Download | A peek inside | License

The ESC-50 dataset is a public labeled set of 2000 environmental recordings (50 classes, 40 clips per class, 5 seconds per clip) suitable for environmental sound classification tasks.

See ESC: Dataset for Environmental Sound Classification - paper replication data for the full paper with a more thorough analysis.

The dataset consists of 50 classes of recordings in 5 loosely defined groups:

Animals

• 101 - Dog
• 102 - Rooster
• 103 - Pig
• 104 - Cow
• 105 - Frog
• 106 - Cat
• 107 - Hen
• 108 - Insects (flying)
• 109 - Sheep
• 110 - Crow

Natural soundscapes & water sounds

• 201 - Rain
• 202 - Sea waves
• 203 - Crackling fire
• 204 - Crickets
• 205 - Chirping birds
• 206 - Water drops
• 207 - Wind
• 208 - Pouring water
• 209 - Toilet flush
• 210 - Thunderstorm

Human, non-speech sounds

• 301 - Crying baby
• 302 - Sneezing
• 303 - Clapping
• 304 - Breathing
• 305 - Coughing
• 306 - Footsteps
• 307 - Laughing
• 308 - Brushing teeth
• 309 - Snoring
• 310 - Drinking - sipping

Interior/domestic sounds

• 401 - Door knock
• 402 - Mouse click
• 403 - Keyboard typing
• 404 - Door - wood creaks
• 405 - Can opening
• 406 - Washing machine
• 407 - Vacuum cleaner
• 408 - Clock alarm
• 409 - Clock tick
• 410 - Glass breaking

Exterior/urban noises

• 501 - Helicopter
• 502 - Chainsaw
• 503 - Siren
• 504 - Car horn
• 505 - Engine
• 506 - Train
• 507 - Church bells
• 508 - Airplane
• 509 - Fireworks
• 510 - Hand saw

Clips have been constructed from public field recordings gathered by the Freesound.org project. The dataset has been prearranged into 5 folds. Clips stemming from the same original source file are contained in a single fold.

File naming scheme:

category_id - category_name/fold_number-Freesound_clip_ID-take_letter.ogg

File details:

5-second-long recordings reconverted to a unified format:
- 44100 Hz,
- single channel (monophonic),
- Vorbis/Ogg compression @ 192 kbit/s. 

The dataset can be downloaded as a single .zip file (~200 MB):

ESC-50 dataset

Waveforms and mel-spectrograms of ESC-50 dataset recordings:

The dataset is available under the terms of the Creative Commons license - Attribution-NonCommercial.

A smaller subset (ESC-10) is available under CC BY (Attribution).

In academic settings please cite:

K. J. Piczak. ESC: Dataset for Environmental Sound Classification. In Proceedings of the ACM International Conference on Multimedia, in press, ACM, 2015.

[DOI: http://dx.doi.org/10.1145/2733373.2806390]

Due to GitHub limitations (README length limit) licensing details for individual clips are available in the plain text README.