Description

t2t-datagen command is not working properly with problem=common_voice . It is due to the URL not being static anymore so the file being downloaded is just the webpage showing Access Denied in a .xml. You can directly enter the static common voice URL in browser and see that it does not download the .tar.gz dataset but instead shows the Access Denied webpage

Environment information

OS: Tested on both Google Cloud Platform and Windows 10

$ pip freeze | grep tensor
mesh-tensorflow==0.0.5
tensor2tensor==1.14.0
tensorboard==1.15.0
tensorflow==1.15.0
tensorflow-cpu==2.1.0
tensorflow-datasets==1.2.0
tensorflow-estimator==1.15.1
tensorflow-gan==2.0.0
tensorflow-hub==0.7.0
tensorflow-metadata==0.21.0
tensorflow-probability==0.7.0
tensorflow-serving-api==2.1.0rc1

$ python -V
# your output here
python 2.7.13

Steps to reproduce:

Use the t2t-datagen command with problem=common_voiceor problem=common_voice_clean
In the code we can see that the URL of common voice is static:
_COMMONVOICE_URL = "https://common-voice-data-download.s3.amazonaws.com/cv_corpus_v1.tar.gz"

while for example the french URL is generated and looks like this:
https://voice-prod-bundler-ee1969a6ce8178826482b88e843c335139bd3fb4.s3.amazonaws.com/cv-corpus-4-2019-12-10/fr.tar.gz

If this gets fixed, it would also be nice to have a flag allowing us to choose which language we want our dataset to be in.