for testing and deploying your application
for finding and fixing issues
for empowering human code reviews
from fuel.datasets.hdf5 import H5PYDataset
from fuel.utils import find_in_data_path
class YouTubeAudio(H5PYDataset):
r"""Dataset of audio from YouTube video.
Assumes the existence of a dataset file with the name
`youtube_id.hdf5`. These datasets don't have any split; the entire
audio sequence is considered training.
Note that the data structured in the form `(batch, time, features)`
where `features` are the audio channels (dimension 1 or 2) and batch is
equal to 1 in this case (since there is only one audiotrack).
Parameters
----------
youtube_id : str
11-character video ID (taken from YouTube URL)
\*\*kwargs
Passed to the `H5PYDataset` class.
"""
def __init__(self, youtube_id, **kwargs):
super(YouTubeAudio, self).__init__(
file_or_path=find_in_data_path('{}.hdf5'.format(youtube_id)),
which_sets=('train',), **kwargs
)
mila-udem /
fuel
