2013:Audio Music Similarity and Retrieval Results

From MIREX Wiki
Revision as of 01:32, 26 October 2013 by Yun Hao (talk | contribs)

Introduction

These are the results for the 2013 running of the Audio Music Similarity and Retrieval task set. For background information about this task set please refer to the Audio Music Similarity and Retrieval page.

Each system was given 7000 songs chosen from IMIRSEL's "uspop", "uscrap" and "american" "classical" and "sundry" collections. Each system then returned a 7000x7000 distance matrix. 50 songs were randomly selected from the 10 genre groups (5 per genre) as queries and the first 5 most highly ranked songs out of the 7000 were extracted for each query (after filtering out the query itself, returned results from the same artist were also omitted). Then, for each query, the returned results (candidates) from all participants were grouped and were evaluated by human graders using the Evalutron 6000 grading system. Each individual query/candidate set was evaluated by a single grader. For each query/candidate pair, graders provided two scores. Graders were asked to provide 1 categorical BROAD score with 3 categories: NS,SS,VS as explained below, and one FINE score (in the range from 0 to 100). A description and analysis is provided below.

The systems read in 30 second audio clips as their raw data. The same 30 second clips were used in the grading stage.


General Legend

Team ID

Sub code Submission name Abstract Contributors
DM1 DM1 PDF Franz de Leon, Kirk Martinez
DM2 DM2 PDF Franz de Leon, Kirk Martinez
GKC1 ILSP Audio Similarity PDF [http:// Aggelos Gkiokas], [http:// Vassilis Katsouros], [http:// George Carayannis]
GKC2 ILSP Audio Similarity 2 PDF [http:// Aggelos Gkiokas], [http:// Vassilis Katsouros], [http:// George Carayannis]
PS1 PS09 PDF Dominik Schnitzer, Tim Pohle
RA1 SimRet2013 PDF [http:// Roman Aliyev]
SS2 cbmr_sim_2013_submission PDF Klaus Seyerlehner, Markus Schedl
SSPK1 cbmr_sim_2010_resubmission PDF Klaus Seyerlehner, Markus Schedl, Tim Pohle, Peter Knees

Broad Categories

NS = Not Similar
SS = Somewhat Similar
VS = Very Similar

Understanding Summary Measures

Fine = Has a range from 0 (failure) to 100 (perfection).
Broad = Has a range from 0 (failure) to 2 (perfection) as each query/candidate pair is scored with either NS=0, SS=1 or VS=2.

Human Evaluation

Overall Summary Results

Measure DM1 DM2 GKC1 GKC2 PS1 RA1 SS2 SSPK
Average Fine Score 46.262 48.074 43.804 48.046 53.808 18.858 55.210 54.100
Average Cat Score 0.926 0.992 0.896 0.978 1.142 0.276 1.184 1.164

download these results as csv

Friedman's Tests

Friedman's Test (FINE Scores)

The Friedman test was run in MATLAB against the Fine summary data over the 50 queries.
Command: [c,m,h,gnames] = multcompare(stats, 'ctype', 'tukey-kramer','estimate', 'friedman', 'alpha', 0.05);

