Part of MONSTER: https://arxiv.org/abs/2502.15122.

WhaleSounds
Category	Audio
Num. Examples	105,163
Num. Channels	1
Length	2,500
Sampling Freq.	250 Hz
Num. Classes	8
License	CC BY 4.0
Citations	[1] [2]

WhaleSounds consists of underwater acoustic recordings around Antarctica, manually annotated for seven different types of whale calls [1, 2]. The recordings are single channel. The original data consists of extended recordings with a mixture of different sampling rates between 250 and 2,500 Hz. The dataset has been processed to extract the annotated whale calls from the original recordings. The extracted whale calls have been resampled to a consistent sampling frequency of 250 Hz. (The whale sounds are typically well below 100 Hz.) The processed dataset contains 105,163 (univariate) time series, each of length 2,500 (i.e., each time series represents 10 seconds of data at 250 Hz, approximately centred on the labelled whale sound), with eight classes representing the seven types of whale call plus a class for unidentified sounds. This version of the dataset has been split into stratified random cross-validation folds.

[1] Brian S Miller, Kathleen M Stafford, Ilse Van Opzeeland, et al. (2020) Whale Sounds. https://data.aad.gov.au/metadata/AcousticTrends_BlueFinLibrary. CC BY 4.0.

[2] Brian S Miller, Kathleen M Stafford, Ilse Van Opzeeland, Danielle Harris, Flore Samaran, Ana Širović, Susannah Buchan, Ken Findlay, Naysa Balcazar, Sharon Nieukirk, Emmanuelle C Leroy, Meghan Aulich, Fannie W Shabangu, Robert P Dziak, Won Sang Lee, and Jong Kuk Hong. (2021). An open access dataset for developing automated detectors of Antarctic baleen whale sounds and performance evaluation of two commonly used detectors. Scientific Reports, 11.