Support for Speech Tasks

[anishacharya]

Support for Speech Tasks
We would like to add support for at least one speech task. Ideally, follow the vision_datasets.py implementation and do a parallel implementation, The output should be similar to that of vision_datasets