BioDCASE 2026 Task 4: ATBFL Dataset

General Description This directory contains the development data for Task 4 of BioDCASE 2026 challenge. This is a curated subset of the ATBFL (Acoustic Trends Blue Fin Library). Overview: 15,633 audio file entries (corresponding to ~21.7 hours) with corresponding pre-computed embeddings Derived from ATBFL (one of the largest annotated datasets in marine bioacoustics, gathering underwater recordings around Antarctica from 2005 to 2017) Recordings from 11 different deployments organized as site-year datasets (e.g., kerguelen-2015, casey2014) Each sample can contain multiple overlapping whale call types: bma, bmb, bmz, bmd, bpd, bp20, bp20plus Folder Structure ATBFL_BASEAL_Data/ └── labels.csv # Main annotation file └── metadata.csv # Metadata file (recording location, annotator, etc.) └── embeddings/ └── perch_v2/ (.npy pre-computed embeddings from Perch v2 model) Dataset Statistics Split Breakdown Split Train Validation Total Count 12,506 (80%) 3,127 (20%) 15,633 Positive Samples (has calls) 9,981 (79.8%) 2,495 (79.8%) 12,476 (79.8%) Negative Samples (no call) 2,525 (20.2%) 632 (20.2%) 3,157 (20.25%) Class Distribution Class Description Count bma ABZ Call (Blue Whale) Constant-frequency tone between 25–28 Hz, no additional units 10,356 bmb ABZ Call Variant (Blue Whale) Similar to BmA but followed by partial/full inter-tone downsweep 7,240 bmz ABZ Call with Z-Shape (Blue Whale) Two tonal units (A and C), often with B downsweep between them, forming "Z" shape 5,432 bmd Downsweep (Blue Whale) Downsweep with continuous frequency modulation 4,065 bp20 Bp Call (Fin Whale) Pulsed calls with peak energy at 20 Hz 3,582 bp20plus Bp Call With Higher Freq. Energy (Fin Whale) Similar to Bp20 with additional energy at 80–100 Hz 2,414 bpd Downsweep (Fin Whale) Downsweep with continuous frequency modulation 2,312 Related Works ATBFL Library: Miller, B.S., Stafford, K.M., Van Opzeeland, I., Harris, D., Samaran, F., šIrović, A., Buchan, S., Findlay, K., Balcazar, N., Nieukirk , S., Leroy, E.C., Aulich, M., Shabangu, F.W., Dziak, R.P., Lee, W., Hong, J. (2020) An annotated library of underwater acoustic recordings for testing and training automated algorithms for detecting Antarctic blue and fin whale sounds, Ver. 1, Australian Antarctic Data Centre. http://dx.doi.org/doi:10.26179/5e6056035c01b. Perch 2.0 Model: van Merriënboer, B., Dumoulin, V., Hamer, J., Harrell, L., Burns, A., & Denton, T. (2025). Perch 2.0: The bittern lesson for bioacoustics. arXiv preprint arXiv:2508.04665. BioDCASE 2025 Challenge: Task 2: Jean-Labadye, L., Parcerisas, C., Miller, B., Carvaillo, P., Dubus, G., Farrugia, N., Gros-Martial, A., Marmoret, A., Moummad, I., Napoli, A., Nguyen Hong Duc, P., Raumer, P.-Y., Schall, E., White, E., Adam, O., Roch, M. A., White, P., & Cazau, D. (2025). BioDCASE 2025 task 2: Development set (Version v1) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.15092732