Xeno-Canto Bird Sound Dataset

This repository provides access to the Xeno-Canto bird sound dataset (checkpoint from 2022-07-18) used in the benchmark BIRB, specifically pre-processed to facilitate training deep learning models.

The dataset has been processed using CNN14 from PANNs, a model pre-trained on AudioSet, to select 6-second windows with the highest bird sound activation. All audio has been downsampled to 16kHz and converted into Pytorch format (.pt), optimizing it for efficient use in training deep learning models.

This version of the dataset was utilized in the study "Domain-Invariant Representation Learning of Bird Sounds" as training data for feature extraction models. The study further evaluated the few-shot learning capabilities of these models using a variety of soundscape datasets, also from the BIRB benchmark, which were preprocessed and saved in .pt format. These pre-processed evaluation datasets are available on Zenodo.

Dataset Structure

• 684,744 audio segments, each 6 seconds long
• 10,127 bird species classes
• Compressed into 13 tar files, each approximately 20GB in size
• Total size ~250GB

Metadata

The dataset includes a .csv file (taxonomy_info_2022-07-18.csv) with metadata for each file:

• Species code
• Scientific name
• Sound type (e.g., call, song)
• Recording details (e.g., recordist name, country)
• Others

Usage

1. Download the dataset using the provided download.py script:

• Ensure you have the huggingface_hub library installed. If not, you can install or upgrade it using the following command:

  pip install --upgrade huggingface_hub
  

• Set the DESTINATION_PATH variable in the script to specify where the data should be saved.

2. For instructions on data loading, model implementation, training, and evaluation, see ProtoCLR Github Repository.

Bonus

For using a feature extractor trained on Xeno-Canto, see ProtoCLR Hugging Face Repository.

Citation

If you use this dataset or our feature extractor ProtoCLR in your research, please cite the following paper:

@misc{moummad2024dirlbs,
      title={Domain-Invariant Representation Learning of Bird Sounds}, 
      author={Ilyass Moummad and Romain Serizel and Emmanouil Benetos and Nicolas Farrugia},
      year={2024},
      eprint={2409.08589},
      archivePrefix={arXiv},
      primaryClass={cs.SD},
      url={https://arxiv.org/abs/2409.08589}, 
}