Hello!
iNaturalist has been an exceptional citizen science platform, where its image classification model helps bridge the gap between casual users and expert reviewers. Through my own use, I learned a lot about plant species, initially by relying on the model’s suggestions for my observations, and later by engaging with specialists who refined or confirmed those identifications.
As I began working in bioacoustics, I often searched for audio samples in iNaturalist sound-based observations to train or validate custom classification models, so I started to think about creating a more accessible approach to sound classification to help iNaturalist users in their sound-only observations.
This project is a first step in that direction: a free and open-source desktop browser extension that runs bioacoustic neural network models directly in the browser, enabling analysis of sound recordings on iNaturalist observation pages. The extension performs local inference to identify species and cross-checks detection against geographic occurrence data from GBIF and iNaturalist.
For this initial release, I chose to support models in ONNX format, using ONNX Runtime for inference. This decision reflects a broader interest in promoting the interoperability and openness encouraged by the ONNX ecosystem. Two well-known bioacoustic models (BirdNET v2.4 and Perch v2.0) have been adapted to ONNX by Justin Chu and are included as built-in options in the extension’s model zoo. A key goal moving forward is to support adapting custom models to ONNX wherever possible. But also studying the options to support other formats in this extension.
Models can also include geographic bounding boxes that define where they are applicable. The extension uses this information to filter models based on the location of each observation. At present, models can be downloaded from repositories such as Hugging Face or Zenodo, and are saved in the browser cache. Users should be mindful of resource requirements when adding models to the extension (for example Perch v2.0 ~400MB, which is larger than BirdNET v2.4 ~60MB, and may require more RAM during inference).
Current available versions:
This is the first release, and feedback is very welcome. The project is open source and available on GitHub, open for contribution.
Cheers,
Kauê