TeamID TeamID Lowerbound Mean Upperbound Significance
SS2 SSPK1 -1.103 0.380 1.863 FALSE
SS2 PS1 -1.063 0.420 1.903 FALSE
SS2 DM2 0.027 1.510 2.993 TRUE
SS2 GKC2 0.027 1.510 2.993 TRUE
SS2 DM1 0.447 1.930 3.413 TRUE
SS2 GKC1 1.187 2.670 4.153 TRUE
SS2 RA1 3.457 4.940 6.423 TRUE
SSPK1 PS1 -1.443 0.040 1.523 FALSE
SSPK1 DM2 -0.353 1.130 2.613 FALSE
SSPK1 GKC2 -0.353 1.130 2.613 FALSE
SSPK1 DM1 0.067 1.550 3.033 TRUE
SSPK1 GKC1 0.807 2.290 3.773 TRUE
SSPK1 RA1 3.077 4.560 6.043 TRUE
PS1 DM2 -0.393 1.090 2.573 FALSE
PS1 GKC2 -0.393 1.090 2.573 FALSE
PS1 DM1 0.027 1.510 2.993 TRUE
PS1 GKC1 0.767 2.250 3.733 TRUE
PS1 RA1 3.037 4.520 6.003 TRUE
DM2 GKC2 -1.483 0.000 1.483 FALSE
DM2 DM1 -1.063 0.420 1.903 FALSE
DM2 GKC1 -0.323 1.160 2.643 FALSE
DM2 RA1 1.947 3.430 4.913 TRUE
GKC2 DM1 -1.063 0.420 1.903 FALSE
GKC2 GKC1 -0.323 1.160 2.643 FALSE
GKC2 RA1 1.947 3.430 4.913 TRUE
DM1 GKC1 -0.743 0.740 2.223 FALSE
DM1 RA1 1.527 3.010 4.493 TRUE
GKC1 RA1 0.787 2.270 3.753 TRUE

download these results as csv

2013 evalutron.fine.friedman.tukeyKramerHSD.png

Friedman's Test (BROAD Scores)

The Friedman test was run in MATLAB against the BROAD summary data over the 50 queries.
Command: [c,m,h,gnames] = multcompare(stats, 'ctype', 'tukey-kramer','estimate', 'friedman', 'alpha', 0.05);

TeamID TeamID Lowerbound Mean Upperbound Significance
SS2 SSPK1 -1.037 0.400 1.837 FALSE
SS2 PS1 -1.067 0.370 1.807 FALSE
SS2 DM2 -0.017 1.420 2.857 FALSE
SS2 GKC2 0.193 1.630 3.067 TRUE
SS2 DM1 0.573 2.010 3.447 TRUE
SS2 GKC1 0.993 2.430 3.867 TRUE
SS2 RA1 3.263 4.700 6.137 TRUE
SSPK1 PS1 -1.467 -0.030 1.407 FALSE
SSPK1 DM2 -0.417 1.020 2.457 FALSE
SSPK1 GKC2 -0.207 1.230 2.667 FALSE
SSPK1 DM1 0.173 1.610 3.047 TRUE
SSPK1 GKC1 0.593 2.030 3.467 TRUE
SSPK1 RA1 2.863 4.300 5.737 TRUE
PS1 DM2 -0.387 1.050 2.487 FALSE
PS1 GKC2 -0.177 1.260 2.697 FALSE
PS1 DM1 0.203 1.640 3.077 TRUE
PS1 GKC1 0.623 2.060 3.497 TRUE
PS1 RA1 2.893 4.330 5.767 TRUE
DM2 GKC2 -1.227 0.210 1.647 FALSE
DM2 DM1 -0.847 0.590 2.027 FALSE
DM2 GKC1 -0.427 1.010 2.447 FALSE
DM2 RA1 1.843 3.280 4.717 TRUE
GKC2 DM1 -1.057 0.380 1.817 FALSE
GKC2 GKC1 -0.637 0.800 2.237 FALSE
GKC2 RA1 1.633 3.070 4.507 TRUE
DM1 GKC1 -1.017 0.420 1.857 FALSE
DM1 RA1 1.253 2.690 4.127 TRUE
GKC1 RA1 0.833 2.270 3.707 TRUE

download these results as csv

2013 evalutron.cat.friedman.tukeyKramerHSD.png

Summary Results by Query

FINE Scores

These are the mean FINE scores per query assigned by Evalutron graders. The FINE scores for the 5 candidates returned per algorithm, per query, have been averaged. Values are bounded between 0 and 100. A perfect score would be 100. Genre labels have been included for reference.

