G. Tzanetakis
GT3 = G. Tzanetakis
LRPPI1 = T. Lidy, A. Rauber, A. Pertusa, P. Peonce de León, J. M. Iñesta 1
LRPPI2 = T. Lidy, A. Rauber, A. Pertusa, P. Peonce de León, J. M. Iñesta 2
LRPPI3 = T. Lidy, A. Rauber, A. Pertusa, P. Peonce de León, J. M. Iñesta 3
LRPPI4 = T. Lidy, A. Rauber, A. Pertusa, P. Peonce de León, J. M. Iñesta 4
ME1 = M. I. Mandel, D. P. W. Ellis 1
ME2 = M. I. Mandel, D. P. W. Ellis 2
ME3 = M. I. Mandel, D. P. W. Ellis 3

Overall Summary Results

MIREX 2008 Audio Classical Composer Classification Summary Results - Raw Classification Accuracy Averaged Over Three Train/Test Folds

Participant Average Classifcation Accuracy
GP1 48.99%
GT1 39.47%
GT2 45.82%
GT3 43.81%
LRPPI1 34.13%
LRPPI2 39.43%
LRPPI3 37.48%
LRPPI4 39.54%
ME1 53.25%
ME2 53.10%
ME3 52.89%

download these results as csv

Accuracy Across Folds
Classification fold GP1 GT1 GT2 GT3 LRPPI1 LRPPI2 LRPPI3 LRPPI4 ME1 ME2 ME3
0 0.501 0.363 0.452 0.457 0.165 0.386 0.379 0.389 0.545 0.538 0.532
1 0.483 0.415 0.464 0.424 0.431 0.406 0.369 0.395 0.523 0.525 0.527
2 0.486 0.407 0.458 0.433 0.429 0.391 0.377 0.403 0.529 0.530 0.527

download these results as csv

Accuracy Across Categories
Class GP1 GT1 GT2 GT3 LRPPI1 LRPPI2 LRPPI3 LRPPI4 ME1 ME2 ME3
bach 0.667 0.516 0.651 0.571 0.563 0.575 0.500 0.583 0.734 0.738 0.738
beethoven 0.321 0.409 0.548 0.425 0.198 0.310 0.266 0.282 0.393 0.385 0.393
brahms 0.290 0.159 0.198 0.230 0.198 0.321 0.310 0.310 0.429 0.429 0.433
chopin 0.913 0.885 0.897 0.810 0.595 0.663 0.627 0.659 0.770 0.774 0.766
dvorak 0.417 0.333 0.488 0.484 0.393 0.369 0.393 0.361 0.456 0.444 0.448
handel 0.492 0.310 0.302 0.321 0.397 0.425 0.377 0.405 0.548 0.548 0.544
haydn 0.655 0.488 0.651 0.556 0.369 0.413 0.397 0.460 0.603 0.607 0.603
mendelssohnn 0.337 0.472 0.492 0.401 0.286 0.401 0.306 0.341 0.488 0.472 0.468
mozart 0.266 0.087 0.079 0.242 0.234 0.246 0.274 0.214 0.353 0.353 0.345
schubert 0.302 0.194 0.230 0.210 0.175 0.218 0.262 0.278 0.417 0.425 0.425
vivaldi 0.730 0.488 0.504 0.567 0.345 0.397 0.413 0.456 0.667 0.667 0.655

download these results as csv

MIREX 2008 Audio Classical Composer Classification Evaluation Logs and Confusion Matrices

