The problem is to classify a case as a right whale or not based on an audio signal.

That positive cases contain a set of right whale up-calls. Up-calls are the most commonly documented right whale vocalisation with an acoustic signature of approximately 60Hz-250Hz, typically lasting 1 second. Each series is labelled as either containing a right whale or not with the aim to correctly identify the series that contain up-calls.

Right whale calls can often be difficult to hear as the low frequency band can become congested with anthropogenic sounds such as ship noise, drilling, piling, or naval operations.

Each case is a two second audio segment sampled at 2kHz, giving a series length of 4000. We have done no preprocessing of the data. The dataset contains 10,934 train cases and 1962 test cases.

The data is from the Marinexplore and Cornell University Whale Detection Challenge.