Datasets:
tags:
- time series
- time series classification
- monster
- audio
pretty_name: CornellWhaleChallenge
size_categories:
- 10K<n<100K
license: other
Part of MONSTER: https://arxiv.org/abs/2502.15122.
| CornellWhaleChallenge | |
|---|---|
| Category | Audio |
| Num. Examples | 30,000 |
| Num. Channels | 1 |
| Length | 4,000 |
| Sampling Freq. | 2 kHz |
| Num. Classes | 2 |
| License | Cornell |
| Citations | [1] |
CornellWhaleChallenge consists of hydrophone recordings [1]. The recordings are single channel with a sampling rate of 2 kHz. The recordings come from an array of buoys near Boston. The processed dataset consists of 30,000 (univariate) time series, each of length 4,000 (i.e., representing recordings of 2 seconds of audio with a sampling rate of 2 kHz). The task is to distinguish right whale calls from other noises. (An abridged version of this dataset is included in the broader UCR archive.) This version of the dataset has been divided into stratified random cross-validation folds.
[1] André Karpištšenko, Eric Spalding, and Will Cukierski. (2013). The Marinexplore and Cornell University whale detection challenge. https://kaggle.com/competitions/whale-detection-challenge. Copyright 2011 Cornell University and the Cornell Research Foundation.