The dataset viewer is not available for this subset.
Exception: SplitsNotFoundError
Message: The split names could not be parsed from the dataset config.
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 286, in get_dataset_config_info
for split_generator in builder._split_generators(
~~~~~~~~~~~~~~~~~~~~~~~~~^
StreamingDownloadManager(base_path=builder.base_path, download_config=download_config)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/webdataset/webdataset.py", line 80, in _split_generators
raise ValueError(
...<2 lines>...
)
ValueError: The TAR archives of the dataset should be in WebDataset format, but the files in the archive don't share the same prefix or the same types.
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
File "/src/services/worker/src/worker/job_runners/config/split_names.py", line 68, in compute_split_names_from_streaming_response
for split in get_dataset_split_names(
~~~~~~~~~~~~~~~~~~~~~~~^
path=dataset,
^^^^^^^^^^^^^
config_name=config,
^^^^^^^^^^^^^^^^^^^
token=hf_token,
^^^^^^^^^^^^^^^
)
^
File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 340, in get_dataset_split_names
info = get_dataset_config_info(
path,
...<6 lines>...
**config_kwargs,
)
File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 291, in get_dataset_config_info
raise SplitsNotFoundError("The split names could not be parsed from the dataset config.") from err
datasets.inspect.SplitsNotFoundError: The split names could not be parsed from the dataset config.Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
ONCship: A Large-Scale Real-World Underwater Acoustic Dataset
Dataset Description
ONCship is a large-scale underwater acoustic dataset constructed from hydrophone recordings provided by Ocean Networks Canada (ONC) and vessel information from the Automatic Identification System (AIS). It is designed to support research on underwater acoustic representation learning, vessel detection and classification, and fine-grained vessel identification.
The dataset comprises approximately 4,521 hours of curated real-world audio, organized into 10-second clips derived from recordings sampled at 32 kHz. The collection covers a variety of vessel types and acoustic conditions, including single-vessel, multiple-vessel, and vessel-absent scenarios as defined by the dataset's AIS-based selection procedure.
This Hugging Face repository distributes audio recordings directly. The final organization of accompanying metadata and evaluation splits will be documented as they become available.
Dataset Statistics
| Subset | Duration (hours) |
|---|---|
| Single-ship (with vessel-type labels) | 3,106.8 |
| Multi-ship | 929.1 |
| USSLP recordings | 439.3 |
| Single-ship (without vessel-type labels) | 38.4 |
| No-ship | 7.4 |
| Total | 4,521.0 |
The dataset includes 22 vessel types. The four most represented vessel types in the labeled single-ship subset are Tug, Cargo, Passenger, and Towing.
Data Collection and Processing
The original audio recordings were collected by ONC hydrophones. Where AIS records were available, vessel trajectories were associated with hydrophone recording intervals to identify candidate single-ship, multi-ship, and no-ship segments. Selected recordings were then processed into shorter audio clips.
The USSLP recordings are retained as a separate subset because their corresponding AIS records were not downloaded during dataset construction.
Labeling note: Vessel presence and vessel-type information are derived from AIS-based association rather than manual verification of every audible event. Consequently, AIS-based labels should not be interpreted as guarantees that a vessel is acoustically audible—or that no other acoustic source is present—in every clip.
Intended Uses
ONCship is intended for research applications, including:
- Self-supervised and supervised underwater acoustic representation learning.
- Vessel detection and vessel-type classification.
- Fine-grained vessel identification using vessel identity information, where available.
- Studies of generalization and robustness in real-world underwater acoustic environments.
Limitations
Hydrophone location, recording conditions, vessel distance, background noise, AIS coverage, and vessel traffic patterns may influence acoustic characteristics and label reliability.
Users should take these factors into account when designing training and evaluation protocols. In particular, where vessel identifiers are available, vessel-disjoint train/test splits are recommended to reduce identity leakage.
Data Source and Acknowledgments
We acknowledge Ocean Networks Canada (ONC) for the hydrophone recordings used to construct ONCship. AIS records were used to associate vessel activity with recording intervals.
License and Redistribution
The license and redistribution conditions for the released ONCship audio are being finalized. Users should consult the license displayed on this dataset repository once it is published. The rights and attribution requirements associated with the original ONC recordings and AIS data must also be respected.
Citation
The associated paper and BibTeX citation will be added upon publication.
Updates
The repository will be updated with release details, metadata documentation, and evaluation instructions.
- Downloads last month
- 56