metadata

dataset_info:
  features:
    - name: file_name
      dtype: large_string
    - name: species_name
      dtype: large_string
    - name: group
      dtype: large_string
    - name: license
      dtype: large_string
    - name: contributor
      dtype: large_string
    - name: observation
      dtype: large_string
    - name: file
      dtype: large_string
    - name: background
      dtype: large_string
    - name: original_name
      dtype: large_string
    - name: subset
      dtype: large_string
    - name: audio
      dtype: audio
  splits:
    - name: train
      num_bytes: 10599233926.978176
      num_examples: 15873
    - name: validation
      num_bytes: 3561890667.7648244
      num_examples: 5307
  download_size: 92104242107
  dataset_size: 14161124594.743
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: validation
        path: data/validation-*
license: cc-by-4.0
task_categories:
  - audio-classification
tags:
  - biology
pretty_name: >-
  InsectSet459: A large dataset for automatic acoustic identification of insects
  (Orthoptera and Cicadidae)

InsectSet459: An Open Dataset of Insect Sounds for Bioacoustic Machine Learning

Overview

InsectSet459 is a comprehensive dataset of insect sounds designed for developing and testing machine learning algorithms for automatic insect identification. It contains 26,399 audio files from 459 species of Orthoptera (crickets, grasshoppers, katydids) and Cicadidae (cicadas), providing 9.5 days of audio material.

Key Features

459 unique insect species (310 Orthopteran and 149 Cicada species)
26,399 audio files in WAV and MP3 formats
9.5 days of total audio duration
Diverse sample rates ranging from 8 to 500 kHz
Worldwide geographic coverage (heavily focused on Europe and North America)
Creative Commons licensed (CC-BY-4.0 or CC0)

Dataset Split

The dataset is divided into training, validation, and test sets with a 60/20/20 split:

Training set: 15,873 files (137.3 hours)
Validation set: 5,307 files (46.2 hours)
Test set: 5,219 files (43.7 hours)

Note: For the 2025 BioDCASE data challenge, the test set is being held back until the challenge concludes. The full version will be published as version 1.0 afterward.

Data Sources

Audio recordings were collected from three primary sources:

xeno-canto: Orthoptera recordings
iNaturalist: Orthoptera & Cicadidae recordings
BioAcoustica: Cicadidae recordings

Curation Process

Only research-grade observations from iNaturalist
One file per observation for those with multiple audio attachments
Deduplication based on MD5 checksums
Pooling recordings by username, species, location, date, and time
Selection of only one recording from a one-hour period
Minimum 10 sound examples per species
Stereo files converted to mono
Standardized formats (WAV and MP3)
Long recordings (>2 minutes) automatically trimmed
Species nomenclature unified using COL24.4 taxonomy

Use Cases

Development of acoustic insect monitoring systems
Biodiversity assessment through passive acoustic monitoring
Testing new audio representation methods for highly variable frequencies
Ecological research and insect population monitoring

Citation

@misc{faiß2025insectset459opendatasetinsect,
      title={InsectSet459: an open dataset of insect sounds for bioacoustic machine learning}, 
      author={Marius Faiß and Burooj Ghani and Dan Stowell},
      year={2025},
      eprint={2503.15074},
      archivePrefix={arXiv},
      primaryClass={cs.SD},
      url={https://arxiv.org/abs/2503.15074}, 
}

Creators

Marius Faiß (Max Planck Institute of Animal Behavior, Konstanz, Germany)
Burooj Ghani (Naturalis Biodiversity Centre, Leiden, The Netherlands)
Dan Stowell (Tilburg University, The Netherlands)