MIREX 2008 Audio Classical Composer Classification Run Times

Participant Runtime (hh:mm) / Fold
GP1 Feat Ex: 04:40 Train/Classify: 00:47
GT1 Feat Ex/Train/Classify: 00:16
GT2 Feat Ex/Train/Classify: 00:34
GT3 Feat Ex: 00:05 Train/Classify: 00:00 (7 sec)
LRPPI1 Feat Ex: 08:00 Train/Classify: 00:02
LRPPI2 Feat Ex: 08:00 Train/Classify: 00:09
LRPPI3 Feat Ex: 08:00 Train/Classify: 00:09
LRPPI4 Feat Ex: 08:00 Train/Classify: 00:14
ME1 Feat Ex: 1:17 Train/Classify: 00:00 (21 sec)
ME2 Feat Ex: 1:17 Train/Classify: 00:00 (21 sec)
ME3 Feat Ex: 1:17 Train/Classify: 00:00 (21 sec)

download these results as csv

CSV Files Without Rounding

audiocomposer_results_fold.csv
audiocomposer_results_class.csv

Results By Algorithm

(.tar.gz)
GP1 = G. Peeters
GT1 = G. Tzanetakis
GT2 = G. Tzanetakis
GT3 = G. Tzanetakis
LRPPI1 = T. Lidy, A. Rauber, A. Pertusa, P. Peonce de León, J. M. Iñesta 1
LRPPI2 = T. Lidy, A. Rauber, A. Pertusa, P. Peonce de León, J. M. Iñesta 2
LRPPI3 = T. Lidy, A. Rauber, A. Pertusa, P. Peonce de León, J. M. Iñesta 3
LRPPI4 = T. Lidy, A. Rauber, A. Pertusa, P. Peonce de León, J. M. Iñesta 4
ME1 = I. M. Mandel, D. P. W. Ellis 1
ME2 = I. M. Mandel, D. P. W. Ellis 2
ME3 = I. M. Mandel, D. P. W. Ellis 3

Friedman's Test for Significant Differences

Classes vs. Systems

The Friedman test was run in MATLAB against the average accuracy for each class.

Friedman's Anova Table
Source SS df MS Chi-sq Prob>Chi-sq
Columns 581.36 10 58.1364 53.09 7.16E-08
Error 623.14 100 6.2314
Total 1204.5 120

download these results as csv

Tukey-Kramer HSD Multi-Comparison

The Tukey-Kramer HSD multi-comparison data below was generated using the following MATLAB instruction. Command: [c, m, h, gnames] = multicompare(stats, 'ctype', 'tukey-kramer', 'estimate', 'friedman', 'alpha', 0.05);

