Fewshot Animal Sound Detection 13 (FASD13)

Fewshot Bioacoustic Sound Event Detection 13 (FASD13) Fewshot Bioacoustic Sound Event Detection (FSBSED) describes the task of detecting animal sounds in recordings based on only a handful of examples. It is of interest to researchers in ecology, animal behavior, and machine learning. A collection of public FSBSED datasets was previously provided in Nolasco et al., 2023 and Liang et al., 2024, but were designated as datasets for model training and validation. We complement these with Fewshot Animal Sound Detection 13 (FASD13), a public benchmark to be used for model evaluation. FASD13 consists of 13 bioacoustics datasets, each of which includes between 2 and 12 audio files. Eleven of these datasets were used from previous studies; they were chosen for their taxonomic diversity, varied recording conditions, and quality of their annotations. Two (CC and JS) are presented here for the first time. All datasets were developed alongside studies of ecology or animal behavior, and represent a range of realistic problems encountered in bioacoustics data. We follow the data format in Nolasco et al., 2023: Each audio file comes with annotations of the onsets and offsets of positive sound events, i.e. sounds coming from a predetermined category (such as a species label or call type). An N-shot detection system is provided with the audio up through the Nth positive event, and must predict the onsets and offsets of positive events in the rest of the recording. Evaluation of N-shot detection systems is described in loc. cit. Summary Table Datasets CC and JS are presented for the first time here. Other datasets have been released previously (see LICENSE.txt for citations). Terrestrial and underwater passive acoustic monitoring are abbreviated TPAM and UPAM, respectively. Dataset Full Name N Files Dur (hr) N Events Recording Type Location Taxa Detection Target AS AnuraSet 12 0.20 162 TPAM Brazil Anura Species CC Carrion Crow 10 10.00 2200 On-body Spain Corvus corone + Clamator glandarius Species+Life Stage GS Gunshot 7 38.33 85 TPAM Gabon Homo Sapiens Production Mechanism HA Hawaiian Birds 12 1.10 628 TPAM Hawaii, USA Aves Species HG Hainan Gibbons 9 72.00 483 TPAM Hainan, China Nomascus hainanus Species HW Humpback Whale 10 2.79 1565 UPAM North Pacific Ocean Megaptera novaeangliae Species JS Jumping Spider 4 0.23 924 Substrate Laboratory Habronattus Sound Type KD Katydid 12 2.00 883 TPAM Panamá Tettigoniidae Species MS Marmoset 10 1.67 1369 Laboratory Laboratory Callithrix jacchus Call Type PM Powdermill 4 6.42 2032 TPAM Pennsylvania, USA Passeriformes Species RG Ruffed Grouse 2 1.50 34 TPAM Pennsylvania, USA Bonasa umbellus Species RS Rana Sierrae 7 1.87 552 UPAM California, USA Rana sierrae Species RW Right Whale 10 5.00 398 UPAM Gulf of St. Lawrence Eubalaena glacialis Species Details Details of dataset collection and preprocessing steps are described in the attached appendix pdf. This file also contains example spectrograms for each dataset. Citation If you use this dataset, please cite our paper as well as the original source (see LICENSE.txt)