BirdNET+ V3.0 model developer preview

This is a preview version of the BirdNET+ V3.0 model family for early access experimentation and development. Most recent version: - BirdNET+ V3.0 Developer Preview 3.1 - 11K Species (Jun 2026) - Improves model performance despite larger species set - available for ONNX, PyTorch, Tensorflow and LiteRT - Still lacks a noise fallback class, to be added in future releases - Species list needs a cleanup and includes some weird entries Key changes vs earlier model versions: - Variable-length input (removed fixed 3 s constraint) - Model takes 32 kHz audio input (compared to 48 kHz previously) - Improved architecture and training procedure - Much larger and more diverse training dataset - Expanded set of non-bird species Pending revisions: - Cross-platform / cross-framework packaging - Species list curation (inclusion/exclusion based on data availability) - Final architecture and model size - Additional non-target / environmental classes (human, rain, wind, engines, etc.) Note: Developer preview; models, labels, and code will change. Trained on a subset of data and may not reflect final performance. Please refer to the TERMS_OF_USE.txt file for detailed terms and conditions regarding the use of the BirdNET+ V3.0 developer preview models. More info and code examples at https://github.com/birdnet-team/birdnet-V3.0-dev