Automatic resampling of speech, noise, and RIR

[jfsantos]

Right now we are not checking for the sampling rate of any of the files until we call add_noise or generate_condition, but ideally we should make all files have the same fs when they are added to a dataset. However, this brings another issue: if the dataset is too large, we cannot store these resampled files in memory, so we would need to save them somewhere else.