Datasets:
Tasks:
Audio Classification
Modalities:
Audio
Languages:
English
Size:
10K<n<100K
Tags:
audio
License:
Dataset Viewer
Full Screen Viewer
Full Screen
The viewer is disabled because this dataset repo requires arbitrary Python code execution. Please consider
removing the
loading script
and relying on
automated data support
(you can use
convert_to_parquet
from the datasets
library). If this is not possible, please
open a discussion
for direct help.
Dataset Card for AudioSet
Dataset Summary
AudioSet is a dataset of 10-second clips from YouTube, annotated into one or more sound categories, following the AudioSet ontology.
Supported Tasks and Leaderboards
audio-classification
: Classify audio clips into categories. The leaderboard is available here
Languages
The class labels in the dataset are in English.
Dataset Structure
Data Instances
Example instance from the dataset:
{
'video_id': '--PJHxphWEs',
'audio': {
'path': 'audio/bal_train/--PJHxphWEs.flac',
'array': array([-0.04364824, -0.05268681, -0.0568949 , ..., 0.11446512,
0.14912748, 0.13409865]),
'sampling_rate': 48000
},
'labels': ['/m/09x0r', '/t/dd00088'],
'human_labels': ['Speech', 'Gush']
}
Data Fields
Instances have the following fields:
video_id
: astring
feature containing the original YouTube ID.audio
: anAudio
feature containing the audio data and sample rate.labels
: a sequence ofstring
features containing the labels associated with the audio clip.human_labels
: a sequence ofstring
features containing the human-readable forms of the same labels as inlabels
.
Data Splits
The distribuion of audio clips is as follows:
balanced
configuration
train | test | |
---|---|---|
# instances | 18685 | 17142 |
unbalanced
configuration
train | test | |
---|---|---|
# instances | 1738788 | 17142 |
Dataset Creation
Curation Rationale
Source Data
Initial Data Collection and Normalization
Who are the source language producers?
The labels are from the AudioSet ontology. Audio clips are from YouTube.
Annotations
Annotation process
Who are the annotators?
Personal and Sensitive Information
Considerations for Using the Data
Social Impact of Dataset
Discussion of Biases
Other Known Limitations
- The YouTube videos in this copy of AudioSet were downloaded in March
2023, so not all of the original audios are available. The number of
clips able to be downloaded is as follows:
- Balanced train: 18685 audio clips out of 22160 originally.
- Unbalanced train: 1738788 clips out of 2041789 originally.
- Evaluation: 17142 audio clips out of 20371 originally.
- Most audio is sampled at 48 kHz 24 bit, but about 10% is sampled at 44.1 kHz 24 bit. Audio files are stored in the FLAC format.
Additional Information
Dataset Curators
Licensing Information
The AudioSet data is licensed under CC-BY-4.0
Citation
@inproceedings{jort_audioset_2017,
title = {Audio Set: An ontology and human-labeled dataset for audio events},
author = {Jort F. Gemmeke and Daniel P. W. Ellis and Dylan Freedman and Aren Jansen and Wade Lawrence and R. Channing Moore and Manoj Plakal and Marvin Ritter},
year = {2017},
booktitle = {Proc. IEEE ICASSP 2017},
address = {New Orleans, LA}
}
- Downloads last month
- 15,551
Models trained or fine-tuned on agkphysics/AudioSet
Audio Classification
•
Updated
•
547
•
2