Better datasets to consider

[greigs]

I'm not sure if these are larger than what you are using, but have a look at the datasets used by speech-to-text-wavenet