---
dataset_info:
  features:
  - name: Selection
    dtype: int64
  - name: BeginTime
    dtype: float64
  - name: EndTime
    dtype: float64
  - name: LowFreq
    dtype: float64
  - name: HighFreq
    dtype: float64
  - name: CallType
    dtype: string
  - name: Filename
    dtype: string
  - name: VocalizationDuration
    dtype: float64
  - name: Audio
    dtype:
      audio:
        sampling_rate: 44100
  splits:
  - name: train
    num_bytes: 2573497129.02
    num_examples: 1514
  download_size: 1825156822
  dataset_size: 2573497129.02
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
---

# Orcasound Humpback Whale Dataset
## Overview
This dataset is built on top of the **Orcasound project** (https://www.orcasound.net/) and specifically utilizes the "Haro Humpback" catalog curated by Emily Vierling and Scott Veirs. The catalog contains audio recordings of humpback whale vocalizations, along with associated annotations. These recordings were collected as part of Orcasound's efforts to monitor and protect marine life through acoustic observation.

The dataset available here has been preprocessed and uploaded to HuggingFace by Liana Napalkova, based on the original catalog from Orcasound.

## Dataset Contents
- Audio Files: This dataset includes recordings of humpback whale vocalizations in WAV format, sampled at a rate of 44,100 Hz. Each recording has a fixed duration of 5 seconds.
- Annotations: The dataset also includes annotations that provide detailed information about the recorded whale sounds.

## Data Source
The original data was retrieved from the public sources provided by Orcasound. The data retrieval process involved using the quilt3 library to download the necessary files from the "acoustic-sandbox" Amazon S3 bucket.

## Code to Retrieve Original Data
If you wish to retrieve the original data directly from Orcasound, you can use the following Python code:

```
import quilt3

# Create a connection to the S3 bucket
b = quilt3.Bucket("s3://acoustic-sandbox")

# Download the annotation files
b.fetch("humpbacks/Emily-Vierling-Orcasound-data/Em_HW_data/Annotations/", f"./raw/annotations/")

# Download the audio data
b.fetch("humpbacks/Emily-Vierling-Orcasound-data/Em_HW_data/flac_files/", f"./raw/audio/")
```

## Related Dataset
The relevant dataset containing noise recordings without whales vocalizations is available on HuggingFace: https://huggingface.co/datasets/Liana/humpbacks-orcasound-em-hW-data-noise. These recordings are also 5 seconds long.

## Acknowledgments
Special thanks to Emily Vierling and Scott Veirs for curating the "Haro Humpback" catalog and to Orcasound for making this data publicly available.

## Usage
This dataset can be used for research purposes, particularly in the fields of marine biology, bioacoustics, and machine learning. 
Please cite the original sources and the creators if you use this dataset in your research or projects.

## License
Please refer to the Orcasound project and the HuggingFace dataset page for licensing information related to the use of this dataset.