FST

Interface to train FST-based pronunciation generation. Phonetisaurus is used https://github.com/AdolfVonKleist/Phonetisaurus under the hood. FST-based model is used for both spelling and regular pronunciation training. For spelling, training data should just specify how each letter should be pronounced. There is also code for evaluation of pronunciation generation algo. In the future it should generalized to work with any pronunciation generation method.