Genre Query DM1 DM2 GKC1 GKC2 PS1 RA1 SS2 SSPK1
BAROQUE d001537 88.8 85.3 76.7 84.6 82.7 14.7 86.8 87.6
BAROQUE d004754 27.3 33.7 24.3 25.7 55.7 0.4 35.6 55.8
BAROQUE d005709 67.0 67.0 61.2 57.0 63.3 4.9 60.0 50.0
BAROQUE d010905 64.3 40.3 61.8 59.3 70.3 9.7 76.0 76.5
BAROQUE d012021 38.7 14.7 20.5 27.0 40.0 0.2 54.5 58.0
BLUES e000029 33.0 35.6 27.0 23.4 36.1 17.7 23.4 29.7
BLUES e001368 75.1 74.4 53.7 43.0 79.1 35.1 73.4 72.6
BLUES e005937 43.1 51.9 37.0 53.0 67.3 55.6 65.3 66.5
BLUES e015582 62.7 49.8 64.1 55.7 38.4 41.8 66.8 59.7
BLUES e018379 63.8 61.6 59.8 63.3 63.5 36.3 58.6 59.4
CLASSICAL d004474 58.5 59.2 55.0 53.9 55.6 8.3 57.7 58.2
CLASSICAL d007653 54.1 58.0 41.0 38.7 51.1 13.7 54.8 48.0
CLASSICAL d007675 62.3 81.0 73.5 72.2 75.6 21.4 80.0 72.4
CLASSICAL d019132 3.5 2.8 3.4 4.9 3 0 7.5 4.5
CLASSICAL d019948 65.5 58.0 53.5 60.0 64.5 7.5 61.5 58.5
COUNTRY e003848 36.8 36.3 29.2 29.9 39.9 15.2 41.4 40.0
COUNTRY e007409 59.0 65.0 67.5 67.0 76.0 39.0 76.5 81.0
COUNTRY e010612 40.7 37.5 35.1 52.9 55.1 14.1 66.0 62.7
COUNTRY e012827 67.9 71.6 68.4 69.4 80.0 29.9 87.5 76.7
COUNTRY e019126 29.2 44.1 33.4 41.6 31.4 22.5 45.6 39.0
EDANCE a000504 86.5 88.1 68.5 80.7 84.5 24.7 87.5 83.9
EDANCE a003842 19.5 28.5 56.8 42.5 56.6 22.7 60.8 53.8
EDANCE a004027 43.8 39.0 25.1 20.7 41.0 21.6 28.4 33.4
EDANCE b002993 57.0 49.0 51.0 57.0 37.0 29.0 68.0 68.0
EDANCE b003885 33.9 58.1 66.6 67.6 56.0 25.4 67.5 67.6
JAZZ e001100 43.3 31.9 59.6 69.1 65.6 12.1 63.6 60.3
JAZZ e002233 56.8 59.0 54.0 59.2 51.9 31.8 52.2 63.3
JAZZ e003036 83.5 79.8 86.1 88.5 82.2 37.7 88.5 87.9
JAZZ e007159 60.2 67.3 41.3 65.5 73.6 16.8 72.2 46.1
JAZZ e010868 61.0 73.0 65.4 72.4 69.7 41.3 71.3 85.5
METAL b005514 9.5 22.6 15.5 27.7 24.4 13.5 28.9 35.9
METAL b005832 63.6 64.5 36.5 50.0 62.8 38.1 73.9 73.3
METAL b008728 13.9 18.3 15.1 26.4 24.1 12.1 24.7 16.6
METAL b015780 39.3 66.0 44.7 46.1 70.3 22.7 50.0 58.5
METAL f004405 17.4 21.6 11.0 22.4 37.7 12.6 15.9 21.2
RAPHIPHOP a005976 52.5 58.0 50.3 56.6 58.4 23.2 62.4 59.2
RAPHIPHOP a007753 37.3 38.1 32.6 50.7 48.3 10.9 48.9 49.1
RAPHIPHOP a008296 31.1 27.2 40.4 40.4 36.1 14.3 38.7 37.1
RAPHIPHOP b007216 12.7 22.1 36.9 44.5 46.0 10.9 40.9 38.5
RAPHIPHOP b009991 62.9 65.2 67.2 69.5 75.6 20.4 62.4 66.3
ROCKROLL b002418 29.5 29.0 34.5 38.5 35.0 14.8 42.0 33.5
ROCKROLL b012404 36.8 52.1 47.1 45.7 46.5 36.3 58.7 61.7
ROCKROLL b012651 17.5 18.0 10.0 21.5 30.5 12.5 41.5 33.5
ROCKROLL b018558 35.5 36.0 11.5 25.0 42.0 11.5 49.0 48.5
ROCKROLL b019150 6 11.2 12.0 11.2 13.6 3.5 12.2 13.6
ROMANTIC d004234 27.7 20.4 17.2 24.6 28.3 11.7 13.8 27.0
ROMANTIC d007659 65.0 62.5 22.5 40.9 72.8 1.5 74.1 67.6
ROMANTIC d011050 54.5 58.8 44.2 44.9 52.5 7.1 53.2 47.3
ROMANTIC d011968 76.1 67.9 74.0 59.3 78.0 6.2 73.7 73.8
ROMANTIC d013072 37.5 42.7 46.5 50.7 60.8 8 56.7 40.2

