Maybe integrate the open source sound to letter system from Facebook AI research

[cbrew]

This could close the loop letters -> sound -> letters.

https://github.com/facebookresearch/wav2letter