Difference between revisions of "2025:Audio Chord Estimation Results"

From MIREX Wiki
Line 38: Line 38:
 
|}
 
|}
  
−
= Main Evaluation Results =
+
= Test Sets =
 +
 
 +
Main Test Sets
 +
* Billboard 2013: The held-out portion of the McGill Billboard dataset, containing mainly western pop songs from the Billboard chart.
 +
* Yamaha_JPOP: A private dataset annotated by Yamaha Corporation. The dataset contains 200 JPOP songs.
 +
* Yamaha_Balanced: A private dataset annotated by Yamaha Corporation. The dataset contains 241 songs. While still biased towards JPOP songs, the dataset contains a wider genre distribution: J.Pop (10.37%), Rock (10.37%), J.Enka (10.37%), J.Kayoukyoku (10.37%), Soundtrack (10.37%), Western Pop (10.37%), Children's Song (10.37%), R&B (6.22%), Hiphop (4.56%), Jazz (2.49%), Dance (2.49%), World (2.07%), Techno (1.24%), Easy listening (1.24%), J.Minyou (1.24%), Others (5.81%).
 +
 
 +
Additional Test Sets
 +
* Billboard 2012: The public portion of the McGill Billboard dataset.
 +
* RWC Popular: 100 pop songs from the RWC (Real World Computing) Music Database. 20% songs with English lyrics and 80% songs with Japanese lyrics.
 +
 
 +
= Main Results =
  
 
The following datasets are served as pure test sets. No system is allowed to train on them.
 
The following datasets are served as pure test sets. No system is allowed to train on them.
Line 267: Line 278:
 
|}
 
|}
  
−
= Additional Test Sets =
+
= Additional Results =
  
−
Below are datasets that are not pure test sets. Some models might have trained on them; please see their extended abstracts for details.
+
Below are results on datasets that may not be strictly held-out test sets. Some models might have been trained on these datasets; for specific details, please refer to the extended abstracts of each model.
  
 
== Billboard2012 ==
 
== Billboard2012 ==

Revision as of 23:53, 7 September 2025

This page is still WIP. More submissions and descriptions may appear.

Submissions

Submission Title PDF
Baseline: Chordino NNLS Chroma v1.1 Link
Baseline: ISMIR2019 Large-Vocabulary Chord Transcription via Chord Structure Decomposition Link
MD1 Degree-Based Automatic Chord Recognition with Enharmonic Distinction TBA
wu-ensemble wu-ensemble TBA
wu-single wu-single TBA
YK1 Semi-Supervised Audio Chord Estimator Based on Disentangled Generative Modeling TBA
BMACE A Mamba-Based Model for Automatic Chord Recognition TBA

Test Sets

Main Test Sets

  • Billboard 2013: The held-out portion of the McGill Billboard dataset, containing mainly western pop songs from the Billboard chart.
  • Yamaha_JPOP: A private dataset annotated by Yamaha Corporation. The dataset contains 200 JPOP songs.
  • Yamaha_Balanced: A private dataset annotated by Yamaha Corporation. The dataset contains 241 songs. While still biased towards JPOP songs, the dataset contains a wider genre distribution: J.Pop (10.37%), Rock (10.37%), J.Enka (10.37%), J.Kayoukyoku (10.37%), Soundtrack (10.37%), Western Pop (10.37%), Children's Song (10.37%), R&B (6.22%), Hiphop (4.56%), Jazz (2.49%), Dance (2.49%), World (2.07%), Techno (1.24%), Easy listening (1.24%), J.Minyou (1.24%), Others (5.81%).

Additional Test Sets

  • Billboard 2012: The public portion of the McGill Billboard dataset.
  • RWC Popular: 100 pop songs from the RWC (Real World Computing) Music Database. 20% songs with English lyrics and 80% songs with Japanese lyrics.

Main Results

The following datasets are served as pure test sets. No system is allowed to train on them.

