Learning Similarity Metrics for Melody Retrieval

Full paper; Supplementary material

Similarity measures are indispensable in music information retrieval. In recent years, various proposals have been made for measuring melodic similarity in symbolically encoded scores. Many of these approaches are ultimately based on a dynamic programming approach such as sequence alignment or edit distance, which has various drawbacks. First, the similarity scores are not necessarily metrics and are not directly comparable. Second, the algorithms are mostly first-order and of quadratic time-complexity, and finally, the features and weights need to be defined precisely.

We propose an alternative approach which employs deep neural networks for end-to-end similarity metric learning. We contrast and compare different recurrent neural architectures (LSTM and GRU) for representing symbolic melodies as continuous vectors, and demonstrate how duplet and triplet loss functions can be employed to learn compact distributional representations of symbolic music in an induced melody space. We contrast this approach with an alignment-based approach. We present results for the Meertens Tune Collections, which consists of a large number of vocal and instrumental monophonic pieces from Dutch musical sources, spanning five centuries, and we demonstrate the robustness of the learned similarity metrics.

Citation

When using the code, or data associated with this paper and repository, please cite:

Karsdorp F, Van Kranenburg P & Manjavacas E. “Learning Similarity Metrics for Melody Retrieval”, 20th International Society for Music Information Retrieval Conference, Delft, The Netherlands, 2019.

@inProceedings{karsdorp_learning_melody_sims,
  author    = {Folgert Karsdorp and Peter {Van Kranenburg} and Enrique Manjavacas},
  title     = {Learning Similarity Metrics for Melody Retrieval},
  booktitle = {20th International Society for Music Information Retrieval Conference},
  address   = {Delft, The Netherlands},
  pages     = {478-485},
  year      = {2019}
}

Requirements

The code requires Python version 3.6 or higher. All dependencies are listed in requirements.txt and can be installed by running the following command in your shell:

pip install -r requirements.txt

Training

The script train.py can be used for training. The script provides many different options, which can be accessed using:

python train.py --help

A typical training run is done using something like:

python train.py --train ../data/datasets/FSINST_train.jsonl \
  --dev ../data/datasets/FSINST_dev.jsonl --test ../data/datasets/FSINST_test.jsonl \
  --categorical_features scaledegree pitch metriccontour imacontour \
  --continuous_features duration beat beatstrength imaweight phrasepos \
  --example_type pairs --n_classes 5 \
  --n_samples 5 --n_layers 2 --model LSTM --emb_dim 8 --hid_dim 64 --dropout 0.5 \
  --loss cosine --margin 0.4 --weight 0.25 --lr 0.001 --batch_size 20 \
  --log_interval 10 --n_workers 5 --cuda --early_stop_score MAP --balanced_batch_sampler \
  --online_sampler --patience 10 --scaler zscore 

License

Karsdorp, Van Kranenburg, Manjavacas. Licensed under a Creative Commons Attribution 4.0 International License (CC BY 4.0). Attribution: Karsdorp, Van Kranenburg, Manjavacas. “Learning Similarity Metrics for Melody Retrieval”, 20th International Society for Music Information Retrieval Conference, Delft, The Netherlands, 2019