CoolFace
Datasetpublic

OpenWhistleNeurIPS26/OpenWhistle-Pretraining

OpenWhistle Pretraining Dataset OpenWhistleNeurIPS26/OpenWhistle-Pretraining is the public unlabeled audio dataset used for OpenWhistle pretraining. It contains 96 kHz dolphin acoustic segments with timing and recording metadata, but no whistle/noise labels. The main default config is the complete pretraining dataset. A smaller deterministic review-sample config is also provided so reviewers can inspect representative examples quickly without downloading the full dataset.… See the full description on the dataset page: https://huggingface.co/datasets/OpenWhistleNeurIPS26/OpenWhistle-Pretraining.

sourceHugging Faceupdated 5mo agoView on Hugging Face
0likes14downloads
Dataset Card

OpenWhistle Pretraining Dataset

OpenWhistleNeurIPS26/OpenWhistle-Pretraining is the public unlabeled audio dataset used for OpenWhistle pretraining. It contains 96 kHz dolphin acoustic segments with timing and recording metadata, but no whistle/noise labels.

The main default config is the complete pretraining dataset. A smaller deterministic review-sample config is also provided so reviewers can inspect representative examples quickly without downloading the full dataset.

Dataset contents

  • Hugging Face repo: OpenWhistleNeurIPS26/OpenWhistle-Pretraining
  • Public columns: audio, start_time, end_time, duration, year, hydrophone
  • Sampling rate: 96 kHz, mono audio
  • Labels: none; this dataset is intended for unsupervised or self-supervised pretraining

Full dataset splits

SplitRowsDuration (s)Duration (h)
train28,410367,792.80102.165
validation3,37043,629.2012.119
Total31,780411,422.00114.284

Full dataset coverage by year

Split20192020202120232024
train9201,42014,3769,8851,809
validation1191791,8201,055197
Total1,0391,59916,19610,9402,006

Full dataset coverage by hydrophone

Splitchannel_0channel_1channel_2
train20,8087,130472
validation2,59974130
Total23,4077,871502

Full dataset coverage by year and hydrophone

Split2019 ch02020 ch02021 ch02023 ch02023 ch12023 ch22024 ch02024 ch1
train9201,42014,3763,9825,4314721101,699
validation1191791,8204645613017180
Total1,0391,59916,1964,4465,9925021271,879

Review sample

The review-sample config is a small deterministic subset of the same public dataset. It was created only to make review and manual inspection easier. It is not a replacement for the full dataset used for model development or reporting.

How the review sample was created

The review sample was designed to preserve the structure of the full pretraining dataset while keeping the download small enough for quick manual inspection. Because the pretraining dataset is unlabeled, the sample is not class-balanced. Instead, it preserves the original train and validation split names and uses the same audio and metadata columns as the full dataset.

The target size was set to 480 rows, following the same review-sample size used for the OpenWhistle CNN dataset. Rows were allocated across splits to keep the same approximate train/validation ratio as the full pretraining dataset: 430 examples for train and 50 examples for validation.

For each split, the dataset was streamed from OpenWhistleNeurIPS26/OpenWhistle-Pretraining, the audio column was cast to Audio(sampling_rate=96000, mono=True, decode=False), and rows were shuffled with a fixed seed and a finite shuffle buffer before taking the requested number of examples. Sampling used base seed 42, shuffle buffer size 64, and split-specific seeds derived from the split order (train=42, validation=43). The resulting config is named review-sample.

Review sample size

SplitRowsDuration (s)Duration (h)
train4306,017.601.672
validation50642.400.178
Total4806,660.001.850

Review sample coverage

SplitYear countsHydrophone counts
train2019: 12; 2020: 33; 2021: 218; 2023: 137; 2024: 30channel0: 336; channel1: 93; channel_2: 1
validation2021: 20; 2023: 30channel0: 37; channel1: 11; channel_2: 2

Loading the data

python
from datasets import load_dataset

full = load_dataset("OpenWhistleNeurIPS26/OpenWhistle-Pretraining")
review = load_dataset("OpenWhistleNeurIPS26/OpenWhistle-Pretraining", "review-sample")