curated audio datasets for animal research
- zebra_finch - 3405 zebra finch calls classified across 11 call types. Additonal labels include name of individual making the vocalization and its age (chick or adult).
- macaques - 7285 macaque coo calls from 8 individuals (4 males and 4 females). There is a collaborative tutorial of techniques to recover identity from voice.
- giant otter - A tutorial demonstrating a complete ML pipeline applied to giant otter bioacoustics, beginning with data preprocessing, proceeding to load the data, and culminating in the construction and training of a CNN-based classifier capable of labeling giant otter vocalizations according to call type.
- Egyptian fruit bats- Approxiamtely 8k Egyptian fruit bat vocalizations classified on interaction context using fastai's pretrained resnet models.
All datasets are accessible by issuing a single command from within the fastai v2 library.
| dataset | architecture |
|---|---|
| giant otter | conv2d classifier with an interactive gui |
| macaques | conv1d classifier on raw audio |
| macaques | xresnet18 classifier with fastai audio |
| macaques | pretrained resnet18 using fastai DataBlock api and error analysis |
| macaques | ROCKET model extracting information from raw audio using conv1d without training |
| zebra finch | pretrained resnet18 classifier with confusion matrix using fastai |