This dataset contains short WAV audio clips curated from field recordings collected at The Wilds (Ohio, USA) for binary audio classification: positive clips contain Pere David’s deer vocalizations, and negative clips contain background/non-target audio. Positives are organized by song member ID (one folder per song member), and negatives are stored in a separate folder with filenames that include song member, timestamp, and negative subtype. The dataset is designed to support:

automated screening of long-duration wildlife recordings
model development for species call detection
ecological acoustics workflows and monitoring prototypes

Dataset Details

This dataset is intended to support analysis of Pere David's Deer behaviors and detection of Pere David's Deers by providing curated examples of the Pere David's Deer vocalizations.

Supported Tasks and Leaderboards

Supported tasks:

binary audio classification: positive (deer vocalization) vs negative (background/non-target)
acoustic event detection / call detection
representation learning and embedding model development

Dataset Structure

Audio is organized under a single top-level folder with two class folders:

positives: deer vocalization clips grouped by song member ID (twXX-smYY/)
negatives: non-target/background clips (e.g., silence, rain). Filenames encode song member, timestamp, and negative subtype.

The positives folder is further organized using the specific song meters, as shown below:

positives/
  tw02-sm04/
    <wav clips...>
  tw05-sm01/
    <wav clips...>
  tw06-sm03/
    <wav clips...>
  tw07-sm02/
    tw07-sm02_SM002_..._95900_1077-1080s.wav
    tw07-sm02_SM002_..._95900_1086-1089s.wav
    ...
  tw08-sm04/
    <wav clips...>

Class Balance

Positives: 369
Negatives: 5423
Total: 5792
Imbalance: negatives are ~14.7× the positives (5423 / 369)

Data Instances

Each instance is a WAV file. Positive instances contain a short Pere David’s deer vocalization segment; negative instances contain background/non-target audio. Clips were extracted from longer field recordings and standardized to 16 kHz mono WAV format.

Data Fields

This dataset does not require structured metadata files to use. The directory structure and filenames provide the primary organization.

Data Splits

No predefined train, validation, or test splits are provided.

Recommended evaluation practice is to split by song member, holding out entire song member folders for testing.

Dataset Creation

Curation Rationale

The dataset was created to support automated tools for detecting Pere David's deer vocalizations in long-duration field recordings. Automated detection can help researchers screen large audio archives and accelerate ecological acoustics workflows.

Source Data

Data Collection and Processing

Audio clips were extracted from longer field recordings collected at The Wilds in 2025. Positive clips were selected based on confirmed Pere David's deer vocalizations. Negative clips were selected as background/non-target audio and labeled with negative subtypes (e.g., silence, rain) in the filename. Audio was standardized to 16 kHz mono WAV format. Clip names were cleaned and renamed consistently.

This dataset includes both positives and negatives and can be used directly for binary training; however, the class distribution is highly imbalanced and should be handled with appropriate sampling/weighting and evaluation metrics.

Original Source Dataset

This dataset is a derived, curated subset of:

https://huggingface.co/datasets/imageomics/thewilds_bioacousticmonitors

Who are the source data producers?

Source audio was collected and processed by members of the Imageomics Institute. Clip extraction and dataset curation were performed as part of the same effort.

Personal and Sensitive Information

The dataset contains no personally identifiable human information. Data consists of environmental audio and wildlife vocalizations.

Considerations for Using the Data

This dataset is best used for binary detection/classification with curated positive vocalizations and curated negative/background clips.

Bias, Risks, and Limitations

Although this dataset includes negative examples, the dataset is highly imbalanced (many more negatives than positives), and models may bias toward predicting the negative class without careful training and evaluation.

Out-of-scope use includes identifying individual animals, detecting species other than Pere David's deer, demographic inference, safety-critical decision making, or population estimation without expert ecological validation.

Recommendations

Pair these clips with background audio collected under similar conditions for training and calibration. Use song member holdouts for evaluation. Report results on separate sessions when available, and consider robustness checks across different noise conditions and devices.

Licensing Information

This dataset is released under CC0 1.0. You may use it without restriction. Attribution is appreciated for academic and research use.

Citation

Viswapriyan, Varun. Pere David's Deer Vocalizations (The Wilds, Ohio). 2025. Hosted on Hugging Face Datasets.

BibTeX (dada):

@misc{viswapriyan_pere_davids_deer_vocalizations_2025,
  author = {Viswapriyan, Varun},
  title  = {Pere David's Deer Vocalizations (The Wilds, Ohio)},
  year   = {2025},
  publisher = {Hugging Face},
  url = {https://huggingface.co/datasets/imageomics/pere-david-deer-vocalizations}
}

Important: This dataset is a derived product from the original source dataset. If you use this dataset, please also cite the original data source:

@misc{thewilds_bioacoustics_2025,
  author    = {Tanishka Wani and Vedant Patil and Rugved Katole and Bharath Pillai and Anirudh Potlapally and Ally Bonney and Jenna Kline},
  title     = {The Wilds Bioacoustic Monitors},
  year      = {2025},
  publisher = {Hugging Face Datasets},
  url       = {https://huggingface.co/datasets/imageomics/thewilds_bioacousticmonitors}
}

Acknowledgements

This work was supported by the Imageomics Institute, which is funded by the US National Science Foundation's Harnessing the Data Revolution (HDR) program under Award #2118240 (Imageomics: A New Frontier of Biological Information Powered by Knowledge-Guided Machine Learning). Any opinions, findings and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.

Downloads last month: 38

Curated by: Varun Viswapriyan

Language(s) (NLP): English (metadata and documentation only)