TeamID TeamID Lowerbound Mean Upperbound Significance
GP1 GT1 -4.6324 -0.0909 4.4506 FALSE
GP1 GT2 -4.1779 0.3636 4.9051 FALSE
GP1 GT3 -3.2233 1.3182 5.8597 FALSE
GP1 LRPPI1 -2.5415 2.0000 6.5415 FALSE
GP1 LRPPI2 -1.6324 2.9091 7.4506 FALSE
GP1 LRPPI3 -0.1324 4.4091 8.9506 FALSE
GP1 LRPPI4 0.0949 4.6364 9.1779 TRUE
GP1 ME1 -0.3597 4.1818 8.7233 FALSE
GP1 ME2 0.6403 5.1818 9.7233 TRUE
GP1 ME3 2.0494 6.5909 11.1324 TRUE
GT1 GT2 -4.0870 0.4545 4.9961 FALSE
GT1 GT3 -3.1324 1.4091 5.9506 FALSE
GT1 LRPPI1 -2.4506 2.0909 6.6324 FALSE
GT1 LRPPI2 -1.5415 3.0000 7.5415 FALSE
GT1 LRPPI3 -0.0415 4.5000 9.0415 FALSE
GT1 LRPPI4 0.1858 4.7273 9.2688 TRUE
GT1 ME1 -0.2688 4.2727 8.8142 FALSE
GT1 ME2 0.7312 5.2727 9.8142 TRUE
GT1 ME3 2.1403 6.6818 11.2233 TRUE
GT2 GT3 -3.5870 0.9545 5.4961 FALSE
GT2 LRPPI1 -2.9051 1.6364 6.1779 FALSE
GT2 LRPPI2 -1.9961 2.5455 7.0870 FALSE
GT2 LRPPI3 -0.4961 4.0455 8.5870 FALSE
GT2 LRPPI4 -0.2688 4.2727 8.8142 FALSE
GT2 ME1 -0.7233 3.8182 8.3597 FALSE
GT2 ME2 0.2767 4.8182 9.3597 TRUE
GT2 ME3 1.6858 6.2273 10.7688 TRUE
GT3 LRPPI1 -3.8597 0.6818 5.2233 FALSE
GT3 LRPPI2 -2.9506 1.5909 6.1324 FALSE
GT3 LRPPI3 -1.4506 3.0909 7.6324 FALSE
GT3 LRPPI4 -1.2233 3.3182 7.8597 FALSE
GT3 ME1 -1.6779 2.8636 7.4051 FALSE
GT3 ME2 -0.6779 3.8636 8.4051 FALSE
GT3 ME3 0.7312 5.2727 9.8142 TRUE
LRPPI1 LRPPI2 -3.6324 0.9091 5.4506 FALSE
LRPPI1 LRPPI3 -2.1324 2.4091 6.9506 FALSE
LRPPI1 LRPPI4 -1.9051 2.6364 7.1779 FALSE
LRPPI1 ME1 -2.3597 2.1818 6.7233 FALSE
LRPPI1 ME2 -1.3597 3.1818 7.7233 FALSE
LRPPI1 ME3 0.0494 4.5909 9.1324 TRUE
LRPPI2 LRPPI3 -3.0415 1.5000 6.0415 FALSE
LRPPI2 LRPPI4 -2.8142 1.7273 6.2688 FALSE
LRPPI2 ME1 -3.2688 1.2727 5.8142 FALSE
LRPPI2 ME2 -2.2688 2.2727 6.8142 FALSE
LRPPI2 ME3 -0.8597 3.6818 8.2233 FALSE
LRPPI3 LRPPI4 -4.3142 0.2273 4.7688 FALSE
LRPPI3 ME1 -4.7688 -0.2273 4.3142 FALSE
LRPPI3 ME2 -3.7688 0.7727 5.3142 FALSE
LRPPI3 ME3 -2.3597 2.1818 6.7233 FALSE
LRPPI4 ME1 -4.9961 -0.4545 4.0870 FALSE
LRPPI4 ME2 -3.9961 0.5455 5.0870 FALSE
LRPPI4 ME3 -2.5870 1.9545 6.4961 FALSE
ME1 ME2 -3.5415 1.0000 5.5415 FALSE
ME1 ME3 -2.1324 2.4091 6.9506 FALSE
ME2 ME3 -3.1324 1.4091 5.9506 FALSE

download these results as csv

2008 composer.perclassaccuracy.friedman.tukeykramerhsd.png

Folds vs. Systems

The Friedman test was run in MATLAB against the accuracy for each fold.

Friedman's Anova Table
Source SS df MS Chi-sq Prob>Chi-sq
Columns 296 10 29.6 26.91 0.0027
Error 34 20 1.7
Total 330 32

download these results as csv

Tukey-Kramer HSD Multi-Comparison

The Tukey-Kramer HSD multi-comparison data below was generated using the following MATLAB instruction. Command: [c, m, h, gnames] = multicompare(stats, 'ctype', 'tukey-kramer', 'estimate', 'friedman', 'alpha', 0.05);

