Thai2Rom

Deep learning thai romanization.

Thai2Rom is trained from 80 % of Thai Romanization (https://www.kaggle.com/wannaphong/thai-romanization) and test on the rest 20 %.

Number of samples: 647352
Number of unique input tokens: 91
Number of unique output tokens: 39
Max sequence length for inputs: 29
Max sequence length for outputs: 57
Train on 517881 samples, validate on 129471 samples

Epoch 11
loss: 0.0062 - val_loss: 0.0100