download these results as csv

BROAD Scores

These are the mean BROAD scores per query assigned by Evalutron graders. The BROAD scores for the 5 candidates returned per algorithm, per query, have been averaged. Values are bounded between 0 (not similar) and 2 (very similar). A perfect score would be 2. Genre labels have been included for reference.

Genre Query DM1 DM2 GKC1 GKC2 PS1 RA1 SS2 SSPK1
BAROQUE d001537 2 2 1.6 1.9 1.9 0.1 2 2
BAROQUE d004754 0.7 0.8 0.7 0.8 1.2 0 0.7 1.2
BAROQUE d005709 1.5 1.5 1.3 1.1 1.5 0 1.4 1
BAROQUE d010905 1.7 0.8 1.5 1.4 1.8 0.1 1.9 2
BAROQUE d012021 0.9 0.3 0.5 0.4 0.7 0 1.1 1.4
BLUES e000029 1 1.1 0.6 0.8 1 0.4 0.8 1
BLUES e001368 1.5 1.5 0.7 0.4 1.8 0.4 1.5 1.6
BLUES e005937 0.8 1.1 0.6 1.2 1.6 1.2 1.5 1.5
BLUES e015582 1.2 0.9 1.3 1 0.6 0.8 1.4 1.2
BLUES e018379 0.9 0.9 1.1 1.2 1.2 0.6 1 1
CLASSICAL d004474 1.3 1.3 1.2 1.2 1.2 0 1.3 1.1
CLASSICAL d007653 1.2 1.3 0.7 0.6 0.9 0 1.2 1
CLASSICAL d007675 1.4 1.9 1.6 1.7 1.7 0.2 1.9 1.7
CLASSICAL d019132 0 0 0 0 0 0 0.1 0
CLASSICAL d019948 1.1 0.9 1 1 1 0 1 1
COUNTRY e003848 0.6 0.6 0.3 0.3 0.6 0 0.7 0.6
COUNTRY e007409 1 1.2 1.3 1.2 1.6 0.6 1.6 1.8
COUNTRY e010612 0.8 0.8 0.7 1.1 1.2 0.1 1.3 1.2
COUNTRY e012827 1.2 1.4 1.4 1.3 1.7 0.5 2 1.7
COUNTRY e019126 0.5 0.8 0.6 0.8 0.7 0.3 1 0.7
EDANCE a000504 2 2 1.5 1.8 1.7 0.4 1.9 1.8
EDANCE a003842 0.1 0.5 1.4 0.9 1.4 0.4 1.6 1.4
EDANCE a004027 0.9 0.8 0.4 0.3 0.9 0.4 0.7 0.8
EDANCE b002993 1 1 0.9 1 0.5 0.3 1.3 1.3
EDANCE b003885 0.5 1.2 1.8 1.8 1.2 0.3 1.6 1.6
JAZZ e001100 0.5 0.3 1.2 1.3 1.2 0 1.2 1.1
JAZZ e002233 1.1 1.2 1.2 1.3 1.1 0.6 1.1 1.3
JAZZ e003036 1.7 1.6 1.9 1.8 1.6 0.3 1.9 2
JAZZ e007159 1.3 1.5 0.7 1.3 1.6 0.3 1.5 1
JAZZ e010868 1.1 1.3 1.2 1.3 1.3 0.7 1.3 1.9
METAL b005514 0.3 0.8 0.6 0.8 0.9 0.6 0.8 1.1
METAL b005832 1.3 1.3 0.5 0.7 1.2 0.6 1.5 1.6
