Embedding learning with triplets created from perceptual decisions of birds about song similarity.

zf_embedding_learning: training (all approaches) and evaluation code.

Python 3 setup: We recommend setting up a Python "virtualenv" before continuing, so that the installed python packages will be stored local to your project rather than system-wide. Then activate the virtualenv and run:

  pip install -r requirements.txt

The Python script zf_embedding_learning.py file shows the full process of training models from triplet data. For this you will need the dataset, available from https://zenodo.org/record/5545872 , and also a computer suitable for deep learning training (e.g. with an NVIDIA GPU and "cuda/cudnn" installed).

To use our pretrained model without having to train something new: The script file apply_embedding.py shows a simple example of how to load the model, and apply it to a previously computed mel-spectrogram.