Freesound Loop Dataset
The Loop Annotator is a Flask web application that allows to annotate loops from Freesound. With it, we created a dataset of Creative Commons loops, with professional quality and perfect for research!
This annotator is hosted online and accessible here.
The full dataset can be found in Zenodo.
The full paper, poster and presentation are available at ISMIR website and, in this page, we present some of the results that we were not able to include in the paper.
If you use this work for research, please cite:
@inproceedings{ramires2020, author = "Antonio Ramires and Frederic Font and Dmitry Bogdanov and Jordan B. L. Smith and Yi-Hsuan Yang and Joann Ching and Bo-Yu Chen and Yueh-Kao Wu and Hsu Wei-Han and Xavier Serra", title = "The Freesound Loop Dataset and Annotation Tool", booktitle = "Proc. of the 21st International Society for Music Information Retrieval (ISMIR)", year = "2020" }
Benchmarking Results
In this page, we present the full results of the benchmarking for the Freesound Loop Dataset.
Tempo Estimation
For tempo estimation, we used the following metrics: Accuracy 1 is the percentage of instances whose estimated BPM is within 4% of the annotated ground truth. Accuracy 2 is the percentage ofinstances whose estimated BPM is within a 4% of 1/3,1/2, 1, 2, or 3 times the ground truth BPM. Accuracy 1e repre-sents the percentage of instances whose estimated BPM is exactly the same as the ground truth after rounding the estimated BPM to the nearest integer.
Various Annotations - Either
| Algorithm | Accuracy1 | Accuracy1e | Accuracy2 | Mean Accuracy |
|---|---|---|---|---|
| Percival14_essentia | 84.39 | 73.81 | 94.18 | 84.13 |
| Percival14 | 85.24 | 73.39 | 96.31 | 84.98 |
| Zapata14 | 83.32 | 60.75 | 89.85 | 77.97 |
| Degara12 | 83.53 | 63.38 | 90.56 | 79.16 |
| Bock15 | 70.05 | 51.88 | 92.12 | 71.35 |
| Bock15ACF | 77.29 | 54.58 | 92.83 | 74.90 |
| Bock15DBN | 67.00 | 51.03 | 94.68 | 70.90 |
Various Annotations - Same
| Algorithm | Accuracy1 | Accuracy1e | Accuracy2 | Mean Accuracy |
|---|---|---|---|---|
| Percival14_essentia | 63.55 | 53.18 | 80.52 | 65.75 |
| Percival14 | 62.92 | 52.32 | 81.23 | 65.49 |
| Zapata14 | 62.84 | 41.01 | 73.13 | 58.99 |
| Degara12 | 62.53 | 42.66 | 72.74 | 59.31 |
| Bock15 | 48.08 | 33.54 | 77.14 | 52.92 |
| Bock15ACF | 53.65 | 35.43 | 75.18 | 54.75 |
| Bock15DBN | 44.78 | 32.99 | 79.18 | 52.32 |
Single Annotations
| Algorithm | Accuracy1 | Accuracy1e | Accuracy2 | Mean Accuracy |
|---|---|---|---|---|
| Percival14_essentia | 60.35 | 50.24 | 78.34 | 62.98 |
| Percival14 | 61.10 | 51.80 | 79.09 | 64.00 |
| Zapata14 | 62.39 | 40.46 | 73.52 | 58.79 |
| Degara12 | 62.25 | 41.00 | 73.05 | 58.77 |
| Bock15 | 50.64 | 27.56 | 75.29 | 51.17 |
| Bock15ACF | 52.68 | 31.77 | 71.42 | 51.96 |
| Bock15DBN | 44.81 | 29.60 | 77.39 | 50.60 |
Automatic Annotations
| Algorithm | Accuracy1 | Accuracy1e | Accuracy2 | Mean Accuracy |
|---|---|---|---|---|
| Percival14_essentia | 56.73 | 46.56 | 70.98 | 58.09 |
| Percival14 | 56.01 | 45.80 | 71.65 | 57.82 |
| Zapata14 | 56.27 | 33.59 | 65.57 | 51.81 |
| Degara12 | 56.43 | 35.07 | 65.47 | 52.32 |
| Bock15 | 40.22 | 25.62 | 67.41 | 44.42 |
| Bock15ACF | 48.74 | 29.98 | 67.24 | 48.65 |
| Bock15DBN | 38.73 | 26.45 | 72.10 | 45.76 |
Key Estimation
For key estimation, we use the MIREX metrics for the Audio Key Detection. These are described as:
Keys will be considered as “close” if they have one of the following relationships: distance of perfect fifth, relative major and minor, and parallel major and minor. A correct key assignment will be given a full point, and incorrect assignments will be allocated fractions of a point according to the following table:
| Relation to Correct Key | Points |
| Same | 1.0 |
| Perfect fifth | 0.5 |
| Relative major/minor | 0.3 |
| Parallel major/minor | 0.2 |
| Other | 0.0 |
The points are counted over all files and averaged. The number of correctly identified keys as well as the distribution of the errors is also reported.
Various Annotations - Different
| Algorithm | Same | Fifth | Relative | Parallel | Mirex |
|---|---|---|---|---|---|
| Edmkey | 80.57 | 6.40 | 2.21 | 5.96 | 85.63 |
| EdmkeyKrumhansl | 76.60 | 9.27 | 1.10 | 7.06 | 82.98 |
| EdmkeyTemperley | 67.33 | 2.87 | 8.83 | 1.77 | 71.77 |
| EdmkeyShaath | 80.57 | 6.40 | 2.21 | 5.96 | 85.63 |
| EssentiaBasic | 82.12 | 3.09 | 2.65 | 4.19 | 85.30 |
| QMULKeyDetector | 39.07 | 6.62 | 5.96 | 10.38 | 46.25 |
Various Annotations - Same
| Algorithm | Same | Fifth | Relative | Parallel | Mirex |
|---|---|---|---|---|---|
| Edmkey | 84.36 | 4.91 | 0.92 | 5.83 | 88.25 |
| EdmkeyKrumhansl | 79.45 | 7.67 | 0.31 | 7.36 | 84.85 |
| EdmkeyTemperley | 67.79 | 2.45 | 8.59 | 0.92 | 71.78 |
| EdmkeyShaath | 84.36 | 4.91 | 0.92 | 5.83 | 88.25 |
| EssentiaBasic | 86.50 | 2.15 | 1.23 | 4.29 | 88.80 |
| QMULKeyDetector | 35.28 | 5.52 | 5.52 | 12.27 | 42.15 |
Single Annotations
| Algorithm | Same | Fifth | Relative | Parallel | Mirex |
|---|---|---|---|---|---|
| Edmkey | 66.24 | 5.44 | 3.20 | 11.68 | 72.26 |
| EdmkeyKrumhansl | 59.84 | 7.20 | 3.20 | 12.96 | 66.99 |
| EdmkeyTemperley | 56.32 | 3.52 | 6.88 | 6.56 | 61.46 |
| EdmkeyShaath | 66.40 | 5.44 | 3.20 | 11.52 | 72.38 |
| EssentiaBasic | 66.88 | 3.20 | 2.72 | 9.76 | 71.25 |
| QMULKeyDetector | 28.64 | 5.44 | 3.36 | 13.60 | 35.09 |
Music Generation
In this section, we present example templates used for the generation of music pieces from loops, and some audio examples of the generated pieces. The code to run this experiments can be obtained at Jordan Smith’s repo.
Example Layouts
Music Examples
Factorial Layout Factorial1 Factorial2