esc-50

Created by K. J. Piczak

version	1.0.1
license	CC-BY-NC-3.0
usage	research
languages
format	wav
channel	1
sampling rate	44100
bit depth	16
duration	0 days 02:46:40
files	2000, duration distribution: each file is 5.0 s
repository	audb-public

Description

The ESC-50 dataset is a labeled collection of 2000 environmental audio recordings suitable for benchmarking methods of environmental sound classification. The dataset consists of 5-second-long recordings organized into 50 semantical classes (with 40 examples per class) loosely arranged into 5 major categories.

Example

audio/2-188822-D-40.wav

Tables

Click on a row to toggle a preview.

ID	Type	Columns
files	filewise	category, clip_id, esc10, fold, major, take	file category clip_id esc10 fold major take audio/1-100032-A-0.wav dog 100032 True 1 animals A audio/1-100038-A-14.wav chirping_birds 100038 False 1 natural A audio/1-100210-A-36.wav vacuum_cleaner 100210 False 1 interior A audio/1-100210-B-36.wav vacuum_cleaner 100210 False 1 interior B audio/1-101296-A-19.wav thunderstorm 101296 False 1 natural A 2000 rows x 6 columns

Schemes

ID	Dtype	Labels
category	str	airplane, breathing, brushing_teeth, can_opening, car_horn, cat, chainsaw, […], snoring, thunderstorm, toilet_flush, train, vacuum_cleaner, washing_machine, water_drops, wind
clip_id	str
esc10	bool
fold	int	1, 2, 3, 4, 5
major	str	animals, exterior, human, interior, natural
take	str	A, B, C, D, E, F, G, H