TeamID TeamID Lowerbound Mean Upperbound Significance
GP1 GT1 -9.0495 -0.3333 8.3828 FALSE
GP1 GT2 -8.3828 0.3333 9.0495 FALSE
GP1 GT3 -6.7162 2.0000 10.7162 FALSE
GP1 LRPPI1 -5.3828 3.3333 12.0495 FALSE
GP1 LRPPI2 -4.7162 4.0000 12.7162 FALSE
GP1 LRPPI3 -2.0495 6.6667 15.3828 FALSE
GP1 LRPPI4 -2.0495 6.6667 15.3828 FALSE
GP1 ME1 -1.7162 7.0000 15.7162 FALSE
GP1 ME2 -0.3828 8.3333 17.0495 FALSE
GP1 ME3 -2.7162 6.0000 14.7162 FALSE
GT1 GT2 -8.0495 0.6667 9.3828 FALSE
GT1 GT3 -6.3828 2.3333 11.0495 FALSE
GT1 LRPPI1 -5.0495 3.6667 12.3828 FALSE
GT1 LRPPI2 -4.3828 4.3333 13.0495 FALSE
GT1 LRPPI3 -1.7162 7.0000 15.7162 FALSE
GT1 LRPPI4 -1.7162 7.0000 15.7162 FALSE
GT1 ME1 -1.3828 7.3333 16.0495 FALSE
GT1 ME2 -0.0495 8.6667 17.3828 FALSE
GT1 ME3 -2.3828 6.3333 15.0495 FALSE
GT2 GT3 -7.0495 1.6667 10.3828 FALSE
GT2 LRPPI1 -5.7162 3.0000 11.7162 FALSE
GT2 LRPPI2 -5.0495 3.6667 12.3828 FALSE
GT2 LRPPI3 -2.3828 6.3333 15.0495 FALSE
GT2 LRPPI4 -2.3828 6.3333 15.0495 FALSE
GT2 ME1 -2.0495 6.6667 15.3828 FALSE
GT2 ME2 -0.7162 8.0000 16.7162 FALSE
GT2 ME3 -3.0495 5.6667 14.3828 FALSE
GT3 LRPPI1 -7.3828 1.3333 10.0495 FALSE
GT3 LRPPI2 -6.7162 2.0000 10.7162 FALSE
GT3 LRPPI3 -4.0495 4.6667 13.3828 FALSE
GT3 LRPPI4 -4.0495 4.6667 13.3828 FALSE
GT3 ME1 -3.7162 5.0000 13.7162 FALSE
GT3 ME2 -2.3828 6.3333 15.0495 FALSE
GT3 ME3 -4.7162 4.0000 12.7162 FALSE
LRPPI1 LRPPI2 -8.0495 0.6667 9.3828 FALSE
LRPPI1 LRPPI3 -5.3828 3.3333 12.0495 FALSE
LRPPI1 LRPPI4 -5.3828 3.3333 12.0495 FALSE
LRPPI1 ME1 -5.0495 3.6667 12.3828 FALSE
LRPPI1 ME2 -3.7162 5.0000 13.7162 FALSE
LRPPI1 ME3 -6.0495 2.6667 11.3828 FALSE
LRPPI2 LRPPI3 -6.0495 2.6667 11.3828 FALSE
LRPPI2 LRPPI4 -6.0495 2.6667 11.3828 FALSE
LRPPI2 ME1 -5.7162 3.0000 11.7162 FALSE
LRPPI2 ME2 -4.3828 4.3333 13.0495 FALSE
LRPPI2 ME3 -6.7162 2.0000 10.7162 FALSE
LRPPI3 LRPPI4 -8.7162 0.0000 8.7162 FALSE
LRPPI3 ME1 -8.3828 0.3333 9.0495 FALSE
LRPPI3 ME2 -7.0495 1.6667 10.3828 FALSE
LRPPI3 ME3 -9.3828 -0.6667 8.0495 FALSE
LRPPI4 ME1 -8.3828 0.3333 9.0495 FALSE
LRPPI4 ME2 -7.0495 1.6667 10.3828 FALSE
LRPPI4 ME3 -9.3828 -0.6667 8.0495 FALSE
ME1 ME2 -7.3828 1.3333 10.0495 FALSE
ME1 ME3 -9.7162 -1.0000 7.7162 FALSE
ME2 ME3 -11.0495 -2.3333 6.3828 FALSE

download these results as csv

2008 composer.perfoldaccuracy.friedman.tukeykramerhsd.png