facebookresearch / voxpopuli

A large-scale multilingual speech corpus for representation learning, semi-supervised learning and interpretation
Other
509 stars 50 forks source link

Download scripts do not work with pytorchaudio>=2.0.0 #43

Open Whyki opened 1 year ago

Whyki commented 1 year ago

Current checkout does not work, because dependency was download_url was removed from pytorchaudio

raivisdejus commented 1 year ago

Please use torchaudio==0.13.1 See https://github.com/facebookresearch/voxpopuli/pull/42

JIE-CHI commented 1 year ago

change https://github.com/facebookresearch/voxpopuli/blob/f7a3bb98d664e1d031763ec4f7639c4a530c64e9/voxpopuli/download_audios.py#L11 to

from torchaudio.datasets.utils import _extract_tar
from torch.hub import download_url_to_file

and https://github.com/facebookresearch/voxpopuli/blob/f7a3bb98d664e1d031763ec4f7639c4a530c64e9/voxpopuli/download_audios.py#L60-L61 to

download_url_to_file(url, tar_path.as_posix(), hash_prefix=None)
_extract_tar(tar_path.as_posix())

works for me

jordanhoskins commented 1 year ago

Changing requirements.txt to the following package list worked for me without any code changes. Note torch==1.13.1 and torchaudio==0.13.1. Numpy was also not included in the original version of the file.

docopt==0.6.2
editdistance==0.6.2
edlib==1.3.9
filelock==3.12.4
Jinja2==3.1.2
MarkupSafe==2.1.3
mpmath==1.3.0
networkx==3.1
num2words==0.5.12
numpy==1.24.4
sympy==1.12
torch==1.13.1
torchaudio==0.13.1
tqdm==4.66.1
typing_extensions==4.8.0