Billboard2013

Group MirexRoot MirexMajMin MirexMajMinBass MirexSevenths MirexSeventhsBass MeanSeg UnderSeg OverSeg
Baseline: Chordino 71.06 67.18 65.09 48.88 47.06 0.82 0.83 0.83
Baseline: ISMIR2019 78.61 76.39 74.72 64.15 62.65 0.83 0.79 0.93
MD1 81.35 79.15 77.91 66.40 65.33 0.86 0.85 0.89
wu-ensemble 74.64 71.97 70.72 55.06 53.96 0.83 0.86 0.82
wu-single 75.77 73.14 71.74 55.41 54.15 0.83 0.85 0.83
YK1 81.01 78.10 75.41 64.53 62.05 0.86 0.85 0.87

YAMAHA_Balanced

Group MirexRoot MirexMajMin MirexMajMinBass MirexSevenths MirexSeventhsBass MeanSeg UnderSeg OverSeg
Baseline: Chordino 77.57 74.64 71.59 56.38 53.90 0.87 0.87 0.87
Baseline: ISMIR2019 82.00 81.16 79.69 66.97 65.77 0.89 0.86 0.93
MD1 81.83 80.22 78.87 64.13 63.14 0.88 0.89 0.88
wu-ensemble 82.54 81.29 78.99 62.84 60.84 0.87 0.89 0.87
wu-single 81.37 79.69 77.61 61.60 59.84 0.87 0.90 0.86
YK1 82.53 79.71 75.60 66.02 62.31 0.89 0.90 0.89

YAMAHA_JPop

Group MirexRoot MirexMajMin MirexMajMinBass MirexSevenths MirexSeventhsBass MeanSeg UnderSeg OverSeg
Baseline: Chordino 74.49 71.99 69.24 52.40 49.97 0.87 0.86 0.88
Baseline: ISMIR2019 81.49 79.99 78.58 62.81 61.61 0.90 0.87 0.94
MD1 79.34 77.10 76.07 55.59 54.71 0.88 0.88 0.88
wu-ensemble 79.58 77.58 75.57 54.36 52.58 0.87 0.88 0.87
wu-single 78.87 76.56 74.66 55.35 53.60 0.87 0.89 0.86
YK1 80.13 77.03 72.85 61.24 57.26 0.89 0.90 0.89

Additional Results

Below are results on datasets that may not be strictly held-out test sets. Some models might have been trained on these datasets; for specific details, please refer to the extended abstracts of each model.

Billboard2012

Group MirexRoot MirexMajMin MirexMajMinBass MirexSevenths MirexSeventhsBass MeanSeg UnderSeg OverSeg
Baseline: Chordino 74.04 72.11 70.05 55.24 53.28 0.84 0.85 0.83
MD1 85.11 83.98 82.76 74.12 73.12 0.89 0.89 0.90
wu-ensemble 78.26 77.15 75.58 59.99 58.79 0.84 0.88 0.83
wu-single 79.23 78.21 76.76 60.23 59.07 0.85 0.87 0.84
YK1 85.90 84.66 81.81 77.22 74.45 0.88 0.88 0.90


RWC-Popular

Group MirexRoot MirexMajMin MirexMajMinBass MirexSevenths MirexSeventhsBass MeanSeg UnderSeg OverSeg
Baseline: Chordino 78.97 77.78 74.13 63.15 59.72 0.89 0.88 0.90
MD1 83.98 81.18 79.42 66.53 64.83 0.89 0.89 0.89
wu-ensemble 81.87 80.30 77.58 62.65 60.25 0.88 0.90 0.86
wu-single 82.48 81.35 78.48 62.86 60.28 0.88 0.89 0.87
YK1 88.76 87.27 81.14 76.88 70.90 0.92 0.92 0.92
Retrieved from "https://music-ir.org/mirex/w/index.php?title=2025:Audio_Chord_Estimation_Results&oldid=14752"