cli for easily building datasets of audio files from google's AudioSet.
Update may 2023: metadata has been extended with number of views, likes, comments, availability
pip install audioset-downloader
note that you'll need to have ffmpeg
installed on your system.
- filter by class names (union or intersection)
- filter by set (train [balanced, unbalanced], eval)
- limit number of downloads
- select most viewed / most liked
Usage: audioset-dl [OPTIONS]
download examples of a specific class from google's AudioSet
Options:
-o, --output-dir TEXT target directory for the downloads
(default='./')
-c, --class-name TEXT the name of the class to download
(default=Snoring)this option can be repeated
to select examples at the intersection of
multiple classese.g. `-c Music -c
Techno`(list of available classes can be
printed out with the command `audioset-
classes`)
-u, --class-union toggle whether class names should intersect
(default) or not
-m, --mixed if provided, the downloaded examples will be
instances of `--class-name` and possibly some
other classes. Otherwise (default behaviour),
downloaded examples have only `--class-name`
as single label.
-xe, --exclude-eval-set if provided, exclude examples from the eval
set (default=False)
-xb, --exclude-balanced-set if provided, exclude examples from the
balanced set (default=False)
-xu, --exclude-unbalanced-set if provided, exclude examples from the
unbalanced set (default=False)
-n, --n-examples INTEGER number of examples to download (default=all
matching)
-f, --full-source if provided, download full examples instead
of 10 sec. segments (default=False)
-mv, --most-viewed if --n-examples is provided, only the n most
viewed examples will be downloaded
-ml, --most-liked if --n-examples is provided, only the n most
liked examples will be downloaded
--help Show this message and exit.
you can also print the available classes names with
audioset-classes
- AudioSet Homepage: https://research.google.com/audioset/index.html
- Dataset classes content: https://research.google.com/audioset/dataset/index.html
MIT