[Audio] Allow resampling for audio datasets in streaming mode

[patrickvonplaten]

Many audio datasets like Common Voice always need to be resampled. This can very easily be done in non-streaming mode as follows:

from datasets import load_dataset

ds = load_dataset("common_voice", "ab", split="test")

ds = ds.cast_column("audio", Audio(sampling_rate=16_000))

However in streaming mode it fails currently:

from datasets import load_dataset

ds = load_dataset("common_voice", "ab", split="test", streaming=True)

ds = ds.cast_column("audio", Audio(sampling_rate=16_000))

with the following error:

AttributeError: 'IterableDataset' object has no attribute 'cast_column'  

It would be great if we could add such a feature (I'm not 100% sure though how complex this would be)

[lhoestq]

This requires implementing cast_column for iterable datasets, it could be a very nice addition !

It can also be useful to be able to disable the audio/image decoding for the dataset viewer (see PR #3430) cc @severo
EDIT: actually following #3145 the dataset viewer might not need it anymore

[patrickvonplaten]

Just to clarify a bit. This feature is always needed when using the common voice dataset in streaming mode. So I think it's quite important