BioDCASE 2026 Task 4: BirdSet Dataset

BirdSet_BASEAL contains exported embedding bundles for three BirdSet subsets used in the BASEAL workflow: HSN_BASEAL, POW_BASEAL, and UHH_BASEAL. Directory Overview HSN_BASEAL POW_BASEAL UHH_BASEAL Each subdirectory contains a ready-to-load embedding bundle with README.md, labels.csv, metadata.csv, and embeddings/perchv2/*.npy. Shared Bundle Format All three subset bundles share the same export layout and metadata conventions: Source label column: ebird_code_multilabel Packaged split scope: train and validation Embedding model: Perch v2 (Perch 2.0) Embedding dimension: 1536 Join key across bundle files: filename Main files: README.md, labels.csv, metadata.csv, embeddings/perchv2/*.npy Detailed column descriptions, class inventories, and loading examples remain in each subset-specific README.md. Embedding Model The exported .npy files in embeddings/perchv2 were generated with the Perch v2 model handle used for this export: https://www.kaggle.com/models/google/bird-vocalization-classifier/tensorFlow2/perch_v2/2. For model attribution, reference the Perch 2.0 paper: https://research.google/pubs/perch-20-the-bittern-lesson-for-bioacoustics/. The official project repository is https://github.com/google-research/perch, which is published under the Apache-2.0 license. Perch model/software attribution is separate from the BirdSet and source-recording provenance documented in each subset's metadata.csv. Subset Summary Subset Config Samples Processed Classes Positive Assignments Avg. Labels/Sample Splits HSN_BASEAL HSN 8400 19 4398 0.524 train=6600, validation=1800 POW_BASEAL POW 2964 41 8398 2.833 train=2280, validation=684 UHH_BASEAL UHH 25646 25 27121 1.058 train=18319, validation=7327 POW is the densest multilabel subset, HSN is the sparsest, and UHH is the largest by sample count. Where to Find More Detail BirdSet_BASEAL/HSN_BASEAL/README.md BirdSet_BASEAL/POW_BASEAL/README.md BirdSet_BASEAL/UHH_BASEAL/README.md These subset-specific documents contain the full class inventory, a typical loading example, and the complete metadata.csv field descriptions for each exported bundle. Citation Information These exported bundles are derived from BirdSet subset exports and were embedded with Perch v2 (Perch 2.0). If you use data from BirdSet_BASEAL, cite BirdSet, cite the Perch 2.0 model/paper, and also cite or acknowledge the upstream source record listed for the subset you use. The metadata.csv files are internally consistent within each subset and point to the following upstream records: Subset Source record License in metadata.csv HSN_BASEAL https://zenodo.org/record/7525805 Creative Commons Attribution 4.0 International Public License POW_BASEAL https://zenodo.org/record/4656848 Creative Commons Zero v1.0 Universal UHH_BASEAL https://zenodo.org/record/7078499 Creative Commons Attribution 4.0 International Public License This overview does not introduce a new citation entry for the datasets themselves. For formal citation text, use the upstream BirdSet dataset or publication together with the relevant upstream source record for the subset you analyze. Licensing The exported bundles preserve source provenance in metadata.csv, including the license and source columns. In the current exports, each subset uses a single source record and a single recorded license across all packaged samples. HSN_BASEAL and UHH_BASEAL record Creative Commons Attribution 4.0 International Public License. POW_BASEAL records Creative Commons Zero v1.0 Universal. These data-license notes are separate from the embedding-model software license. The official Perch repository is published under Apache-2.0; users should review the model source and distribution page when reusing the embedding model itself. Use of these exports remains subject to the licensing and usage terms associated with BirdSet and the underlying source records listed above. This directory does not assign a single new data license across all bundled subsets. Disclaimer We carefully selected and composed this dataset's content. If you believe that any of this content violates licensing agreements or infringes on intellectual property rights, please contact us immediately (cf. contact information). In such a case, we will promptly investigate the issue and remove the implicated data records from our dataset if necessary. Users are responsible for ensuring that their use of the dataset complies with all licenses, applicable laws, regulations, and ethical guidelines. We make no representations or warranties of any kind and accept no responsibility in the case of violations.