MULE

The Musicset Unsupervised Large Embedding (MULE) module is your music-audio workhorse!

This module contains SCOOCH configurable code to run a simple analysis pipeline to extract audio embeddings from audio files which may then be used for downstream music understanding purposes.

This module requires FFMpeg to read audio files, which may be downloaded here.

In order to create MULE embeddings, you will need a SCOOCH configuration describing the pipeline, and the model weights. Both are licensed under the CC BY-NC 4.0 license, and can be found in this module's github repository.

To create embeddings for a single audio file, e.g., test.wav in the current directory, you can use this module in conjunction with the provided configuration and model weights:

pip install sxmp-mule
git clone https://github.com/PandoraMedia/music-audio-representations.git
cd ./music-audio-representations
mule analyze --config ./supporting_data/configs/mule_embedding.yml -i ../test.wav -o ./embedding.npy

For more information on this module, please check out the publication:

Supervised and Unsupervised Learning of Audio Representations for Music Understanding, M. C. McCallum, F. Korzeniowski, S. Oramas, F. Gouyon, A. F. Ehmann.