METAL b008728 0.3 0.5 0.5 0.6 0.7 0.2 0.7 0.5
METAL b015780 0.9 1.7 1 1 1.7 0.4 1.1 1.3
METAL f004405 0.7 0.7 0.3 0.7 1.2 0.5 0.5 0.6
RAPHIPHOP a005976 0.9 0.9 0.7 0.9 1 0.3 1 1
RAPHIPHOP a007753 0.8 0.7 0.5 1 0.9 0 1 1.1
RAPHIPHOP a008296 0.4 0.3 0.7 0.8 0.7 0.2 0.6 0.6
RAPHIPHOP b007216 0.2 0.5 0.8 1 1.1 0.1 1 0.8
RAPHIPHOP b009991 1.3 1.3 1.6 1.6 1.7 0 1.2 1.4
ROCKROLL b002418 0.4 0.4 0.6 0.7 0.5 0.1 0.6 0.4
ROCKROLL b012404 0.7 1.3 0.9 0.7 0.9 0.5 1.5 1.4
ROCKROLL b012651 0.2 0.3 0.2 0.4 0.5 0.3 0.7 0.5
ROCKROLL b018558 0.8 0.8 0.2 0.7 1.1 0.4 1.3 1.3
ROCKROLL b019150 0.2 0.5 0.5 0.4 0.6 0.4 0.5 0.5
ROMANTIC d004234 0.2 0.2 0 0 0 0 0.1 0.2
ROMANTIC d007659 1.2 1.1 0.5 1 1.6 0 1.4 1.3
ROMANTIC d011050 1.1 1.2 1 1 1.1 0.1 1.1 1
ROMANTIC d011968 2 1.7 1.7 1.5 1.9 0 1.8 1.9
ROMANTIC d013072 0.9 0.9 1.1 1.2 1.4 0.1 1.3 0.8

download these results as csv

Raw Scores

The raw data derived from the Evalutron 6000 human evaluations are located on the 2013:Audio Music Similarity and Retrieval Raw Data page.

Metadata and Distance Space Evaluation

The following reports provide evaluation statistics based on analysis of the distance space and metadata matches and include:

  • Neighbourhood clustering by artist, album and genre
  • Artist-filtered genre clustering
  • How often the triangular inequality holds
  • Statistics on 'hubs' (tracks similar to many tracks) and orphans (tracks that are not similar to any other tracks at N results).

Reports

DM1 = Franz de Leon, Kirk Martinez
DM2 = Franz de Leon, Kirk Martinez
GKC1 = Aggelos Gkiokas, Vassilis Katsouros, George Carayannis
GKC2 = Aggelos Gkiokas, Vassilis Katsouros, George Carayannis
PS1 = Dominik Schnitzer, Tim Pohle
RA1 = Roman Aliyev
SS2 = Klaus Seyerlehner, Markus Schedl
SSPK1 = Klaus Seyerlehner, Markus Schedl, Tim Pohle, Peter Knees