Add text normalization

[vvolhejn]

For cases like #73, we should pre-process the text (apply text normalization) to make the job easier for the text to speech.

See e.g. https://developer.nvidia.com/blog/text-normalization-and-inverse-text-normalization-with-nvidia-nemo

Complicated examples:

• 2/3 -> two thirds
• $100 -> hundred dollars (or "one hundred dollars")
• 50g -> fifty grams
• Sunset Blvd -> Sunset Boulevard

The model might learn some of these from the training data, but we can make its job easier.

[sanjaymarison]

Different English version pronounce the numbers differently, like some say 1 lakh dollars but americans might say hundred thousand dollars. So maybe account for that as well?

[teddybear082]

This may help (MIT licensed): https://github.com/keithito/tacotron/blob/master/text/cleaners.py