2017:Audio Chord Estimation Results

From MIREX Wiki
Revision as of 05:18, 14 October 2020 by JohanPauwels (talk | contribs) (Remove outdated reference)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)

Introduction

This page contains the results of the 2017 edition of the MIREX automatic chord estimation tasks. This edition was the fifth one since the reorganization of the evaluation procedure in 2013. The results can therefore be directly compared to those of the last four years. Chord labels are evaluated according to five different chord vocabularies and the segmentation is also assessed. Additional information about the used measures can be found on the page of the 2013 edition.

What’s new?

  • This year the algorithms have additionally been evaluated on the "RWC-Popular" and "USPOP2002Chords" dataset annotated at the Music and Audio Research Lab of NYU, whose annotations are publicly available. The RWC-Popular dataset contains 100 pop songs recorded specifically for music information retrieval research. The USPOP2002Chords set is the 195 file subset of the USPOP2002 dataset that have been annotated with chord sequences.

Software

All software used for the evaluation has been made open-source. The evaluation framework is described by Pauwels and Peeters (2013). The corresponding binaries and code repository can be found on GitHub and the used measures are available as presets. The raw algorithmic output provided below makes it possible to calculate the additional measures from the paper (separate results for tetrads, etc.), in addition to those presented below. More help can be found in the readme.

The statistical comparison between the different submissions is explained in Burgoyne et al. (2014). The software is available at BitBucket. It uses the detailed results provided below as input.

Submissions

Abstract Contributors
CM2 PDF Chris Cannam, Matthias Mauch
JLW1, JLW2 PDF Junyan Jiang, Wei Li, Yiming Wu
KBK1, KBK2 PDF Filip Korzeniowski, Sebastian Böck, Florian Krebs
WL1 PDF Yiming Wu, Wei Li

Results

Summary

All figures can be interpreted as percentages and range from 0 (worst) to 100 (best).

Isophonics2009
Algorithm MirexRoot MirexMajMin MirexMajMinBass MirexSevenths MirexSeventhsBass MeanSeg UnderSeg OverSeg
CM2 78.66 75.51 72.58 54.78 52.36 85.87 87.22 85.98
JLW1 81.45 78.97 77.18 67.64 66.11 86.35 83.73 91.08
JLW2 80.93 78.49 76.71 67.22 65.70 85.59 82.22 91.33
KBK1 83.62 82.24 79.27 72.06 69.35 87.11 83.82 92.48
KBK2 87.38 86.80 83.43 75.56 72.60 89.29 87.24 92.35
WL1 81.38 79.70 74.33 69.01 64.11 84.03 79.19 91.90

download these results as csv

Billboard2012
Algorithm MirexRoot MirexMajMin MirexMajMinBass MirexSevenths MirexSeventhsBass MeanSeg UnderSeg OverSeg
CM2 74.23 72.31 70.25 55.44 53.48 83.60 85.33 83.31
JLW1 78.00 76.42 75.08 62.06 60.93 84.23 82.33 87.86
JLW2 78.66 77.14 75.79 62.49 61.32 84.52 81.88 89.22
KBK1 79.81 79.19 76.71 56.90 54.84 85.78 82.61 90.72
KBK2 86.30 86.02 83.12 61.12 58.75 88.38 86.78 90.97
WL1 79.49 78.54 74.67 62.54 59.57 83.61 79.64 89.80

download these results as csv

Billboard2013
Algorithm MirexRoot MirexMajMin MirexMajMinBass MirexSevenths MirexSeventhsBass MeanSeg UnderSeg OverSeg
CM2 71.23 67.36 65.28 49.07 47.25 81.50 83.13 82.53
JLW1 75.00 72.30 70.93 57.11 55.87 81.28 79.15 87.35
JLW2 75.40 72.77 71.41 57.49 56.25 81.58 78.57 89.02
KBK1 75.17 72.31 69.88 52.27 50.13 81.25 77.78 89.40
KBK2 80.60 78.37 75.85 55.81 53.57 83.59 83.01 87.54
WL1 75.22 72.53 69.06 57.87 55.14 79.69 75.54 88.89

download these results as csv

JayChou29
Algorithm MirexRoot MirexMajMin MirexMajMinBass MirexSevenths MirexSeventhsBass MeanSeg UnderSeg OverSeg
CM2 72.82 72.15 65.55 54.46 49.05 86.58 86.94 86.84
JLW1 76.26 75.85 72.47 58.10 55.42 89.71 88.79 90.95
JLW2 76.06 75.70 72.36 57.67 55.08 89.72 88.23 91.56
KBK1 78.11 77.41 66.03 50.86 41.66 86.93 83.46 91.47
KBK2 81.42 80.48 69.66 51.77 43.18 89.08 87.67 90.87
WL1 83.04 82.34 78.57 62.02 58.60 89.01 86.38 91.93

download these results as csv

RobbieWilliams
Algorithm MirexRoot MirexMajMin MirexMajMinBass MirexSevenths MirexSeventhsBass MeanSeg UnderSeg OverSeg
CM2 81.94 78.29 76.09 57.97 55.94 87.95 89.00 87.39
JLW1 83.72 80.48 78.81 70.08 68.40 88.80 87.26 91.12
JLW2 84.00 80.68 79.04 70.07 68.41 89.01 86.97 91.91
KBK1 84.36 81.98 79.38 77.98 75.66 88.33 85.61 92.14
KBK2 89.38 88.12 85.06 83.38 80.68 90.96 89.52 92.84
WL1 83.34 80.65 77.55 71.23 68.38 86.38 82.19 92.14

download these results as csv

RWC-Popular
Algorithm MirexRoot MirexMajMin MirexMajMinBass MirexSevenths MirexSeventhsBass MeanSeg UnderSeg OverSeg
CM2 79.12 77.95 74.30 63.33 59.89 88.62 88.11 89.67
JLW1 81.84 81.40 79.12 69.76 67.50 89.20 86.65 92.75
JLW2 82.22 81.76 79.44 70.06 67.79 89.56 86.54 93.72
KBK1 81.32 80.48 74.79 55.46 50.10 87.85 84.67 91.84
KBK2 87.20 86.58 80.92 58.04 52.78 89.94 87.94 92.37
WL1 84.05 82.87 79.08 70.12 66.91 87.48 83.54 92.51

download these results as csv

USPOP2002Chords
Algorithm MirexRoot MirexMajMin MirexMajMinBass MirexSevenths MirexSeventhsBass MeanSeg UnderSeg OverSeg
CM2 78.41 76.24 72.74 59.84 56.67 85.81 86.81 86.02
JLW1 80.85 79.33 77.34 67.30 65.54 86.53 84.50 89.97
JLW2 81.47 80.06 78.12 68.00 66.29 87.11 84.48 91.32
KBK1 79.56 77.19 72.47 62.52 58.30 85.49 82.25 90.52
KBK2 82.34 80.72 76.04 64.50 60.30 86.59 85.49 89.17
WL1 81.97 81.20 77.30 68.91 65.48 85.49 81.41 91.70

download these results as csv


Detailed Results

More details about the performance of the algorithms, including per-song performance and supplementary statistics, are available from this repository.

Algorithmic Output

The raw output of the algorithms are available in this repository. They can be used to experiment with alternative evaluation measures and statistics.