Animal Sounds

Animal sounds is a repository that provides an end-to-end audio classification pipeline for animal vocalizations, including audio preprocessing, feature extraction, data augmentation, training, evaluation, and prediction.

What you can find here

• Getting Started: environment setup, data layout, and configuration file creation.
• Training workflow: preprocessing, feature extraction, and training.
• Prediction workflow: prepare new data, extract features, and run inference or evaluation.
• Methods and data: dataset summary, preprocessing choices, feature representations, and classifier details.
• Model comparison: rendered notebook with tables and plots.

Project at a glance

• Two model families are supported: SVM for lightweight deployment and CNN models for improved performance when sufficient training data and compute power is available.
• The pipeline was developed for chimpanzee vocalization detection and evaluated across distinct acoustic environments.
• Synthetic data generation is available to improve robustness on target background noise.

Software stack

• Python >=3.12
• Sklearn ~1.7.0
• Numpy ~2.2.6
• Pandas ~2.3.0
• torch~2.6.0
• torchaudio~2.0.2
• torchlibrosa~0.1.0

References

• Zwerts, J. A., Treep, J., Kaandorp, C. S., Meewis, F., Koot, A. C., and Kaya, H. (2021). Introducing a central African primate vocalisation dataset for automated species classification.
• Schuller, B. W., Batliner, A., Bergler, C., Mascolo, C., Han, J., Lefter, I., and others (2021). The INTERSPEECH 2021 Computational Paralinguistics Challenge: COVID-19 cough, COVID-19 speech, escalation and primates.
• Zwerts, J., Treep, J., Zahedi, P., and Kaandorp, C. (2024). Central African Primate vocalization bioacoustics dataset.

License

Project code is released under Apache 2.0. Some feature extraction scripts retain their original CeCILL 1.1 licensing; those files contain their own header notices.