iNaturalist Sounds Dataset Natural Species Sound Dataset

iNaturalist Sounds Dataset (iNatSounds) is a collection of audio files submitted by contributors to the global citizen science platform iNaturalist in 2024. The dataset collects 230,000 audio files, capturing sounds from more than 5,500 species, contributed by more than 27,000 recorders worldwide. This dataset contains the sounds of birds, mammals, insects, reptiles, and amphibians, and the audio and species labels are derived from observation records submitted to iNaturalist.

Each recording in the dataset is of varying length and contains annotations for a single species. Despite the weak labeling, the study demonstrates that iNatSounds is robust as a pre-training resource and outperforms strongly labeled downstream evaluation datasets. The dataset is provided in a single, freely accessible archive, promoting accessibility and research in this important field.

The applications of iNatSounds are promising, and models trained on this data are expected to power the next generation of public engagement applications and assist biologists, ecologists, and land managers in processing large audio collections, thereby helping to understand species composition in different soundscapes.