- Model v 4.4: https://github.com/plauha/BSG_classifier_builder/tree/main/Run%20BSG%20models/models/Finland/model_v4
- Model v 3.5: https://github.com/plauha/BSG_classifier_builder/tree/main/Run%20BSG%20models/models/Finland/model_v3
This repository directs to the BSG – Finnish Birds Model, which is a pretrained bird sound classification model fine-tuned for Finland. The model has been developed at the University of Jyväskylä and is based on the BirdNET model. It uses an EfficientNet-B0 backbone (from BirdNET-Analyzer) with a custom head trained on vocalizations of 263 Finnish bird species, covering all breeders, non-breeding migrants, and most common vagrants.
The model processes audio in 3-second overlapping segments (spectrogram-based) and outputs species-wise detection probabilities, calibrated per species via logistic regression. Training data combined expert-annotated clips from Xeno-canto, Finnish soundscapes, targeted field recordings, and selected mobile phone recordings. Predictions are further filtered by species' seasonal and geographic plausibility to reduce obvious misclassifications.
See the LICENSE file for more information on how the model can be used. Usage is limited to non-commercial purposes.
The code for using the model is available in another repository: https://github.com/luomus/bird-identification.
Nokelainen, O., Lauha, P., Andrejeff, S., Hänninen, J., Inkinen, J., Kallio, A., Lehto, H.J., Mutanen, M., Paavola, R., Schiestl-Aalto, P., Somervuo, P., Sundell, J., Talaskivi, J., Vallinmäki, M., Vancraeyenest, A., Lehtiö, A. and Ovaskainen, O. (2024) ‘A Mobile Application–Based Citizen Science Product to Compile Bird Observations’, Citizen Science: Theory and Practice, 9(1), p. 24. https://doi.org/10.5334/cstp.710
Patrik Lauha, Meeri Rannisto, Panu Somervuo, et al. Bird Sounds Global - model builder: An end-to-end workflow for building locally fine-tuned bird classifiers. Authorea. December 17, 2025. https://doi.org/10.22541/au.176599468.88578746/v1